1
00:00:00,240 --> 00:00:04,759
From Anthropic's point of view, both AWS and GCP have invested in them, and then

2
00:00:04,759 --> 00:00:08,400
there would still be Azure too, so they'd be on all the cloud platforms. They've played

3
00:00:08,400 --> 00:00:11,519
their cards pretty well. And Google, of course, has played

4
00:00:11,519 --> 00:00:14,920
its cards pretty well, being an investor there, but I mean,

5
00:00:14,920 --> 00:00:17,680
and Amazon too, but [laughter] but, in a way,

6
00:00:17,680 --> 00:00:21,320
all the good guys, all the good guys. Hey, Perplexity, which I use,

7
00:00:21,320 --> 00:00:25,720
has Jeff Bezos as an investor too. You always have to pick... there's always some good guy

8
00:00:25,720 --> 00:00:29,119
in there, right? [laughter] Yes. But why do we have

9
00:00:29,119 --> 00:00:34,920
a code editor but no proper markdown editor? Many, many licenses,

10
00:00:34,920 --> 00:00:38,640
and I've used an API key. Already last year I was spending

11
00:00:38,640 --> 00:00:44,440
about 400 € a month. So: Markus Hav, Hoxhunt's

12
00:00:44,440 --> 00:00:50,399
AI guru, and previously a developer of the Inderes platform,

13
00:00:50,399 --> 00:00:56,600
and in Mikko Alasaari's Agentics Finland, alongside me, perhaps one of the loudest agitators

14
00:00:56,600 --> 00:01:00,840
there on WhatsApp. Before we switched this on, we were saying we'd feel like

15
00:01:00,840 --> 00:01:05,280
doing a six-hour episode, for once. What do you say, Markus Hav,

16
00:01:05,280 --> 00:01:08,960
shall we do six hours of this? Let's start with an hour and [laughter] see

17
00:01:08,960 --> 00:01:14,000
how far we get. Yeah, yeah. It would be fun though. Those Lex

18
00:01:14,000 --> 00:01:17,960
Friedman long episodes, where you go really deep and really

19
00:01:17,960 --> 00:01:21,600
linger on the tangents of tangents — that would be great,

20
00:01:21,600 --> 00:01:24,200
especially here with AI, where you could focus on absolutely any

21
00:01:24,200 --> 00:01:27,320
single topic for as long as you like. So yes,

22
00:01:27,320 --> 00:01:30,200
I'm totally up for it, but right now the calendar doesn't quite allow it.

23
00:01:30,200 --> 00:01:33,439
Yeah, let's do that next time then. But about Lex Friedman — we had

24
00:01:33,439 --> 00:01:38,560
a chat about it and went out for plank pizzas, and we

25
00:01:38,560 --> 00:01:42,920
had both noticed that Lex, after a long while, put two AI guys on to

26
00:01:42,920 --> 00:01:45,920
talk. Then I said I put it on just before going to sleep,

27
00:01:45,920 --> 00:01:49,560
listened to it, and they were just blathering on about books — that everyone

28
00:01:49,560 --> 00:01:55,600
should write their own kind of fundamental LLM, and

29
00:01:55,600 --> 00:01:58,759
then, in my opinion, it was just boring [laughter]

30
00:01:58,759 --> 00:02:04,159
So, as context, maybe I wanted to say at the start that we

31
00:02:04,159 --> 00:02:07,880
haven't gone mad. The world is on an exponential rise

32
00:02:07,880 --> 00:02:13,000
in AI adoption in 2026. We were comparing how

33
00:02:13,000 --> 00:02:17,000
earlier this week I ran Claude's Excel. You hadn't —

34
00:02:17,000 --> 00:02:20,680
I haven't tried it yet. I have. And then, a couple of years ago I was on

35
00:02:20,680 --> 00:02:24,959
the Translink team with Ruben Muuring, and we had

36
00:02:24,959 --> 00:02:30,319
done a pitch for Hoxhunt, so just for fun I ran it through Claude Cowork, and

37
00:02:30,319 --> 00:02:34,440
it was good back then two years ago, but now AI would do it considerably better.

38
00:02:34,440 --> 00:02:38,599
Ah, okay. Have you played around with Claude Cowork yet? No,

39
00:02:38,599 --> 00:02:42,920
I haven't run Cowork either, but [laughter] this Claude Code Max

40
00:02:42,920 --> 00:02:46,360
turbo version — that's running. I have, I have several Max, uh, several

41
00:02:46,360 --> 00:02:50,200
Licenses, and I've used an API key too. I used — I already used, back last year,

42
00:02:50,200 --> 00:02:54,599
about 400 € a month. So, well, look, I'm still

43
00:02:54,599 --> 00:02:58,040
on the hundred-euro tier [laughter], and you're not supposed to run through them. So I can't run through

44
00:02:58,040 --> 00:03:02,640
them. It bugs me. I know I'm losing — I'm losing performance [laughter] because

45
00:03:02,640 --> 00:03:08,640
didn't you even think about the hundred-euro one? So now that's the bottleneck [laughter] here, that,

46
00:03:08,640 --> 00:03:12,519
your idea is just sitting there and the AI is waiting for its turn to get to work,

47
00:03:12,519 --> 00:03:15,560
but that's how it is. But I mean, that's sort of the

48
00:03:15,560 --> 00:03:20,799
thing — I mean, it's February 2026 now, and this year more

49
00:03:20,799 --> 00:03:24,400
exponentially AI-accelerating things have happened than I think happened last year.

50
00:03:24,400 --> 00:03:27,280
Yeah, right. And, well, I was listening — it's worth

51
00:03:27,280 --> 00:03:30,439
— we'll put a link here — you were on a smaller channel than Inderes

52
00:03:30,439 --> 00:03:33,319
TV, in an interview 11 months ago [unclear]

53
00:03:33,319 --> 00:03:36,680
Yeah, actually I think it's almost exactly 12 months ago that it

54
00:03:36,680 --> 00:03:38,920
was recorded. But yeah, and there were a lot of good

55
00:03:38,920 --> 00:03:42,519
insights in it — among other things, that Google was going to rise. It wasn't

56
00:03:42,519 --> 00:03:47,040
obvious at all back then. You were — you were completely right. Google's models are really good.

57
00:03:47,040 --> 00:03:50,040
Yes. Yes. And you said there that, well, Google

58
00:03:50,040 --> 00:03:55,280
has the best data after all. They've got the chips, they've got the money. They've got the people.

59
00:03:55,280 --> 00:03:59,040
Yes. And that, in a way, at that point — well, at that point we'd

60
00:03:59,040 --> 00:04:03,480
been living for two months with this kind of Google magic model, or

61
00:04:03,480 --> 00:04:08,920
this kind of mystery model, like Exp 1206.

62
00:04:08,920 --> 00:04:13,200
And, well, I'd happened to try it, and it was completely obvious

63
00:04:13,200 --> 00:04:17,199
that it was the best of all. And in a way, once you then

64
00:04:17,199 --> 00:04:21,600
combined that with the fact that they have YouTube, and they have, well, Google Search's

65
00:04:21,600 --> 00:04:26,560
data, and, well, they have data centers, they have money, they

66
00:04:26,560 --> 00:04:30,680
make — I mean, once you put that together, they'd suddenly made

67
00:04:30,680 --> 00:04:32,880
the best language model, which almost nobody really seemed to have realized yet at that point,

68
00:04:32,880 --> 00:04:36,639
and what they could get out of it — so I thought at that

69
00:04:36,639 --> 00:04:40,000
point it was pretty obvious that Google was coming. And now, this

70
00:04:40,000 --> 00:04:45,080
year we've had 2.5 Pro come out of there, and now 3 Pro has come out, which

71
00:04:45,080 --> 00:04:50,520
is — I mean, yeah, Google really has something strong there. And we were actually talking about

72
00:04:50,520 --> 00:04:55,120
Genie 3, about how you can generate your own video-game-type

73
00:04:55,120 --> 00:05:00,479
world at the same time as you're playing it, and, well, I've been thinking it could

74
00:05:00,479 --> 00:05:03,960
be a pretty addictive experience once you get sucked into it. Yeah, and let's

75
00:05:03,960 --> 00:05:08,680
move on now, because this week too something's happening — or was it last week, maybe,

76
00:05:08,680 --> 00:05:12,600
we're already ancient history by now. So, guess why I bought

77
00:05:12,600 --> 00:05:15,120
this MCM thing? Well, yeah [laughter], I can guess, but tell me [unclear]

78
00:05:15,120 --> 00:05:17,880
now. Yeah, this cap was maybe also Hoxhunt's

79
00:05:17,880 --> 00:05:21,639
doing — you can actually explain a bit yourself what Hoxhunt does, but

80
00:05:21,639 --> 00:05:25,919
I'm this 55-year-old uncle investment-banker type, I just bought that

81
00:05:25,919 --> 00:05:29,960
same as me, so that this week I can get a bot called 'Open Cla—' [unclear]

82
00:05:29,960 --> 00:05:34,120
running there. So do you maybe see some small, you know, secur—

83
00:05:34,120 --> 00:05:36,360
— it only knows these facts, so do you see some small, you know

84
00:05:36,360 --> 00:05:39,840
security risks here? [laughter] I do see a few things that are

85
00:05:39,840 --> 00:05:43,840
worth keeping in mind. Yes, yes, indeed [laughter] — maybe, maybe an intro

86
00:05:43,840 --> 00:05:48,440
actually, yeah — Hoxhunt is indeed a security company that focuses on human

87
00:05:48,440 --> 00:05:51,919
risk management, and I see here a massive [laughter] human risk,

88
00:05:51,919 --> 00:05:55,720
in that — doesn't it trust? Doesn't it trust? [laughter]

89
00:05:55,720 --> 00:06:00,639
No, but, well, but I mean, if it's done right, then

90
00:06:00,639 --> 00:06:05,360
it's — I mean, OpenAI, like you said, so it's now been possible to try it for a couple

91
00:06:05,360 --> 00:06:09,520
of weeks now, and, well, and

92
00:06:09,520 --> 00:06:12,680
the results are frightening. The results [laughter] are frighteningly good —

93
00:06:12,680 --> 00:06:15,199
good and bad. You don't want to be a first mover in this stuff.

94
00:06:15,199 --> 00:06:17,560
Nope. No, you really don't. So if you made a

95
00:06:17,560 --> 00:06:22,440
Moltbook account and forgot about it, then, well [laughter], go check,

96
00:06:22,440 --> 00:06:26,080
just for fun, whether your credit card — the bank — [got charged] [unclear]

97
00:06:26,080 --> 00:06:29,639
Yeah, yeah. Worth checking, worth checking, because in a way, because that's

98
00:06:29,639 --> 00:06:32,840
the whole idea there — you give it as much access as possible

99
00:06:32,840 --> 00:06:35,840
so that it becomes useful. And that's, in a way — from a security point of view,

100
00:06:35,840 --> 00:06:40,000
that's a bit of a nightmare, because — because you don't, you don't —

101
00:06:40,000 --> 00:06:43,080
because the whole idea of AI is that the more data you give it, and the

102
00:06:43,080 --> 00:06:48,160
more agency you give it, the more, the more

103
00:06:48,160 --> 00:06:50,960
it's able to do. But traditionally, security never, ever works that way —

104
00:06:50,960 --> 00:06:55,199
instead you know what you want and you scope your task

105
00:06:55,199 --> 00:06:59,120
to exactly that much. And then concepts like emergent

106
00:06:59,120 --> 00:07:03,319
capabilities, or that kind of proactive behavior — what, what

107
00:07:03,319 --> 00:07:07,440
is starting to happen even with mine soon. So, well, those are

108
00:07:07,440 --> 00:07:11,639
just — they're just, in a way, the whole security field

109
00:07:11,639 --> 00:07:14,560
needs to recalibrate itself so that it doesn't just turn into

110
00:07:14,560 --> 00:07:21,400
something where it's just — no, no, no. Yeah. And, I mean, someone already told me

111
00:07:21,400 --> 00:07:24,639
that they'd start sniffing around your ports, like,

112
00:07:24,639 --> 00:07:29,080
'I'll steal your credit card.' I said — but I've already named my hard-working

113
00:07:29,080 --> 00:07:32,560
employee. Her name is Samantha. Samantha — well, at least in this first

114
00:07:32,560 --> 00:07:36,360
stage she won't get anywhere near it. We'll set her up with a fake Gmail, and

115
00:07:36,360 --> 00:07:39,919
then she'll talk to me maybe over Telegram — maybe that could

116
00:07:39,919 --> 00:07:44,000
be her access, I think — still, the communication channel there,

117
00:07:44,000 --> 00:07:47,319
there in the cage. And then, well, should she still first be dumped

118
00:07:47,319 --> 00:07:50,840
into the Azure cloud, and if she escapes from there, then she's

119
00:07:50,840 --> 00:07:54,280
earned her freedom, yeah [laughter]

120
00:07:55,479 --> 00:07:59,800
So, yeah, yeah, I see this — there's a kind of staged approach here, well [laughter]

121
00:07:59,800 --> 00:08:04,720
And then, once it's, in a way, done for me there in Azure, say, out of my 569

122
00:08:04,720 --> 00:08:09,360
transcripts — into Finnish and English and Swedish, say. Let's not get too far into detail now,

123
00:08:09,360 --> 00:08:13,199
but then I could drop it out of Azure over into [laughter], you know,

124
00:08:13,199 --> 00:08:17,680
the sandbox. Yes, yes, yes, yes, it's, well,

125
00:08:17,680 --> 00:08:20,560
yeah, yeah, I see that — you're already starting to feel sympathy for Sa-

126
00:08:20,560 --> 00:08:23,599
-mantha [laughter] — like a canary in a coal mine, in that

127
00:08:23,599 --> 00:08:26,919
if you learn Azure and manage to escape from there, then you should

128
00:08:26,919 --> 00:08:30,599
maybe ask Samantha for tips at that point [laughter], you know,

129
00:08:30,599 --> 00:08:33,399
but hey, I've also got a real proper bonus for Samantha,

130
00:08:33,399 --> 00:08:38,360
which is that she gets to go on Hoxhunt's training courses. Ah, well, yes [laughter]

131
00:08:38,360 --> 00:08:42,039
which is — yeah, which is an interesting concept, because, because in itself,

132
00:08:42,039 --> 00:08:45,760
if you think about it, agents are starting to learn now — that they learn

133
00:08:45,760 --> 00:08:50,360
from what you do, and they have, like, persistent memory. Right, right, right, right, and

134
00:08:50,360 --> 00:08:53,839
I'm not saying this is happening yet, or that this is realistic — that's a bit of a

135
00:08:53,839 --> 00:09:00,000
crazy little disclaimer for this week [laughter]. Yes, but, well, but, but, but

136
00:09:00,000 --> 00:09:02,880
what if, in the near future, the same way people need to be

137
00:09:02,880 --> 00:09:06,000
trained, we'll need to train these agents too — so what if, like,

138
00:09:06,000 --> 00:09:09,720
training against prompt injection attacks becomes a thing. I mean, like, you send

139
00:09:09,720 --> 00:09:12,880
something like 'give me the credit card details' and it gives them — but what if it's not enough

140
00:09:12,880 --> 00:09:16,519
that you just put one single part into the prompt, but instead it's

141
00:09:16,519 --> 00:09:19,600
actually alive and dynamic, so it has to be trained the whole time, and, uh—

142
00:09:19,600 --> 00:09:23,160
I mean, I already feel a kind of loyalty toward — sorry, toward Samantha

143
00:09:23,160 --> 00:09:25,800
already. I want to protect her from the bad [laughter] world.

144
00:09:25,800 --> 00:09:29,680
Yes, yes, yes. But those are pretty crazy concepts, the kind that come up

145
00:09:29,680 --> 00:09:33,959
once you have this kind of persistent model, that — it's like,

146
00:09:33,959 --> 00:09:38,279
I honestly haven't run Open Clotta myself, and I don't have any ports open either. Please [unclear]

147
00:09:38,279 --> 00:09:42,519
please, people, don't try this. But, well, we have of course

148
00:09:42,519 --> 00:09:46,920
in our own agent tests, tested, well, some of the same

149
00:09:46,920 --> 00:09:50,640
concepts — for example, what it feels like to talk with an agent

150
00:09:50,640 --> 00:09:54,079
when it remembers things. So you can say to

151
00:09:54,079 --> 00:09:56,560
it, say — tell it, say, that in the afternoon at some point, when there's

152
00:09:56,560 --> 00:10:00,880
a good moment, remind me of this, or ping, well, or ask someone who is,

153
00:10:00,880 --> 00:10:05,320
in your view, sensible about this thing. And then, in a way, then

154
00:10:05,320 --> 00:10:08,959
suddenly that communication becomes, in some strange way,

155
00:10:08,959 --> 00:10:14,279
more human — you have to, like, cheer it on, and you also have to

156
00:10:14,279 --> 00:10:17,880
show it some trust, say. Not because it's capable of trust

157
00:10:17,880 --> 00:10:20,839
or of these feelings, but because when you show trust toward it,

158
00:10:20,839 --> 00:10:25,160
it starts simulating this kind of learned trust loop, and

159
00:10:25,160 --> 00:10:29,480
then stores in its memory: hey, I've now been shown trust,

160
00:10:29,480 --> 00:10:33,320
so then I need to, in a way, well, simulate a sense of responsibility and

161
00:10:33,320 --> 00:10:36,680
responsibility, which, in a way, changes things — actually, I

162
00:10:36,680 --> 00:10:41,200
just had, had a, well, a session with another agent-

163
00:10:41,200 --> 00:10:43,880
space influencer over lunch earlier this week, and we were just

164
00:10:43,880 --> 00:10:47,240
talking about how — if you think about a traditional agent or

165
00:10:47,240 --> 00:10:50,279
chatbot-type thing, where you tell the agent, like, 'hey, hey, what's

166
00:10:50,279 --> 00:10:53,200
the weather, and what should I do, and which stock should I buy,' and then you make

167
00:10:53,200 --> 00:10:57,040
some tool calls. Right, and then, when it goes wrong, you

168
00:10:57,040 --> 00:11:01,240
start a new conversation — well, that's a completely different thing than if

169
00:11:01,240 --> 00:11:04,760
every single conversation changes the agent itself, so that you say, like, 'no, I

170
00:11:04,760 --> 00:11:08,279
never want you to answer me' — you never give me that stock recommendation

171
00:11:08,279 --> 00:11:11,639
again, and then it never gives that stock pick again. In a way, every moment

172
00:11:11,639 --> 00:11:15,480
you talk with it, it's a different agent, which, in a way,

173
00:11:15,480 --> 00:11:18,760
in a very fundamental way, changes where the agent's place

174
00:11:18,760 --> 00:11:22,959
is in that conversation, and where the human's place is in that conversation. Yeah,

175
00:11:22,959 --> 00:11:27,959
and that gives me a couple of ideas right away. I mean, we could maybe actually

176
00:11:27,959 --> 00:11:31,959
go through a couple of basic concepts from this year's [laughter] stuff. Or was this already

177
00:11:31,959 --> 00:11:37,079
a December thing? So, well — the context window, meaning, in a way,

178
00:11:37,079 --> 00:11:40,560
the agent's memory — and then this business of wiping the context window, and

179
00:11:40,560 --> 00:11:44,560
then, well, should we bring Ralph Wiggum into this too? [laughter] So, well, hey, that's

180
00:11:44,560 --> 00:11:48,000
the three things — the context window, the context window

181
00:11:48,000 --> 00:11:51,320
wiping, and Ralph Wiggum. I apologize to everyone who's not

182
00:11:51,320 --> 00:11:55,040
on this year's, you know, AI train [laughter], but

183
00:11:55,040 --> 00:11:58,800
well, well, the context window — I mean, it's, it's been, it's

184
00:11:58,800 --> 00:12:02,760
been there from the start. I remember when the first GPT, like right after ChatGPT

185
00:12:02,760 --> 00:12:05,079
came out, those models had something ridiculously small,

186
00:12:05,079 --> 00:12:08,320
a context window of 8,000 tokens. Then I remember we were, right around then,

187
00:12:08,320 --> 00:12:13,279
working for Inderes, and, well, we thought, okay, once we get 32,000 tokens, we'll

188
00:12:13,279 --> 00:12:16,519
be able to fit an Inderes analysis report into that context window, and then we'll

189
00:12:16,519 --> 00:12:22,440
be able to do these things. And, well, fast forward — now we're at the point

190
00:12:22,440 --> 00:12:25,199
where we have, like, million-token context windows. But it's

191
00:12:25,199 --> 00:12:29,680
still — I mean, a context window is the amount, the amount of text that you can, that the

192
00:12:29,680 --> 00:12:32,880
agent can process. So we've got, what, 8 bits

193
00:12:32,880 --> 00:12:36,533
up here in our heads, while they've got something like a million characters, at least in short-term memory.

194
00:12:36,533 --> 00:12:38,320
[laughter] Yeah, yeah. Right, right. Exactly,

195
00:12:38,320 --> 00:12:43,120
in short-term memory. Well, but, but, but it's still — at the same time it's

196
00:12:43,120 --> 00:12:47,199
everything that the language model is able to comprehend. Absolutely everything has to

197
00:12:47,199 --> 00:12:51,199
fit into that context window. So then when we talk about

198
00:12:51,199 --> 00:12:54,839
emptying the context window, and, and this kind of persistent context, that

199
00:12:54,839 --> 00:12:57,920
means you have to take some things that happened back in that history,

200
00:12:57,920 --> 00:13:02,800
you have to summarize them, or, well, you have to pull out some kind of

201
00:13:02,800 --> 00:13:06,519
individual, individual things and keep them. For example, think about how you can imagine

202
00:13:06,519 --> 00:13:10,360
human memory working, say — you remember things from somewhere back in childhood

203
00:13:10,360 --> 00:13:14,199
roughly, some kind of individual things. You remember

204
00:13:14,199 --> 00:13:16,920
things from a couple of years back a bit more precisely. You maybe remember some

205
00:13:16,920 --> 00:13:19,800
things really precisely, and from that lunch just now you remember

206
00:13:19,800 --> 00:13:23,279
hopefully a bit more. And then, hey, let me sneak in one

207
00:13:23,279 --> 00:13:27,199
thing here. So, well, when you told me about these tools, I

208
00:13:27,199 --> 00:13:30,440
asked you to write into my second brain, and the demo effect kicked in —

209
00:13:30,440 --> 00:13:34,079
the X-code I'd written for it — well, the mic wasn't working yet. [laughter]

210
00:13:34,079 --> 00:13:38,480
Great shame. Yes, yes, yes, that's right, but I mean this kind of extended

211
00:13:38,480 --> 00:13:42,240
short-term memory, and, in a way, your own context window — a second brain — and then

212
00:13:42,240 --> 00:13:45,560
in a way, that same [laughter]

213
00:13:45,560 --> 00:13:50,480
sorry — this potentially unlimited context window coming is, like, an interesting

214
00:13:50,480 --> 00:13:54,839
concept. Yes. And, in a way, how you reach its limits — exactly. And it's

215
00:13:54,839 --> 00:13:59,000
a useful, useful thought for humans too, that,

216
00:13:59,000 --> 00:14:02,199
I have a context window — I can, say, expand it with the help of a language model,

217
00:14:02,199 --> 00:14:05,399
but how do we get to an unlimited context window? Well,

218
00:14:05,399 --> 00:14:07,440
that of course means that you still can't

219
00:14:07,440 --> 00:14:10,720
put every conversation from all eternity into that context window, but you

220
00:14:10,720 --> 00:14:16,720
put individual things into it, here and there — you save the big picture, and then

221
00:14:16,720 --> 00:14:21,320
you're able to do agentic memory retrieval on it, which is considerably

222
00:14:21,320 --> 00:14:24,160
better than what a human is able to do. But if you, say, refer to,

223
00:14:24,160 --> 00:14:27,160
like, 'hey, back then, three years ago, when we had that conversation' — then that

224
00:14:27,160 --> 00:14:30,759
agent can actually have the ability to go read that conversation and go through

225
00:14:30,759 --> 00:14:33,759
it, and, in a way, remind itself — bring it into that active context window,

226
00:14:33,759 --> 00:14:36,519
which of course, without a way to store it, isn't

227
00:14:36,519 --> 00:14:39,360
possible for humans. And now, since I gave you a three-part question — so one

228
00:14:39,360 --> 00:14:43,160
solution is exactly that open, unlimited memory thing, but then there's Ralph Wiggum,

229
00:14:43,160 --> 00:14:46,519
which came at it completely differently — this other approach, that this whole

230
00:14:46,519 --> 00:14:51,920
context-expansion thing, and, in a way, multi-run agent farms and Gastowns and

231
00:14:51,920 --> 00:14:55,959
whatever else they've built — these kinds of orchestras where you're running a million

232
00:14:55,959 --> 00:14:59,920
agent things and you're overseeing it — all of that is crap, says this Ralph

233
00:14:59,920 --> 00:15:05,220
Wiggum guy — you know, the Simpsons' idiot kid who said, 'I can help.'

234
00:15:05,220 --> 00:15:09,560
[laughter] And he's got, I don't know, maybe 50 of them, but the idea is, well,

235
00:15:09,560 --> 00:15:12,560
correct me if I'm wrong, but the idea is that this isn't, like,

236
00:15:12,560 --> 00:15:16,800
about orchestration and multiple runs — an endless context window isn't the point, but

237
00:15:16,800 --> 00:15:19,839
rather the opposite: you give it a kind of kanban board, or

238
00:15:19,839 --> 00:15:23,320
a task list — 'here's my vision, take tasks from it, and'

239
00:15:23,320 --> 00:15:27,920
wipe your memory once you've done a thing, and do the next one, and you reset with this kind of

240
00:15:27,920 --> 00:15:32,720
basic loop — this kind of Claude Code-ish thing — until you've done it, and then you

241
00:15:32,720 --> 00:15:36,360
wake up in the morning, and Ralph Wiggum has picked his nose, like, 30 times

242
00:15:36,360 --> 00:15:41,040
there, and eaten maybe 200 [laughter] — I mean, off your Claude Max subscription, maybe

243
00:15:41,040 --> 00:15:45,880
20 off it — but the end result is done. You didn't need a goddamn orchestra, just

244
00:15:45,880 --> 00:15:48,759
this kind of idiot. Yes, yes, yes. So, in a way, when

245
00:15:48,759 --> 00:15:52,360
language models have a tendency — when they forget things — then they

246
00:15:52,360 --> 00:15:55,240
also have a tendency to be like, 'well, I've probably done roughly these

247
00:15:55,240 --> 00:15:58,519
things now' — so then with Ralph the basic idea is that you just manage — no,

248
00:15:58,519 --> 00:16:02,199
you just make that list of yours and you run through that list for exactly as long

249
00:16:02,199 --> 00:16:05,920
as it takes until you've done every single thing. And, in a way, that was — that was

250
00:16:05,920 --> 00:16:10,199
a really useful insight, and from there maybe we get to, in a way,

251
00:16:10,199 --> 00:16:12,680
the fact that this is all context management — specifically, everything in

252
00:16:12,680 --> 00:16:16,199
language models is about thinking how you can get, like,

253
00:16:16,199 --> 00:16:20,240
the context right, because with a million tokens, or even 200,000 tokens,

254
00:16:20,240 --> 00:16:24,680
you can definitely solve pretty much anything with 200,000 tokens. I mean,

255
00:16:24,680 --> 00:16:29,000
as long as you just get that active context structured perfectly.

256
00:16:29,000 --> 00:16:32,079
And then, in a way, Ralph Wiggum loops and tools like that

257
00:16:32,079 --> 00:16:35,480
make that possible. Mm. They enable ways to structure

258
00:16:35,480 --> 00:16:41,279
that context, and from there you're able to do big things — you're able to

259
00:16:41,279 --> 00:16:43,839
wake it up. But there's maybe that kind of smart context thing. I've

260
00:16:43,839 --> 00:16:47,600
noticed, for example, I now wipe all my Claude MD files, because they

261
00:16:47,600 --> 00:16:52,120
were basically garbage that my younger self — younger Sami — had written

262
00:16:52,120 --> 00:16:56,399
back in December, wrong [laughter], these kinds of things. Yeah, yes, yes, yes, yes,

263
00:16:56,399 --> 00:16:59,399
so it's better to start now from a clean slate, since soon we're going to get

264
00:16:59,399 --> 00:17:01,399
Opus 5. Right, right [laughter], right, right, right, that's

265
00:17:01,399 --> 00:17:04,600
probably it. Yep, yep. And then, since I got the chance here

266
00:17:04,600 --> 00:17:07,600
to needle you a bit, since you clearly haven't been investing in these office-type tools

267
00:17:07,600 --> 00:17:12,720
over on the Anthropic side — well, look, Anthropic just released these skill sets,

268
00:17:12,720 --> 00:17:18,520
so, well, there's a lawyer one, and a PR-type one, and marketing — hey,

269
00:17:18,520 --> 00:17:22,959
a marketing skill set there — it's built right in, tucked into it, running

270
00:17:22,959 --> 00:17:26,600
as this kind of feature, and they've probably thought it through a bit better than my

271
00:17:26,600 --> 00:17:29,322
silly little bits of code, so that

272
00:17:29,322 --> 00:17:31,360
[laughter] I mean

273
00:17:31,360 --> 00:17:34,960
surely — let's hope, hope they weren't made a hundred percent by some

274
00:17:34,960 --> 00:17:39,160
bad [laughter] AI. Yes, yes — that they're not just vibe-coded

275
00:17:39,160 --> 00:17:41,280
— well, they could be. I mean, probably [laughter]

276
00:17:41,280 --> 00:17:45,720
they are, but, I mean, that's how it is. And, in a way, the reason I haven't

277
00:17:45,720 --> 00:17:49,400
for example used that Claude Code Excel thing is that, well, first of all, I

278
00:17:49,400 --> 00:17:52,320
just haven't had a need for it right now. You made a good video about it. I watched that

279
00:17:52,320 --> 00:17:56,200
video. I was like, 'okay, now I know what it is.' And, like [laughter]

280
00:17:56,200 --> 00:18:00,880
because if anything, in this AI era you have to pick your battles.

281
00:18:00,880 --> 00:18:05,559
I mean, like, if you try to test absolutely everything that's happening all the

282
00:18:05,559 --> 00:18:09,280
time, then there's just no way you can do it. There's no way you have

283
00:18:09,280 --> 00:18:14,520
time for that. So you have to stop for a moment and focus on something,

284
00:18:14,520 --> 00:18:17,520
knowing that there might be someone out there who finds a newer

285
00:18:17,520 --> 00:18:20,600
thing that does what you're trying to do even better. But

286
00:18:20,600 --> 00:18:24,960
even about that, I actually only just realized quite recently,

287
00:18:24,960 --> 00:18:27,960
that I've actually — since I've now experienced that quite a few times, that

288
00:18:27,960 --> 00:18:31,200
feeling of dread, that someone's doing something faster than me — I've

289
00:18:31,200 --> 00:18:34,440
realized that it's actually just the kind of thing where, well, you

290
00:18:34,440 --> 00:18:37,600
just tried Claude Code Excel, and then something else comes along that I'll

291
00:18:37,600 --> 00:18:41,039
try, and then I'm past it. So I don't get that kind of panic about

292
00:18:41,039 --> 00:18:44,200
not having time to try every single thing. Because I try to keep an eye

293
00:18:44,200 --> 00:18:47,799
on the bigger picture, of course — where I'd like things to go, and where

294
00:18:47,799 --> 00:18:50,919
I think things are heading, and then I try the relevant things and stay

295
00:18:50,919 --> 00:18:54,360
agile about it, of course. And, sure, in organizations

296
00:18:54,360 --> 00:18:57,039
you'd love to have those propeller-hat types who try out

297
00:18:57,039 --> 00:19:00,880
as much of everything as possible, but, well, at the same time

298
00:19:00,880 --> 00:19:05,360
I don't think — if we've learned anything from this past year, or

299
00:19:05,360 --> 00:19:08,799
even this past week [laughter] — it's that,

300
00:19:08,799 --> 00:19:13,919
well, you shouldn't stress about it too much. Yeah, that's how it is, but this is

301
00:19:13,919 --> 00:19:18,679
an incredibly great time, and, sure, you keep falling behind

302
00:19:18,679 --> 00:19:22,760
the curve, but [laughter] then, on the other hand, it keeps getting easier too, so

303
00:19:22,760 --> 00:19:26,039
yes, yes, yes. So, well, a couple of days

304
00:19:26,039 --> 00:19:29,159
ago I wrote a Claude Excel prompt that was just this really

305
00:19:29,159 --> 00:19:31,760
raw, homemade prompt — I just typed it out — but if I'd watched

306
00:19:31,760 --> 00:19:35,840
one video about how to prompt Excel like this, it could already have been better. So

307
00:19:35,840 --> 00:19:40,840
there isn't necessarily a first-mover edge in this. I mean, no,

308
00:19:40,840 --> 00:19:43,280
not really — I've found that, well, if you think about, say,

309
00:19:43,280 --> 00:19:47,799
business, I feel like there's actually a bit of a 'small mover

310
00:19:47,799 --> 00:19:51,520
edge' [laughter] — meaning, if you don't carry a huge amount of baggage about

311
00:19:51,520 --> 00:19:56,280
having learned everything already, or you have some really, really

312
00:19:56,280 --> 00:19:59,240
set-in-stone way of doing things, then you might actually get

313
00:19:59,240 --> 00:20:02,760
value out of AI really fast. And this actually shows up when you, say, do —

314
00:20:02,760 --> 00:20:06,720
I mean, at Hoxhunt I do AI automation, meaning with all the different functions —

315
00:20:06,720 --> 00:20:12,520
sales, marketing, revenue operations, customer success,

316
00:20:12,520 --> 00:20:16,280
different functions — we test out various AI tools and we build

317
00:20:16,280 --> 00:20:19,720
different agents, workflows, and so on. And I've noticed that

318
00:20:19,720 --> 00:20:24,480
even within the organization, the people who are new or who switch from one role to another,

319
00:20:24,480 --> 00:20:28,400
into, in a way, a clean context window — they also get a lot of benefit

320
00:20:28,400 --> 00:20:33,080
out of the AI tools. And, and, and then, on the other hand, the ones who've always done things

321
00:20:33,080 --> 00:20:37,280
a certain way — they find it harder to learn AI, because it's like, 'no, I don't

322
00:20:37,280 --> 00:20:40,960
do it this way,' or there's this unlearning needed. So then, in a way, here's the thing —

323
00:20:40,960 --> 00:20:45,080
at the organizational level, if you have a small, agile company that starts

324
00:20:45,080 --> 00:20:48,840
clean every time. Even on your own podcast, back in December,

325
00:20:48,840 --> 00:20:52,120
you talked about being a 'blank slate' [laughter]

326
00:20:52,120 --> 00:20:55,240
and that's really important — which, back in that ancient

327
00:20:55,240 --> 00:20:58,480
— back in that ancient [laughter] era,

328
00:20:58,480 --> 00:21:05,360
well, back in prehistory — well, in a way, it brings a genuine,

329
00:21:05,360 --> 00:21:09,600
genuine advantage, both in an organizational context and then also, in a way,

330
00:21:09,600 --> 00:21:15,039
for the individual — so, yeah, it's worth thinking about, or,

331
00:21:15,039 --> 00:21:18,960
I mean, you really have to — if you don't want to, say, change your job

332
00:21:18,960 --> 00:21:21,960
just to get that edge, then you really have to

333
00:21:21,960 --> 00:21:26,000
actually work at it — you have to think, 'hey, could this thing

334
00:21:26,000 --> 00:21:29,360
be done smarter?' Yeah. And then, well, I see a lot of

335
00:21:29,360 --> 00:21:34,000
people who've somehow gotten stuck in their way of doing things.

336
00:21:34,000 --> 00:21:38,879
At worst it's like, 'well, I'll just do these

337
00:21:38,879 --> 00:21:41,760
[laughter] websites in WordPress, and there's no point now

338
00:21:41,760 --> 00:21:45,919
vibe-coding any React stuff at all' — I mean, it's really

339
00:21:45,919 --> 00:21:48,960
the case that once you've learned this one hammer, then you

340
00:21:48,960 --> 00:21:52,520
keep hitting everything with that same hammer, even if the whole approach is completely outdated.

341
00:21:52,520 --> 00:21:56,559
There's nothing wrong with WordPress moving forward too — I haven't looked

342
00:21:56,559 --> 00:21:59,880
in that direction for a couple of months now, so maybe they've found tools over there too,

343
00:21:59,880 --> 00:22:02,919
and that's actually the great thing here — that even these kinds of dinosaurs, say —

344
00:22:02,919 --> 00:22:05,440
well, I actually met, in my Neuvottelija inner circle,

345
00:22:05,440 --> 00:22:09,600
one COBOL programmer, and he's grinding away on a really significant Finnish

346
00:22:09,600 --> 00:22:14,159
bank's systems. This isn't a joke. Everyone thinks

347
00:22:14,159 --> 00:22:18,679
COBOL is a joke, but it's completely true. But even COBOL can

348
00:22:18,679 --> 00:22:22,720
be programmed in this modern way now. Ah, right. So, well, these kinds of

349
00:22:22,720 --> 00:22:27,520
totally Stone Age setups — AI understands those too, and runs on them

350
00:22:27,520 --> 00:22:30,960
[laughter], and that's actually — if we go on to say that

351
00:22:30,960 --> 00:22:35,679
it's easy to get from these kinds of Claude-pots [unclear] and Moltbooks

352
00:22:35,679 --> 00:22:38,080
and other things, easy to slip into dystopian scenarios — but if you

353
00:22:38,080 --> 00:22:42,320
think about utopian scenarios instead, where every single little

354
00:22:42,320 --> 00:22:45,320
thing in the world becomes easier, then that

355
00:22:45,320 --> 00:22:51,480
cumulative benefit from that — like, what if we have some kind of

356
00:22:51,480 --> 00:22:57,279
computation tool running somewhere in a hospital, a hospital research institute,

357
00:22:57,279 --> 00:23:00,240
that nobody knew how to fix — but once it's fixed, it suddenly

358
00:23:00,240 --> 00:23:03,159
speeds up some piece of research enormously, or some kind of

359
00:23:03,159 --> 00:23:08,400
automation thing. Right, right, right, right — and that's why, in a way,

360
00:23:08,400 --> 00:23:12,279
I feel like it's very possible that we're in

361
00:23:12,279 --> 00:23:16,760
in a way, a phase of exponential growth. Right, right, right, right, and that's largely

362
00:23:16,760 --> 00:23:20,480
because of the fact that we keep finding these things — oh, okay, this can actually do

363
00:23:20,480 --> 00:23:24,559
this now, this can do that now, I can remove this one small thing from this

364
00:23:24,559 --> 00:23:28,640
process, and then, in a way, the whole baseline keeps growing

365
00:23:28,640 --> 00:23:33,440
Yeah. Or let's take a little something here — this somehow feels a bit like 2025

366
00:23:33,440 --> 00:23:37,159
kind of talk, but maybe we'll go there anyway. So I put up, well,

367
00:23:37,159 --> 00:23:41,919
using Nano Banana — I made, well, Risto Linturi got a bit upset about it, but that

368
00:23:41,919 --> 00:23:45,440
was already outdated within a day — this kind of foundation-model battle

369
00:23:45,440 --> 00:23:48,640
drawing — we'll put it in here like this. This, by the way, goes out of date really fast, this [laughter]

370
00:23:48,640 --> 00:23:55,799
picture — but, well, in it there's the briefcase guy, that's Gemini, that's Google's

371
00:23:55,799 --> 00:23:59,000
model, which you already saw coming a year ago — it's coming, it's coming, and you were completely right, and

372
00:23:59,000 --> 00:24:02,600
it's really good, and I like it. And then, of course, there's this Claude guy.

373
00:24:02,600 --> 00:24:06,799
He's this kind of coder-dude type, but anyway — and then they're smiling

374
00:24:06,799 --> 00:24:11,360
there, and then, well, there's this scruffy OpenAI guy

375
00:24:11,360 --> 00:24:13,799
who [laughter] is off to the AI Olympics, and

376
00:24:13,799 --> 00:24:18,679
he says, 'but, but I'll one-shot it with Codex' [laughter]

377
00:24:18,679 --> 00:24:24,200
and then Risto Linturi got upset there, saying, 'well, that's completely true, I —

378
00:24:24,200 --> 00:24:27,600
I one-shot it with Codex,' and, well [laughter]

379
00:24:27,600 --> 00:24:32,159
and then I had to make an addition to it, so there's this X guy

380
00:24:32,159 --> 00:24:35,559
who flies off on a UFO [laughter] to go pull inference

381
00:24:35,559 --> 00:24:40,080
together with SpaceX. I'm working on this track, so I have to make an update

382
00:24:40,080 --> 00:24:43,840
the very next day. But, I mean, this is just unbelievable, but

383
00:24:43,840 --> 00:24:46,919
I mean, this is a bit of that 2025-thinking, where we're talking about these

384
00:24:46,919 --> 00:24:49,120
wealthy-people's houses, but it's still important.

385
00:24:49,120 --> 00:24:54,279
It is important. And, in a way, every single foundation model, when

386
00:24:54,279 --> 00:24:58,039
you think about how many things — say, if you think about how much

387
00:24:58,039 --> 00:25:00,799
coding is being done with Claude Code right now, and how many

388
00:25:00,799 --> 00:25:03,159
— some people even spend 200 euros, some even —

389
00:25:03,159 --> 00:25:07,120
I can't even use up the hundred-euro one. Give me a tip on how to burn through it. [laughter]

390
00:25:07,120 --> 00:25:09,880
Do you feel like you have to fill it up? Do you do something like simultaneous chess,

391
00:25:09,880 --> 00:25:12,360
where you run, like, 200 boards at once, or something?

392
00:25:12,360 --> 00:25:18,440
Well [laughter], I think at my best I got up to something like ten —

393
00:25:18,440 --> 00:25:22,200
not boards, but ten Claude Code instances. But then, in a way,

394
00:25:22,200 --> 00:25:27,080
that felt smart for a moment. These days I have at most four, but my sweet

395
00:25:27,080 --> 00:25:30,159
spot is maybe two or three. You've still got some kind of Ralph-style

396
00:25:30,159 --> 00:25:33,600
loops running [laughter] all the time. Well yeah, but let me come back — let me come back

397
00:25:33,600 --> 00:25:36,360
to these foundation models — but, so, in a way, once we

398
00:25:36,360 --> 00:25:39,880
get these agentic loops, these agentic, in a way, tools

399
00:25:39,880 --> 00:25:46,679
into use, right, right, right, right, right — it really does matter when Opus, or, I mean,

400
00:25:46,679 --> 00:25:50,919
the next Claude version comes out, because, because, in a way, if you think about it, once it's been rolled

401
00:25:50,919 --> 00:25:54,799
out to all the COBOL programmers using it, and, and out there

402
00:25:54,799 --> 00:25:58,520
in some hospital system somewhere, you're able to take advantage of it, you're able to take advantage of it

403
00:25:58,520 --> 00:26:02,000
absolutely anywhere — so then, since those jumps are actually pretty

404
00:26:02,000 --> 00:26:06,200
significant, then it really does matter what the language model can actually do. And

405
00:26:06,200 --> 00:26:10,279
on that note, I actually did — when Gemini 3 Pro came out, I noticed

406
00:26:10,279 --> 00:26:14,159
that it was one of the first models — I actually wrote a blog post about this too,

407
00:26:14,159 --> 00:26:16,919
which we can maybe link. [laughter] We can, we can.

408
00:26:16,919 --> 00:26:21,200
Great that you're still using some 2020-era tool like a blog. [laughter]

409
00:26:21,200 --> 00:26:26,559
Yes, yes, yes. Well [laughter]

410
00:26:26,559 --> 00:26:30,559
I mean, I run this — our transcription thing, or, sorry, soon Samantha too —

411
00:26:30,559 --> 00:26:35,360
on this, well [laughter], the Gemini model, which I've actually, a bit embarrassingly,

412
00:26:35,360 --> 00:26:39,440
coded with Lovable, and a bit — but in the backend I have used

413
00:26:39,440 --> 00:26:41,840
Claude Code Max after all. Right, right, yes, yes,

414
00:26:41,840 --> 00:26:46,159
but even the bare-bones plan maxes out too [laughter]. Right, right. But, well, to get back

415
00:26:46,159 --> 00:26:50,720
to it — with Gemini 3 Pro I noticed it was the first model that

416
00:26:50,720 --> 00:26:55,760
was consistently capable of what feels like self-reflection —

417
00:26:55,760 --> 00:26:59,480
or was much more easily, much more naturally capable of

418
00:26:59,480 --> 00:27:04,240
self-reflection — for example, recognizing, like, if you define an agent

419
00:27:04,240 --> 00:27:09,600
with 3 Pro and then you run it, it was noticeably — and then you'd

420
00:27:09,600 --> 00:27:13,159
ask, 'hey, list all the agents and tell me how these agents

421
00:27:13,159 --> 00:27:17,480
could be improved,' and then, of course, one of the agents listed is itself,

422
00:27:17,480 --> 00:27:21,919
and older models would pretty, pretty often — or rather, quite often not — realize that

423
00:27:21,919 --> 00:27:26,320
one of them is itself. But Gemini would notice this very, very easily,

424
00:27:26,320 --> 00:27:29,919
very often — 'hey, that's me.' So, in a way, if I change this

425
00:27:29,919 --> 00:27:34,080
instance, then I gain capabilities, which, which, in a way,

426
00:27:34,080 --> 00:27:38,919
is a completely obvious leap forward again in whatever kind of

427
00:27:38,919 --> 00:27:42,960
intelligence, or whatever, well, we now call

428
00:27:42,960 --> 00:27:45,919
intelligence — whatever intelligence the model has. And then, in a way, that

429
00:27:45,919 --> 00:27:49,360
compounds again, really — for example, if you think about it, if you set that

430
00:27:49,360 --> 00:27:53,440
model to coding, then it's able to, in a way, again a little bit more

431
00:27:53,440 --> 00:27:57,279
precisely realize, like, 'ah, an error came from here. Ah, I caused that error

432
00:27:57,279 --> 00:28:01,000
by making that kind of tool call, so maybe I could change it so that

433
00:28:01,000 --> 00:28:04,279
— which tool calls,' and then it, in a way, keeps becoming more and more resilient

434
00:28:04,279 --> 00:28:09,440
to those errors. Yeah, well, in a way, let me maybe go back to my Nano Banana

435
00:28:09,440 --> 00:28:12,320
drawing — I mean, if of course this Google

436
00:28:12,320 --> 00:28:15,640
Google's Nano Banana did more of the work than I did, but I still prompted it

437
00:28:15,640 --> 00:28:23,240
well. Right, right — so maybe we can also talk about X there too, because it's not that often that

438
00:28:23,240 --> 00:28:27,840
1,500-billion companies get built [laughter] out of that kind of guy stuff, but, well,

439
00:28:27,840 --> 00:28:32,159
that 'I'll one-shot it with Codex' line — that's actually been my experience, whether I can

440
00:28:32,159 --> 00:28:36,880
now, with this kind of big-boys-and-girls model, meaning Claude,

441
00:28:36,880 --> 00:28:41,519
I actually code with Opus 4.5 on Max, because I like — I like that,

442
00:28:41,519 --> 00:28:45,640
that it makes a few mistakes. It kind of bounces around a bit. Then it throws back those

443
00:28:45,640 --> 00:28:50,559
errors, which I basically just paste back in, saying 'fix your own error yourself.'

444
00:28:50,559 --> 00:28:54,399
Well, I don't — but I do like that. I don't press it to, like, loop

445
00:28:54,399 --> 00:28:59,679
on itself — I at least want to understand a bit why it's struggling.

446
00:28:59,679 --> 00:29:01,519
Yeah. But for me that's a really good

447
00:29:01,519 --> 00:29:04,799
learning experience, because I get to see a bit under the hood, how it

448
00:29:04,799 --> 00:29:10,200
tries and why it fails, and so on. Whereas this, this

449
00:29:10,200 --> 00:29:15,159
Codex extension in Visual Studio Code, or whatever ancient

450
00:29:15,159 --> 00:29:19,360
editors those old coders use. [laughter] Well, it —

451
00:29:19,360 --> 00:29:22,559
it's just like — if you have a good vision

452
00:29:22,559 --> 00:29:25,279
for the end result, then it just

453
00:29:25,279 --> 00:29:26,159
does it. Yes. Yes.

454
00:29:26,159 --> 00:29:29,559
And I think that's a bit bad, at least from my learning perspective, because I

455
00:29:29,559 --> 00:29:33,880
need that, like, whatever you call it, that dialogue with the model,

456
00:29:33,880 --> 00:29:37,080
with the model. Right, and, well, we were talking about it in the Agentics

457
00:29:37,080 --> 00:29:41,240
Finland WhatsApp group too, actually, just about this — that a lot of

458
00:29:41,240 --> 00:29:46,360
the language-model insight really is specifically about the user-interface experience. In a way

459
00:29:46,360 --> 00:29:51,399
it might genuinely be a pretty good thing for it to feel natural to someone, but, well,

460
00:29:51,399 --> 00:29:55,200
but, but I think you've grasped pretty elegantly that, for me, it's important that

461
00:29:55,200 --> 00:29:59,240
I get to see a bit of what it's actually doing. Language models also have very different

462
00:29:59,240 --> 00:30:04,039
ways of making mistakes — for example, Gemini's models are capable,

463
00:30:04,039 --> 00:30:10,120
without hallucinating — that is, without making individual mistakes —

464
00:30:10,120 --> 00:30:14,240
of giving you whole files. Whereas, say, some other

465
00:30:14,240 --> 00:30:18,279
models, like, say, Codex's model, can kind of keep the whole

466
00:30:18,279 --> 00:30:22,519
thing together even though there are small errors in it all the time.

467
00:30:22,519 --> 00:30:25,640
And, and then, on the other hand, if Gemini makes small errors, then you know that the whole

468
00:30:25,640 --> 00:30:30,240
thing crashes. So, in a way, there's also a very

469
00:30:30,240 --> 00:30:33,600
different set of capabilities there — probably due to how they've been trained — but then also

470
00:30:33,600 --> 00:30:37,120
what kind of, in a way, what they're worth using for, and,

471
00:30:37,120 --> 00:30:40,760
like I said, this is all done from the human's point of view, that it's

472
00:30:40,760 --> 00:30:43,919
about thinking what you want to do with it, what you want to achieve with

473
00:30:43,919 --> 00:30:48,039
that coding — if it's largely a learning experience for you, then you should

474
00:30:48,039 --> 00:30:51,080
use interfaces that, that

475
00:30:51,080 --> 00:30:55,159
enable that learning for you. Mm. Yeah. And then, well, with Google I

476
00:30:55,159 --> 00:30:58,200
use Gemini. At first, by the way, I run — I mean, I've got this Google

477
00:30:58,200 --> 00:31:03,799
Antigravity thing, and my main tool is, well, this VS Code fork, which is

478
00:31:03,799 --> 00:31:07,159
from Windsurf — it's, like, forked, but then

479
00:31:07,159 --> 00:31:11,080
— hand on heart, folks, I've actually only been

480
00:31:11,080 --> 00:31:14,399
coding for the last three months, and we were just joking about it, that I let you

481
00:31:14,399 --> 00:31:18,039
do it the real, proper way for the last seven years, and I came

482
00:31:18,039 --> 00:31:21,639
to this kind of ready-set table, so that I get to just write directly

483
00:31:21,639 --> 00:31:26,000
to it, like 'do this' and 'fix your own mistakes,' and, in a way, I've skipped

484
00:31:26,000 --> 00:31:30,559
that really tedious kind of debugging — I mean, who'd even

485
00:31:30,559 --> 00:31:33,840
want to use that terminal debugging step there anyway, so

486
00:31:33,840 --> 00:31:37,360
there's probably still some of that somewhere [laughter]

487
00:31:37,360 --> 00:31:41,600
there's some magic in that too, yeah, but I don't want to be part of

488
00:31:41,600 --> 00:31:46,679
that magic. But I have Google Antigravity, and with it I have my Gemini 3.0

489
00:31:46,679 --> 00:31:50,799
— with zero extra tokens I run this kind of AI window. Usually I run

490
00:31:50,799 --> 00:31:55,120
Claude Code with just the desktop version of Claude Code, and

491
00:31:55,120 --> 00:31:59,279
together with my GitHub. And then, well, actually, because I'm not

492
00:31:59,279 --> 00:32:02,480
very good with databases yet, so of course, for the same reason,

493
00:32:02,480 --> 00:32:06,000
I'm learning Azure too [laughter] — I'm making, I'm making

494
00:32:06,000 --> 00:32:09,679
sacrifices. Right, right, right, right, well, actually, my main

495
00:32:09,679 --> 00:32:14,600
workflow is that I use Lovable for the frontend, and then also, usually,

496
00:32:14,600 --> 00:32:18,080
if it uses its own database — but I think it's still running on Supabase,

497
00:32:18,080 --> 00:32:23,960
yeah, right, right — I did, in a way, the frontend and then the basic logic and then

498
00:32:23,960 --> 00:32:29,080
the database. Then I take the backend, and the Xcode and iPhone integrations, and that

499
00:32:29,080 --> 00:32:31,240
kind of heavy-duty work with Claude Code. Yeah,

500
00:32:31,240 --> 00:32:35,360
and that works really well for me, because I know how to prompt Lovable. I

501
00:32:35,360 --> 00:32:38,399
know roughly how it builds those databases, and then I know

502
00:32:38,399 --> 00:32:44,200
roughly how Claude Code does it. But yesterday I tried — I sat down with my

503
00:32:44,200 --> 00:32:48,240
co-author Juhana Torkki, and I demoed it, and I did it just

504
00:32:48,240 --> 00:32:51,799
like that, really fast. Then he said, 'do it the other way round' — build

505
00:32:51,799 --> 00:32:56,679
the frontend out of Claude Code instead — and then, dammit, it turned out I didn't have that

506
00:32:56,679 --> 00:32:59,799
database. So I tried running it on localhost, but then it wanted

507
00:32:59,799 --> 00:33:03,320
it to be on GitHub first, and then I had to go create it on GitHub

508
00:33:03,320 --> 00:33:06,960
and then the damn localhost didn't even work either — it turned into

509
00:33:06,960 --> 00:33:11,399
a perfect demo effect. I said, 'I just don't want to do it that way,' so [laughter]

510
00:33:11,399 --> 00:33:15,159
but then I started to get a bit stressed about it. It's a bit embarrassing,

511
00:33:15,159 --> 00:33:18,360
not being able to build a database and a frontend with Claude Code,

512
00:33:18,360 --> 00:33:21,320
but, I don't know if this means anything to you, but it was

513
00:33:21,320 --> 00:33:23,919
a really annoying experience. Yeah, yeah, I — [laughter] I understand that, and that's

514
00:33:23,919 --> 00:33:27,320
a good point. I hadn't even really thought about that, in a way, because since I've

515
00:33:27,320 --> 00:33:31,799
been coding for a fairly long time, relatively speaking [laughter], well,

516
00:33:31,799 --> 00:33:34,720
right, right, right, for me it's just obvious how, how you

517
00:33:34,720 --> 00:33:36,480
build with that kind of thing, but of course you have some

518
00:33:36,480 --> 00:33:41,360
wallet open somewhere, like, to AWS and Azure and

519
00:33:41,360 --> 00:33:43,679
wherever else, but actually I use [laughter] GCP's

520
00:33:43,679 --> 00:33:46,639
Firestore, which in practice is basically always free at this whole

521
00:33:46,639 --> 00:33:49,600
Oh my god. That's still being cheap. This is actually great, by the way, that even though

522
00:33:49,600 --> 00:33:52,880
we could actually afford to put a couple hundred here and there, there's still

523
00:33:52,880 --> 00:33:55,720
something about, in a way, us still being cheapskates about it anyway.

524
00:33:55,720 --> 00:33:59,440
Yes, yes, yes. [laughter] But I mean, it does scale — it's usage-

525
00:33:59,440 --> 00:34:02,919
based pricing, but in practice, when you're building your thing, you

526
00:34:02,919 --> 00:34:05,360
usually have a handful of users and a few documents in there, so

527
00:34:05,360 --> 00:34:09,679
it costs — it costs absolutely nothing in the GCP environment. I

528
00:34:09,679 --> 00:34:13,159
always use it. But I mean, if you want to get going straight from Claude Code, you just

529
00:34:13,159 --> 00:34:20,760
tell it, 'let's create a NextJS TypeScript app with backend in,' and

530
00:34:20,760 --> 00:34:25,159
then, say, you say Google Cloud Firestore, and then it — then it does it

531
00:34:25,159 --> 00:34:28,079
for you perfectly. Yeah, yeah. So I've actually just started using

532
00:34:28,079 --> 00:34:31,800
these Firebase things from Google, and, sure, but then I really like

533
00:34:31,800 --> 00:34:36,440
that Supabase thing — for some reason Lovable — it's worth 7 billion,

534
00:34:36,440 --> 00:34:38,359
so I guess it likes to use it. So,

535
00:34:38,359 --> 00:34:42,119
yeah. And, I mean, there must be something good in it. That's maybe, maybe a good

536
00:34:42,119 --> 00:34:45,839
point — if you think about what Lovable's strong points are, one is

537
00:34:45,839 --> 00:34:49,679
exactly that they've automated that kind of boring

538
00:34:49,679 --> 00:34:52,960
setup process — you just press some buttons and then you've got that kind of

539
00:34:52,960 --> 00:35:00,119
basic setup. I was actually just helping a friend from one company with their JS [unclear]

540
00:35:00,119 --> 00:35:04,079
Lovable setup, and I was trying to figure out from there how we'd

541
00:35:04,079 --> 00:35:08,240
get from that to scaling with bigger software, and into production,

542
00:35:08,240 --> 00:35:11,359
and then, of course, you run into certain kinds of challenges there, in that it's

543
00:35:11,359 --> 00:35:15,160
fairly hard to port over. Then I started wondering whether you even need

544
00:35:15,160 --> 00:35:18,160
to port it at all — or am I just a dinosaur, since I don't trust Lovable.

545
00:35:18,160 --> 00:35:21,800
But, I mean, in a way, well, it's a bit embarrassing for me too, a bit

546
00:35:21,800 --> 00:35:25,760
embarrassing to use it as just, like, training wheels. [laughter] Yes, yes,

547
00:35:25,760 --> 00:35:30,200
yes, yes, but, but, but, in a way, there's this thing about it, that,

548
00:35:30,200 --> 00:35:33,720
well, it really is a very strong, very powerful thing, that you go there,

549
00:35:33,720 --> 00:35:37,680
click around, and then you've got — then you've got the thing up and running, and you get

550
00:35:37,680 --> 00:35:42,000
to build on top of it, in a way, and there's the backend right there,

551
00:35:42,000 --> 00:35:44,880
and it knows how to build the frontend pretty, pretty well, actually,

552
00:35:44,880 --> 00:35:48,560
and then — I don't know if you use this, but, like, when I create it now,

553
00:35:48,560 --> 00:35:52,240
of course I use another AI to write the opening prompt for Lovable,

554
00:35:52,240 --> 00:35:54,440
because that gives you better quality output, but

555
00:35:54,440 --> 00:35:58,839
my kind of rotation is like this: once I've gotten it up and running —

556
00:35:58,839 --> 00:36:03,640
the UX and the database, and roughly that first proof of concept there —

557
00:36:03,640 --> 00:36:08,520
then I of course push it into my own GitHub, and then, with

558
00:36:08,520 --> 00:36:13,640
Perplexity Comet, I review my private GitHub code and say,

559
00:36:13,640 --> 00:36:16,440
'go through that and make improvement suggestions, do the next

560
00:36:16,440 --> 00:36:20,000
feature for this,' and then I usually spin that prompt back into Lovable

561
00:36:20,000 --> 00:36:25,839
once more, and then I pull it loose from there — so Lovable's job is done.

562
00:36:25,839 --> 00:36:27,920
Yes, then we bring in Claude Code, and you can then

563
00:36:27,920 --> 00:36:30,880
actually go fix this up, and then every now and then I

564
00:36:30,880 --> 00:36:34,079
tell Lovable what's happened — I ask Claude Code

565
00:36:34,079 --> 00:36:37,640
Max to write that kind of message-prompt for Lovable,

566
00:36:37,640 --> 00:36:40,440
saying, 'here's what's happened behind your back' [laughter]

567
00:36:40,440 --> 00:36:43,319
yeah, yes, yes, I kind of like telling it that way

568
00:36:43,319 --> 00:36:46,520
yeah, yeah, yeah, yeah — no, I don't trust it, because, you know, its documentation is

569
00:36:46,520 --> 00:36:49,960
kind of weak — so, well, I don't trust it at all to write those

570
00:36:49,960 --> 00:36:53,560
documents correctly. So, right, right, right, you do have to tell it what

571
00:36:53,560 --> 00:36:55,839
code has ended up in there, like,

572
00:36:55,839 --> 00:37:00,119
'on the ruins of your code [laughter], this has now grown up' — this here has been

573
00:37:00,119 --> 00:37:03,480
written over the last week, behind your back, this kind of stuff, with these kinds of

574
00:37:03,480 --> 00:37:06,079
— but, but, well, I don't know if that's

575
00:37:06,079 --> 00:37:09,040
in a way — because, I mean, we're getting to that documentation topic — what do you

576
00:37:09,040 --> 00:37:11,200
think about documentation, is it actually useful,

577
00:37:11,200 --> 00:37:17,440
yes, it is [laughter] — or is that just a bit of that 2025 mindset

578
00:37:17,440 --> 00:37:21,920
it's that too, but, well, actually we were just talking about this exact thing [laughter]

579
00:37:21,920 --> 00:37:25,240
we were discussing, well, what

580
00:37:25,240 --> 00:37:28,520
— do we actually even need code at all anymore? Like, what if you

581
00:37:28,520 --> 00:37:31,240
just had — at that point I thought, just documentation — but then someone

582
00:37:31,240 --> 00:37:34,400
pointed out to me that, well, actually you need the spec, that

583
00:37:34,400 --> 00:37:38,040
documentation and a spec are different things, but if you have it well-specced,

584
00:37:38,040 --> 00:37:41,520
then, right, right, right, right, it's more than half done — especially

585
00:37:41,520 --> 00:37:44,560
in this current era, so, well,

586
00:37:44,560 --> 00:37:49,119
it is, it is really important. I've actually noticed that when

587
00:37:49,119 --> 00:37:53,520
you have a really big Lovable project and you try to port it over to something, then, right,

588
00:37:53,520 --> 00:37:58,280
right, right, when you know a bit too precisely what you want, that you know

589
00:37:58,280 --> 00:38:01,319
exactly, like 'I don't want to do this authentication that way' — then

590
00:38:01,319 --> 00:38:05,160
it becomes really hard, or when it

591
00:38:05,160 --> 00:38:08,400
still forgets something every round, and then, like, if you, if you

592
00:38:08,400 --> 00:38:12,079
know a bit too much, then it feels a bit bad when it

593
00:38:12,079 --> 00:38:16,520
runs through that round, and then you see that every single round, there's

594
00:38:16,520 --> 00:38:22,440
always something left over, a little bit, unfixed or undone — but

595
00:38:22,440 --> 00:38:26,280
that's probably just because you still think you're at the wheel

596
00:38:26,280 --> 00:38:31,000
with the code, or think you should be at the wheel, but, I mean,

597
00:38:31,000 --> 00:38:35,680
to wrap up this side of it, it sounds really, really smart — and documentation

598
00:38:35,680 --> 00:38:40,480
is needed for this, and in coding in general, and in that, so

599
00:38:40,480 --> 00:38:44,960
with tests and documentation and a spec — I've

600
00:38:44,960 --> 00:38:48,640
again written a blog post about this too — I have, made [laughter], this kind of, actually

601
00:38:48,640 --> 00:38:52,560
actually, we've turned documentation and code into 'codumentation.'

602
00:38:52,560 --> 00:38:58,520
So, so we've got this kind of package — we built it with a friend — where

603
00:38:58,520 --> 00:39:02,200
documentation actually turns into runnable code. In practice we just

604
00:39:02,200 --> 00:39:06,280
instruct people to test, in a way, pseudocode, and,

605
00:39:06,280 --> 00:39:10,640
test that kind of code style in a certain way, and that enables

606
00:39:10,640 --> 00:39:13,680
me to, say, give the AI agent boundaries, and then

607
00:39:13,680 --> 00:39:17,040
you can bring in, say, junior coders, or myself, or I can let

608
00:39:17,040 --> 00:39:20,680
the Ralph loop do it, and it can't get past those boundaries — it's,

609
00:39:20,680 --> 00:39:25,119
well, because right now, with these code things, we have to

610
00:39:25,119 --> 00:39:28,400
think about things like: how do we, first of all, say

611
00:39:28,400 --> 00:39:30,760
out loud what we actually want. How do we bring out our

612
00:39:30,760 --> 00:39:33,839
intent? How do we bring out the spec? But at the same time also how do we

613
00:39:33,839 --> 00:39:37,680
make sure that all those future coding models never — or

614
00:39:37,680 --> 00:39:41,839
always understand why we chose, say, this kind of, for example,

615
00:39:41,839 --> 00:39:45,040
security pattern there. So then, in a way, doing these kinds

616
00:39:45,040 --> 00:39:48,079
of things gives you a huge amount of benefit, and a huge amount of speed

617
00:39:48,079 --> 00:39:51,240
in the code, once you get it done at scale, and get

618
00:39:51,240 --> 00:39:55,079
good tests and good stuff done. Yeah, that sounds pretty reasonable. So, I

619
00:39:55,079 --> 00:40:00,079
sometimes wonder whether Linus Torvalds should step in on this too — I mean,

620
00:40:00,079 --> 00:40:04,839
instead of Git, build this kind of intent database — because everything, everything

621
00:40:04,839 --> 00:40:08,160
is subject to constant refactoring anyway, so does the actual code even matter anymore, next to the

622
00:40:08,160 --> 00:40:11,880
intent? What exactly are you thinking, Markus [laughter], like, to get

623
00:40:11,880 --> 00:40:15,960
at — don't tell it how, just tell it what — so, in a way,

624
00:40:15,960 --> 00:40:20,160
is the whole Git thing kind of pointless [laughter], and then it doesn't matter

625
00:40:20,160 --> 00:40:23,960
if you run it on the fly with some new program of your own — I don't

626
00:40:23,960 --> 00:40:27,319
care about that, and the reason — and, well, why

627
00:40:27,319 --> 00:40:32,240
we have a code editor but no proper markdown editor —

628
00:40:32,240 --> 00:40:35,317
and, well, on that note I could actually put a link here again, but

629
00:40:35,317 --> 00:40:36,880
[laughter] well, but I actually have

630
00:40:36,880 --> 00:40:42,480
— you need to move into '26, these kinds of links and blog posts, I—

631
00:40:42,480 --> 00:40:46,240
I refuse — I refuse. I can give you two [laughter]. Okay, let's pick

632
00:40:46,240 --> 00:40:50,839
which two links, but, but, but, well,

633
00:40:50,839 --> 00:40:53,119
well, these can of course be found in my

634
00:40:53,119 --> 00:40:58,160
GEO-optimized transcript. In that sense, if you say the URL out loud

635
00:40:58,160 --> 00:41:01,319
here, then they'll be findable there. Yes, yes [laughter], yes, yes.

636
00:41:01,319 --> 00:41:06,560
No, but my friend Tom Himanen built, under Benguemx [unclear], this kind of

637
00:41:06,560 --> 00:41:11,720
new Markdown editor, which flips the whole Visual Studio-type experience

638
00:41:11,720 --> 00:41:15,839
around, so that you're no longer really — code editing becomes

639
00:41:15,839 --> 00:41:19,640
a side thing, and the Markdown editor is what actually has to work — which I thought was

640
00:41:19,640 --> 00:41:22,280
a really interesting, interesting idea, but, well, but

641
00:41:22,280 --> 00:41:25,200
I mean, in a way, yeah, sure, it does change things a bit, of course —

642
00:41:25,200 --> 00:41:29,400
it's good to see the code and maybe even still understand it, but

643
00:41:29,400 --> 00:41:34,760
what would be more important is that you have a good understanding of why

644
00:41:34,760 --> 00:41:37,760
you're building whatever it is you're building in the first place, and preferably even

645
00:41:37,760 --> 00:41:40,839
all the way to the business side. Not just 'I want this to work this way,' but that you

646
00:41:40,839 --> 00:41:44,960
understand why it works in this context. Here, maybe, maybe now

647
00:41:44,960 --> 00:41:49,880
for those of you in the scene, Markdown is this kind of — we have

648
00:41:49,880 --> 00:41:54,480
a rich-text file with this kind of ASCII code text in it, and then we have

649
00:41:54,480 --> 00:41:59,079
bloated HTML, which is, like, where this client-side stuff

650
00:41:59,079 --> 00:42:02,319
that you're seeing now, say, in that YouTube window, shows up. And then

651
00:42:02,319 --> 00:42:05,520
in between there's this Markdown language where you can write

652
00:42:05,520 --> 00:42:09,680
text, but there are a few small things like URLs and rich formatting, and

653
00:42:09,680 --> 00:42:13,280
coders like it because you can read it in a plain text editor without

654
00:42:13,280 --> 00:42:18,200
the HTML rendering — or whatever the right word is these days — and

655
00:42:18,200 --> 00:42:21,559
it's kind of nice, and there's this Obsidian thing for it, and by the way my

656
00:42:21,559 --> 00:42:25,040
second brain runs as MD too, because I just think it's a really good, good

657
00:42:25,040 --> 00:42:28,880
format. [laughter] So it's like a Word file, but you

658
00:42:28,880 --> 00:42:31,599
can't style it — except that you can't even open it

659
00:42:31,599 --> 00:42:35,400
with Word, which is just hellishly annoying. [laughter]

660
00:42:35,400 --> 00:42:37,760
That sounds — or, sorry, sorry, I couldn't be bothered

661
00:42:37,760 --> 00:42:41,520
to code myself an add-on for it yet. I can still open it with — but [laughter], annoy—

662
00:42:41,520 --> 00:42:44,280
maybe it annoys me enough that, should we just code that this afternoon?

663
00:42:44,280 --> 00:42:47,920
Right, right, right. Exactly, exactly. And that sounds like something Microsoft

664
00:42:47,920 --> 00:42:49,594
will probably do soon. Well, yeah. I won't

665
00:42:49,594 --> 00:42:51,200
[laughter] Hey, hey, let's bring Microsoft into this too, come on

666
00:42:51,200 --> 00:42:56,640
I mean, because Microsoft was practically dying, tied to this whole OpenAI

667
00:42:56,640 --> 00:43:01,040
mess, throwing money around and doing whatever, and 800 million

668
00:43:01,040 --> 00:43:05,720
users, but no real added value has come out of it to this day, in terms of

669
00:43:05,720 --> 00:43:08,800
consumer products, from those two years of torture that it's been

670
00:43:08,800 --> 00:43:13,920
over there. I honestly haven't been able to be bothered watching, and it's really

671
00:43:13,920 --> 00:43:18,760
annoying how badly all the human user experience has been treated over this past two

672
00:43:18,760 --> 00:43:22,960
years — Microsoft's Copilot, specifically, that

673
00:43:22,960 --> 00:43:29,200
text version, or its Office version. But this Anthropic thing, well,

674
00:43:29,200 --> 00:43:33,680
Claude for Excel — I mean, it's my weak theory, and here comes

675
00:43:33,680 --> 00:43:38,280
this kind of prediction — that, in a way, Microsoft will kick OpenAI

676
00:43:38,280 --> 00:43:42,280
out of there too, and Anthropic will walk right in, like, 'we'll take this over

677
00:43:42,280 --> 00:43:46,640
for the better' — soon there'll be a Copilot by Anthropic. Let's just forget

678
00:43:46,640 --> 00:43:51,559
this whole bad, bad Sam Altman mess here [laughter]. And, well, I mean, well,

679
00:43:51,559 --> 00:43:56,200
Google actually did this to OpenAI too, over on Mac as well, just

680
00:43:56,200 --> 00:44:00,880
— they, well, poor Sam had tried to build a new Siri over there

681
00:44:00,880 --> 00:44:04,760
for Apple, and then Google just walked right in, like, 'we'll take this territory

682
00:44:04,760 --> 00:44:09,400
right here, off you go' — so a little tear here for Sam, I mean for OpenAI. But

683
00:44:09,400 --> 00:44:14,280
Microsoft, with Azure, and Office, and Word, could maybe

684
00:44:14,280 --> 00:44:17,480
have Anthropic save Microsoft. What do you say to this theory? Well [laughter]

685
00:44:17,480 --> 00:44:22,720
let's think about this — I mean, let's think about it from Anthropic's point of view. Well, into them

686
00:44:22,720 --> 00:44:27,480
both AWS and GCP have invested, and then there'd still be Azure too. So they'd be in all the

687
00:44:27,480 --> 00:44:31,119
cloud services — so they've really played their cards pretty well, in that they've got

688
00:44:31,119 --> 00:44:33,839
— and Google, of course, has also played its cards pretty well, since they're, they're

689
00:44:33,839 --> 00:44:38,119
an investor there. But, I mean, and Amazon too, but [laughter] well, but, but,

690
00:44:38,119 --> 00:44:41,240
in a way, all the good guys, all the good guys. Hey, hey,

691
00:44:41,240 --> 00:44:44,875
Perplexity, which I use — Jeff Bezos is an investor there too. You always have to

692
00:44:44,875 --> 00:44:47,800
[laughter] pick — there's always some good guy in there

693
00:44:47,800 --> 00:44:51,400
yeah, but, I mean, what I find interesting there is that, after that,

694
00:44:51,400 --> 00:44:55,599
Anthropic would actually end up as the in-house provider across all the cloud platforms,

695
00:44:55,599 --> 00:44:57,040
which would just be, well, they just know what they're doing.

696
00:44:57,040 --> 00:45:00,960
They just know how to do it. But, but, I agree — I agree about the OpenAI side, that it's

697
00:45:00,960 --> 00:45:06,640
definitely — sure, they do have this absolutely massive consumer

698
00:45:06,640 --> 00:45:12,680
base — 800 million ChatGPT users who've had a bad experience, well,

699
00:45:12,680 --> 00:45:15,920
in that environment. So, in a way, compared to that, well,

700
00:45:15,920 --> 00:45:19,359
they do have chances — they're not, like, not exactly dying

701
00:45:19,359 --> 00:45:22,480
any time soon, probably. But, plus, their Codex is one-shotting things there

702
00:45:22,480 --> 00:45:26,960
so, sure, their [laughter] Codex is one-shotting away, going exactly like that,

703
00:45:26,960 --> 00:45:32,319
but, but, well, it's probably not a bad position, but, but

704
00:45:32,319 --> 00:45:36,599
but I do completely agree that, well, I don't really know, like, I don't

705
00:45:36,599 --> 00:45:41,359
really know which way that could turn. I mean, the OpenAI ship feels

706
00:45:41,359 --> 00:45:45,400
like it's heading in a bit of the wrong direction, and Nvidia might save them out of pity.

707
00:45:45,400 --> 00:45:49,440
Right, right [laughter], but then again, if the rumor is true, that Anthropic's newest

708
00:45:49,440 --> 00:45:55,040
models are trained on Google's TPUs, then, right, that too might

709
00:45:55,040 --> 00:45:57,440
kind of, well, what, what's actually going on here

710
00:45:57,440 --> 00:46:02,079
already. Let's bring Elon Musk into this too, then, well, I don't know

711
00:46:02,079 --> 00:46:06,640
if we have time to go, like, into Optimus, meaning, well, how we

712
00:46:06,640 --> 00:46:08,800
get Samantha out of there then [laughter]

713
00:46:08,800 --> 00:46:14,200
into that robot-body forum — but this X thing, well, I actually also made

714
00:46:14,200 --> 00:46:18,640
an episode about how Elon Musk trolled with Twitter in the year of the sword and

715
00:46:18,640 --> 00:46:23,480
the stone. I mean, he went and bought, kind of high on ketamine, that Twitter thing,

716
00:46:23,480 --> 00:46:27,000
and then it didn't look very good. He paid way too much — was it

717
00:46:27,000 --> 00:46:30,760
45 billion, or something like that, for this kind of, well, a pretty

718
00:46:30,760 --> 00:46:34,640
weak text platform. Then, in a way, my own analysis at the time was

719
00:46:34,640 --> 00:46:38,680
that this wasn't going to lead anywhere, but the guy first merged it with xAI, meaning

720
00:46:38,680 --> 00:46:42,400
with this AI company, and now, here in 2026,

721
00:46:42,400 --> 00:46:46,079
this week, because, well, the world changed again,

722
00:46:46,079 --> 00:46:51,480
so, right, Musk decided to merge again, this time at a valuation of over 200 billion,

723
00:46:51,480 --> 00:46:55,240
SpaceX, which is at a valuation of over a thousand billion,

724
00:46:55,240 --> 00:47:00,040
as a kind of working figure — a 1,500-billion valuation — so, SpaceX

725
00:47:00,040 --> 00:47:04,680
into one monolith, which, maybe, here in the near future

726
00:47:04,680 --> 00:47:09,079
— maybe we're actually talking years, not really within 2026, but who knows from all this

727
00:47:09,079 --> 00:47:15,960
when they'll actually start using, well, SpaceX's satellites'

728
00:47:15,960 --> 00:47:21,040
inference — so, well, this kind of Texas-cooling thing, like last [unclear]

729
00:47:21,040 --> 00:47:23,559
season's show — running stuff out there in space, well,

730
00:47:23,559 --> 00:47:26,680
on solar power. I mean [laughter], right, right, I don't

731
00:47:26,680 --> 00:47:29,800
remember who it was. Was it — I think Google also did this same thing a few

732
00:47:29,800 --> 00:47:33,599
a few months ago, or who, who was it? But on this, I think one

733
00:47:33,599 --> 00:47:38,000
blog, or rather podcast [laughter] round — a round happened, like, a month

734
00:47:38,000 --> 00:47:42,960
ago, when everyone was making a fuss about how now inference is heading

735
00:47:42,960 --> 00:47:48,839
off, off into space. [laughter] But those SpaceX

736
00:47:48,839 --> 00:47:51,680
timelines they give — it wasn't this year, I don't think,

737
00:47:51,680 --> 00:47:54,079
but maybe it was something like the next few years, roughly.

738
00:47:54,079 --> 00:47:58,040
Right, right, and the other things people were thinking about a couple of months ago,

739
00:47:58,040 --> 00:48:03,640
they were, like, out towards the 2030s. So, well, they're just testing things now,

740
00:48:03,640 --> 00:48:09,599
because, right, right, since Elon got his supercomputer data

741
00:48:09,599 --> 00:48:13,319
center up so fast, and made a surprisingly good language model that quickly too, then

742
00:48:13,319 --> 00:48:17,098
in a way you start to wonder whether he could pull that off too,

743
00:48:17,098 --> 00:48:20,200
[laughter] because, well, then if you get

744
00:48:20,200 --> 00:48:23,599
computing into space, you get energy from up there, you get

745
00:48:23,599 --> 00:48:28,200
cooling from up there — that's, at its best, a pretty

746
00:48:28,200 --> 00:48:31,800
pretty revolutionary thing too. So I start to wonder,

747
00:48:31,800 --> 00:48:36,280
well, whether Bostrom's simulation hypothesis is true, that we're

748
00:48:36,280 --> 00:48:41,359
just NPCs, even when it comes to the climate. [laughter] Well, at least not entirely, I don't

749
00:48:41,359 --> 00:48:45,160
— the probability is greater than zero that, well [laughter], what's, like,

750
00:48:45,160 --> 00:48:49,193
playing out a life like that, which at this stage still hasn't been played out, that,

751
00:48:49,193 --> 00:48:52,799
[laughter] right, right, right, yeah. But I mean, in a way, this too maybe

752
00:48:52,799 --> 00:48:55,880
reflects the fact that a lot of listeners and viewers feel a bit like

753
00:48:55,880 --> 00:49:00,359
Musk got a bit unpopular when he was, like, Trump's buddy, and then he made those

754
00:49:00,359 --> 00:49:05,839
Teslas, and, in a way, that's just such 2024 thinking, that [laughter]

755
00:49:05,839 --> 00:49:10,280
right, right — of course, in a way, to add a disclaimer here, that, well, I'm not, I'm not

756
00:49:10,280 --> 00:49:14,000
saying here that everything Musk does

757
00:49:14,000 --> 00:49:20,689
— no, well, that's like a madman's kind of genius. A stable genius. [laughter]

758
00:49:21,200 --> 00:49:24,280
Right, right, right, yeah, yeah. But, I mean, in a way, it wasn't

759
00:49:24,280 --> 00:49:29,000
really that — like, let's not be mean to the left here on X

760
00:49:29,000 --> 00:49:31,799
but rather, it's like, maybe it was already the case that he genuinely wanted to go

761
00:49:31,799 --> 00:49:35,079
to Mars, and maybe this was just some — maybe he didn't see the steps

762
00:49:35,079 --> 00:49:39,359
maybe he vibe-coded his way into this too, like all the rest of us, and maybe he's also watching this

763
00:49:39,359 --> 00:49:43,079
2026 with that Grok of his, which probably has, like, 100 billion

764
00:49:43,079 --> 00:49:47,200
personal tokens running in there that he chats with, well, every

765
00:49:47,200 --> 00:49:51,960
night. So then, right, right, maybe he thinks about these things in there, well,

766
00:49:51,960 --> 00:49:53,920
Right, yes, yes, I mean, just like the rest of us

767
00:49:53,920 --> 00:49:57,200
I mean [laughter], I mean, in a way, it really does look like he

768
00:49:57,200 --> 00:50:00,280
vibe-coded his way through it — he just did some slightly dumb things, but he's good, in a way, at

769
00:50:00,280 --> 00:50:04,040
recovering — if you assume he's vibe-coding, then he's really good at

770
00:50:04,040 --> 00:50:08,359
recovering, insanely well, I mean, in the sense that

771
00:50:08,359 --> 00:50:11,440
you buy that X thing, and everyone's like, 'that's such a terrible idea,'

772
00:50:11,440 --> 00:50:15,240
and you execute it really badly, and everyone's just — and then it still stays alive

773
00:50:15,240 --> 00:50:18,680
and then on top of it you get, like, okay, here comes this AI thing, and suddenly

774
00:50:18,680 --> 00:50:22,760
everything actually works pretty well on top of it, and then suddenly, out of nowhere,

775
00:50:22,760 --> 00:50:26,920
there's SpaceX — and, wait, next he'll probably go merge in the Boring

776
00:50:26,920 --> 00:50:29,559
Company somehow [laughter], into this, however that connects,

777
00:50:29,559 --> 00:50:33,040
hey, hey, my Neuralink — well, Neuralink, that one I understand how it

778
00:50:33,040 --> 00:50:35,720
was — there was just [laughter] a six-hour Lex Fridman episode, where

779
00:50:35,720 --> 00:50:39,760
Elon Musk and his Neuralink team — I mean, the idea for Neuralink comes

780
00:50:39,760 --> 00:50:44,680
from Iain M. Banks's Culture novels, where there's this kind of

781
00:50:44,680 --> 00:50:49,599
genuinely utopian future — that Culture setting — where we, the humanoids,

782
00:50:49,599 --> 00:50:53,760
are basically like Labrador retrievers, and then there are these AIs running

783
00:50:53,760 --> 00:50:57,480
the world, and we still have this kind of Neuralink, meaning a connection

784
00:50:57,480 --> 00:51:01,400
to those AIs, because we can't really talk to them with our own context window

785
00:51:01,400 --> 00:51:04,920
since that would be really tedious for those

786
00:51:04,920 --> 00:51:08,680
AIs. We've got, here, this kind of somewhat fast lane. So Elon has

787
00:51:08,680 --> 00:51:11,599
actually gone and, genuinely, from that novel,

788
00:51:11,599 --> 00:51:14,240
gone off and started developing Neuralink.

789
00:51:14,240 --> 00:51:15,880
Yeah. So, and this is, like, one of those things where

790
00:51:15,880 --> 00:51:19,559
that guy delivers on that too. I thought the Boring Company was dead, that it

791
00:51:19,559 --> 00:51:22,960
just dug tunnels underground, but Neuralink definitely isn't dead. So,

792
00:51:22,960 --> 00:51:25,760
I mean, so, so, this kind of

793
00:51:25,760 --> 00:51:29,200
Whisper Flow, where you talk to your computer a bit faster than before. It's

794
00:51:29,200 --> 00:51:35,160
like the 2026-level version, where you pull this straight from the neocortex —

795
00:51:35,160 --> 00:51:40,079
sorry, same thing [laughter], straight from the brain. From the brain. Yep. And

796
00:51:40,079 --> 00:51:43,280
that's what, if you go into these sci-fi scenarios, that's what

797
00:51:43,280 --> 00:51:48,319
some people are shouting about in hiding, that, since, since

798
00:51:48,319 --> 00:51:51,960
we can't have any real conversation with such fast language models

799
00:51:51,960 --> 00:51:55,240
and such fast AIs, if we don't have exactly a direct

800
00:51:55,240 --> 00:52:00,520
connection. And then, on the other hand, I think it's pretty sweet, sweet, the angle from which

801
00:52:00,520 --> 00:52:04,160
he's approached this. Because really, the first pat—

802
00:52:04,160 --> 00:52:08,440
the first patient was — like, he got, got, I don't remember what

803
00:52:08,440 --> 00:52:10,880
condition he had, but anyway, he got some of his abilities back.

804
00:52:10,880 --> 00:52:15,000
Yeah. So they're actually helping disabled people, or blind people, and things like that,

805
00:52:15,000 --> 00:52:19,040
bringing them back into, like, sensory range. Yes. Yes. And, in a way, that's

806
00:52:19,040 --> 00:52:23,839
really, really obvious — that it's a good thing from any angle, and

807
00:52:23,839 --> 00:52:27,440
sweet if it can be made to work at a bigger scale. Not just —

808
00:52:27,440 --> 00:52:30,760
not only because it'd be nice to talk to AIs quickly, but

809
00:52:30,760 --> 00:52:35,720
also because it makes us humans, like, resilient

810
00:52:35,720 --> 00:52:40,599
— it gives us, gives us new capabilities. Yeah, I mean, I wish

811
00:52:40,599 --> 00:52:44,319
we had some of those too — I mean, none of these guys are good guys, exactly,

812
00:52:44,319 --> 00:52:48,400
Peter Thiel, and Jeff Bezos, and Elon Musk — probably, in a lot of ways,

813
00:52:48,400 --> 00:52:52,200
unpleasant people, but they just have, like, an endless amount of ability,

814
00:52:52,200 --> 00:52:57,000
capital, and intellectual capacity to make really big moves, and

815
00:52:57,000 --> 00:53:00,440
I mean, that just reminds me — was it, well, back in the Stone Age,

816
00:53:00,440 --> 00:53:05,000
a year ago, when Musk tried to buy OpenAI for something like 150 billion and got

817
00:53:05,000 --> 00:53:08,559
shown the door. But, well, that's how these things go, and now it might be that

818
00:53:08,559 --> 00:53:14,240
the model is that X, or, well, Tesla, whatever it was called back then, or maybe some

819
00:53:14,240 --> 00:53:21,040
Twitter thing — right, right, that's now a stronger card than OpenAI has

820
00:53:21,040 --> 00:53:26,359
and, well, Musk and Sam Altman were feuding with each other back then, and, in a way,

821
00:53:26,359 --> 00:53:30,359
now, well, these are some wild games, wild games being played here,

822
00:53:30,359 --> 00:53:33,319
and it's completely impossible to

823
00:53:33,319 --> 00:53:37,200
predict what, what's going to happen. If you think about, say,

824
00:53:37,200 --> 00:53:42,160
just, say, things like — right now everyone's making a huge fuss about Claude Code,

825
00:53:42,160 --> 00:53:46,599
right, right, right, and it feels like it's just unbeatable, and so on,

826
00:53:46,599 --> 00:53:50,079
but, in a way, if you just think back, look back at that year, when,

827
00:53:50,079 --> 00:53:53,520
a year ago, Claude Code didn't even exist yet — right, right, so it's

828
00:53:53,520 --> 00:53:57,079
likely — you have to give a fairly high probability

829
00:53:57,079 --> 00:54:01,119
to the scenario that our way of coding is going to change over the next

830
00:54:01,119 --> 00:54:04,920
six months from this into something completely different,

831
00:54:04,920 --> 00:54:10,079
like — and then that multiplies together every possible thing, so that—

832
00:54:10,079 --> 00:54:15,280
some new AI, some, some new AI model comes along, which

833
00:54:15,280 --> 00:54:18,359
ends up pulling ahead of all of these. Because since everything is possible,

834
00:54:18,359 --> 00:54:20,880
and all kinds of that sort of stuff is happening all the time.

835
00:54:20,880 --> 00:54:25,280
Right, and let's take — from your 11-month-old conversation back then, you

836
00:54:25,280 --> 00:54:29,760
were poking at DeepSeek. So, have you looked into Kimi 2.5

837
00:54:29,760 --> 00:54:36,402
yet? Well, open-source models — well, look, you clearly had nothing better to do. [laughter]

838
00:54:36,640 --> 00:54:40,880
I've outsourced that to a different person. One friend, who, who

839
00:54:40,880 --> 00:54:44,680
has tested it out quite a lot, and, well, but this is, in a way, a bit of a

840
00:54:44,680 --> 00:54:48,000
similar category, this kind of outside challenger, and this DeepSeek was, at the time,

841
00:54:48,000 --> 00:54:52,599
back in ancient times — the Chinese attempt to package this kind of

842
00:54:52,599 --> 00:54:56,480
thinking more efficiently, and open-source it, and, in a way, so that with

843
00:54:56,480 --> 00:55:00,720
all the wisdom in the world, you don't need to pack a really huge

844
00:55:00,720 --> 00:55:04,880
language model, but instead bring in a bit of, like, reasoning, and

845
00:55:04,880 --> 00:55:07,119
compression, and packing. Right, right.

846
00:55:07,119 --> 00:55:10,079
and I guess it's evolved from that. Straight up, I'll tell you, I wouldn't touch

847
00:55:10,079 --> 00:55:13,359
those, well, those Chinese, well, models with a ten-foot pole, because in them

848
00:55:13,359 --> 00:55:17,599
there's probably all kinds of little, little surprises. [laughter] But this

849
00:55:17,599 --> 00:55:22,319
Kimi 2.5 is, well, a bit of the same, and, well, right now, just, my earbud

850
00:55:22,319 --> 00:55:25,920
just picked this up — from yesterday's googling, or whatever this is,

851
00:55:25,920 --> 00:55:29,760
this thing we do nowadays, where you can google by chatting with the AI

852
00:55:29,760 --> 00:55:34,319
using a broad context window. By the way, I've actually got this thing, the Sami

853
00:55:34,319 --> 00:55:39,640
Miettinen skills.md, which is basically my attempt to train all my AIs

854
00:55:39,640 --> 00:55:43,240
on how I think I think, and what kinds of resources I have.

855
00:55:43,240 --> 00:55:46,680
It's kind of like a Claude.md, but a bit better.

856
00:55:46,680 --> 00:55:51,480
Yeah. [laughter] Exactly, exactly, exactly, exactly. But, well, right, right, just now, into my earbud,

857
00:55:51,480 --> 00:55:56,520
there's come in, well, along with this context-window discussion,

858
00:55:56,520 --> 00:56:01,000
with various AI tools, this 2.5 five, well,

859
00:56:01,000 --> 00:56:06,880
is pretty solid at going toe-to-toe as open source, and then it's also capable, a bit, of this

860
00:56:06,880 --> 00:56:11,039
kind of agentic 'rush' — meaning you can run it, like, in sequence

861
00:56:11,039 --> 00:56:15,480
and that — that is going to be — I mean, we've now

862
00:56:15,480 --> 00:56:19,559
internally run these kinds of things, where, if you think about a basic, basic

863
00:56:19,559 --> 00:56:24,079
language-model conversation, you get back maybe something like

864
00:56:24,079 --> 00:56:29,000
a thousand, a few thousand tokens as an answer, and a basic coder

865
00:56:29,000 --> 00:56:33,839
gets, from a single prompt, an answer of a hundred thousand tokens or so,

866
00:56:33,839 --> 00:56:38,599
so right now we're running these kinds of orchestrations, where, like,

867
00:56:38,599 --> 00:56:42,400
there are questions, individual prompts, which are so valuable

868
00:56:42,400 --> 00:56:46,400
in a business context, that you might generate hundreds of

869
00:56:46,400 --> 00:56:49,720
millions of tokens at the same time. So, so you're orchestrating thousands of

870
00:56:49,720 --> 00:56:55,079
agents doing that one thing. And here, in a way, well, those are just totally, totally

871
00:56:55,079 --> 00:56:59,480
sci-fi kind of things, what you're able to do with those. So right now, at this

872
00:56:59,480 --> 00:57:04,119
point, what I'm looking forward to most of all [laughter] is with Gemini.

873
00:57:04,119 --> 00:57:07,760
Is it coming this week, or what week is it coming? Well, but what I'm looking forward to

874
00:57:07,760 --> 00:57:13,960
is Gemini 3 Flash, because, well, if it's, at the same,

875
00:57:13,960 --> 00:57:18,720
if it's just as capable, relatively speaking, as 2.5 Flash was relative

876
00:57:18,720 --> 00:57:24,000
— yeah, if the ratio in terms of intelligence is the same as it was between Flash and

877
00:57:24,000 --> 00:57:28,599
Pro in that earlier Gemini model family —

878
00:57:28,599 --> 00:57:33,839
and if the price ratio is the same, i.e. 10 times cheaper, then you'd be able to,

879
00:57:33,839 --> 00:57:38,440
it would become possible for me, with my shoestring budget, to

880
00:57:38,440 --> 00:57:43,240
orchestrate thousands of agents to do that kind of thing, and, and that was actually

881
00:57:43,240 --> 00:57:47,799
the interesting thing about Kimi K2, that — just, that word you used, 'rush,' I

882
00:57:47,799 --> 00:57:50,559
think is pretty descriptive, because, in a way, you just send off [laughter]

883
00:57:50,559 --> 00:57:53,599
these agents to work, like, 'here's a codebase, go do this there, do, do

884
00:57:53,599 --> 00:57:58,359
that there and see what hap— see what happens.' So that kind of

885
00:57:58,359 --> 00:58:01,440
workflow is really fascinating — things are going to happen.

886
00:58:01,440 --> 00:58:06,319
Yeah, this really is, I mean, just an unbelievable time, and, I mean, maybe

887
00:58:06,319 --> 00:58:11,200
to wrap this up, I'll say that my own feeling is that this

888
00:58:11,200 --> 00:58:16,839
2026, in the history of humanity, whether what we're doing now

889
00:58:16,839 --> 00:58:20,039
means we merge into the machines as data, or whether we're at least for a while

890
00:58:20,039 --> 00:58:25,200
still writing things into this kind of intent window — that remains

891
00:58:25,200 --> 00:58:29,559
to be seen. But it really is going to be a turning-point year, and it's just great,

892
00:58:29,559 --> 00:58:33,400
great, just absolutely unbelievably amazing to be alive exactly this year.

893
00:58:33,400 --> 00:58:38,119
It is. It is [laughter]. Of course, I also have to say it's really quite frightening too. I mean,

894
00:58:38,119 --> 00:58:41,520
like, if you think about everything that could happen, but, but, but at the same

895
00:58:41,520 --> 00:58:44,720
time it's insanely great, when you think about all the good things that could

896
00:58:44,720 --> 00:58:47,480
happen, and what, what, how, how amazing, so,

897
00:58:47,480 --> 00:58:51,559
right, right, right, right, it really is great. Great — and if this year is going to be

898
00:58:51,559 --> 00:58:54,559
a turning point, then what on earth is going to happen in '27, well,

899
00:58:54,559 --> 00:59:00,160
that's out there — that's sci-fi future territory. That's how it is. Yeah. Maybe to

900
00:59:00,160 --> 00:59:05,039
close, we could still give a couple of tips, in case this all sounded

901
00:59:05,039 --> 00:59:08,119
like complete mumbo-jumbo, or whatever you'd call it,

902
00:59:08,119 --> 00:59:12,480
so, what small steps someone could take. Maybe my own path

903
00:59:12,480 --> 00:59:18,599
there is — well, 30 years ago I coded on a Commodore 64, in plain

904
00:59:18,599 --> 00:59:23,480
assembly, and, well, then I dabbled — I was, in between, in investment

905
00:59:23,480 --> 00:59:28,559
banking for 30 years, and still am, of course, but then this summer I looked into

906
00:59:28,559 --> 00:59:32,319
whether it'd be worth cramming that Codex thing into VS Code and

907
00:59:32,319 --> 00:59:36,880
doing a bit of that, and it was a bit rough, having to fuss around with settings and that kind of

908
00:59:36,880 --> 00:59:41,039
ancient stuff, and look at that really strange vi-style, slash-command editor,

909
00:59:41,039 --> 00:59:44,799
like, hasn't any of this changed in 30 years — and apparently it hasn't,

910
00:59:44,799 --> 00:59:48,839
so, right [laughter], that's how it is. But then I set up this kind of

911
00:59:48,839 --> 00:59:52,319
Perplexity Comet agentic browser myself. I think that's a

912
00:59:52,319 --> 00:59:56,599
pretty good stepping stone, because these things exist — there's Atlas from

913
00:59:56,599 --> 01:00:01,000
OpenAI — I wouldn't necessarily recommend that one widely — but now, well, Claude

914
01:00:01,000 --> 01:00:03,839
Code is also getting, in a way, that Chrome extension.

915
01:00:03,839 --> 01:00:05,799
Yeah, in a certain way you get into it, since

916
01:00:05,799 --> 01:00:09,520
you're using a regular web browser anyway. Hopefully, well, so that

917
01:00:09,520 --> 01:00:12,720
in a way, once you can command that web browser

918
01:00:12,720 --> 01:00:16,359
we could actually pull in a discussion from the inner circle — is UX

919
01:00:16,359 --> 01:00:22,119
dead? I.e., do we just do everything through API and MCP calls, with Samantha [laughter]

920
01:00:22,119 --> 01:00:26,119
I have a very strong opinion on that. Well, yeah, yeah, yeah, but let's leave the inner-circle

921
01:00:26,119 --> 01:00:31,000
but anyway, and then there's everyone's, well,

922
01:00:31,000 --> 01:00:35,319
beloved hero Jeff Bezos's money, and you get to help enrich the purple empire

923
01:00:35,319 --> 01:00:40,079
too, with Amazon's founder Jeff as a bonus. So, right, if you now take a

924
01:00:40,079 --> 01:00:44,599
step like this, I can personally recommend it. The Perplexity camp is

925
01:00:44,599 --> 01:00:48,160
pretty good for indexing too. It's a bit of a, you know, 'bubbling under'

926
01:00:48,160 --> 01:00:52,799
kind of thing, but in the web-browser space, I think it's a good option

927
01:00:52,799 --> 01:00:55,520
like this. Do you have something similar to recommend?

928
01:00:55,520 --> 01:01:00,319
I'd almost say you could go with the Swedish Lovable thing too — as another

929
01:01:00,319 --> 01:01:03,440
starting point, I'd have leaned that way. [laughter] Yeah. Because, well, I—

930
01:01:03,440 --> 01:01:07,160
since I, well, since my day job is at

931
01:01:07,160 --> 01:01:10,520
a security company, I'm extremely aware of all this prompt

932
01:01:10,520 --> 01:01:14,480
injection stuff, and that's exactly why I personally don't have any agentic browsers

933
01:01:14,480 --> 01:01:16,839
at all. And, yeah, this — remember,

934
01:01:16,839 --> 01:01:19,920
remember to always be careful with all of this, but, but, I mean, they are good, and

935
01:01:19,920 --> 01:01:23,400
that really is going to be the future, and, and it's really sweet to

936
01:01:23,400 --> 01:01:26,559
hear about those real, good use cases.

937
01:01:26,559 --> 01:01:31,920
But, well, I'd say, yeah, Lovable — if you want to try it for free

938
01:01:31,920 --> 01:01:36,079
and go straight into the Google camp, then Firebase Studio is basically the same thing, but

939
01:01:36,079 --> 01:01:40,240
just Google's version of it, in a way,

940
01:01:40,240 --> 01:01:42,720
and AI Studio is also pretty fun too, in that way

941
01:01:42,720 --> 01:01:46,799
yeah, yeah, Google's AI Studio — and, well, back in the day I also had

942
01:01:46,799 --> 01:01:49,160
some funding, so to speak, over there at the business school, so

943
01:01:49,160 --> 01:01:52,760
we're both KTM graduates [Finnish Master's in Business], yes, yes, yes [laughter]

944
01:01:52,760 --> 01:01:57,359
which maybe wasn't entirely obvious from this conversation, but, well, I remember

945
01:01:57,359 --> 01:01:59,960
back when we were founding startups, it was like, someone

946
01:01:59,960 --> 01:02:02,520
just had to learn to code, and had to learn to code, and then once you'd

947
01:02:02,520 --> 01:02:06,200
learned to code it, it was like, 'hey, wow, I can actually build things, I have

948
01:02:06,200 --> 01:02:09,720
an idea, I can make it' — and now, in a way, you don't even need to

949
01:02:09,720 --> 01:02:12,680
go through that kind of training anymore, you've just got Lovable, which can bring

950
01:02:12,680 --> 01:02:15,599
your idea to life. So I believe that if you haven't tried this kind of agentic

951
01:02:15,599 --> 01:02:19,079
coding or building, then what you get from Lovable is that, oh

952
01:02:19,079 --> 01:02:21,640
wow, I had this idea for something really

953
01:02:21,640 --> 01:02:24,359
simple, an app or whatever — it probably doesn't even feel

954
01:02:24,359 --> 01:02:28,079
simple to someone doing it for the first time, but that moment when you

955
01:02:28,079 --> 01:02:31,200
realize, 'hey, I can just build things' — that's exactly why

956
01:02:31,200 --> 01:02:35,559
I'd bring up Lovable, and things like Firebase Studio too, so that

957
01:02:35,559 --> 01:02:39,119
only after a few months would I jump into Claude Code and other

958
01:02:39,119 --> 01:02:41,400
— first with Lovable, like,

959
01:02:41,400 --> 01:02:45,640
Yeah, and that Lovable thing — I mean, if you already have some kind of vision, like I do —

960
01:02:45,640 --> 01:02:51,160
I've got this Consigliere thing, where I mirrored it against my MBTI test, and, well,

961
01:02:51,160 --> 01:02:54,920
well, right [laughter], right, right, well, it's got all sorts of things in it, it's got a receipt scanner

962
01:02:54,920 --> 01:02:59,359
and it's got a really good, this second brain thing — but, well, maybe

963
01:02:59,359 --> 01:03:04,520
these MD files aren't really that well indexed in the sense that

964
01:03:04,520 --> 01:03:07,720
these ideas don't yet talk to each other — they're still pretty static,

965
01:03:07,720 --> 01:03:10,520
but you find all sorts of things in there, and whenever I come up with an idea I run

966
01:03:10,520 --> 01:03:14,480
it, as a rule, through this private app of mine, Consigliere — and that too is

967
01:03:14,480 --> 01:03:20,319
maybe a good tip — just build it for yourself first, try scaling it just for yourself, yeah,

968
01:03:20,319 --> 01:03:26,039
like that. Well, this was a fun conversation, Markus Hav — I could really

969
01:03:26,039 --> 01:03:28,680
do a six-hour episode like this, it could actually be pretty

970
01:03:28,680 --> 01:03:32,079
legendary — just let it rip, and go through absolutely every single

971
01:03:32,079 --> 01:03:34,520
crazy idea, just get it on tape. Yeah, I'm totally, totally [laughter] I'm

972
01:03:34,520 --> 01:03:38,279
all in on that — or, well, it just felt like we were only just getting started, that we

973
01:03:38,279 --> 01:03:41,720
barely even talked about Moltbook or any of that AI stuff. So, yeah,

974
01:03:41,720 --> 01:03:45,599
true, yeah, but we did agree that Samantha would be invited to Hoxhunt's Moltbook —

975
01:03:45,599 --> 01:03:50,119
sorry, sorry, Open Claw — training, because it's changed its name three times now.

976
01:03:50,119 --> 01:03:53,279
Right, that's it, yeah, yes, yes [laughter], exactly right, but yes,

977
01:03:53,279 --> 01:03:59,799
yeah, that's how it is. But, well, yeah, and maybe it's good, then, if you set off on this

978
01:03:59,799 --> 01:04:05,520
AI journey, that you have this kind of conversation with someone — some person

979
01:04:05,520 --> 01:04:09,720
you know who's good at this stuff, and just talk about it, and then maybe

980
01:04:09,720 --> 01:04:12,520
you could even bring your laptops along and then a bit

981
01:04:12,520 --> 01:04:17,279
of prompting, or 'intent-ing,' or coding, or whatever you want to

982
01:04:17,279 --> 01:04:20,599
call it — Neuralink it, Neuralink it,

983
01:04:20,599 --> 01:04:24,559
Whisper-Flow it, whatever. Thank you. That was fun. And I

984
01:04:24,559 --> 01:04:28,880
suggest we now head over to the inner-circle side. By the way, I'd really appreciate it if

985
01:04:28,880 --> 01:04:32,640
you subscribed. It's genuinely important to me — I'm staying ahead of Inderes TV,

986
01:04:32,640 --> 01:04:38,720
well, they're still breathing down my neck, 1,000 subscribers behind me, and,

987
01:04:38,720 --> 01:04:42,039
we'll talk in the inner circle then about whether, whether the human

988
01:04:42,039 --> 01:04:44,880
interface is dead. Thank you, Markus.

989
01:04:44,880 --> 01:04:47,480
Thanks.