---
title: "AI Makes the Mediocre Free | Kari Angeria, Janne Ikola | Negotiator 414"
summary: "Kari Angeria and Janne Ikola join Sami Miettinen to discuss their new book Vedenjakaja (The Watershed: AI and the Redistribution of Work). The book's basic claim is that occupations are made of tasks that change at different speeds; the episode's sharpest is that AI makes the mediocre free: once you must write down for an agent what a role is meant to achieve, it turns out nobody ever defined it for the person either. The episode weighs whether this time is different, why organisations change as a slow avalanche, whether people want orders from machines, whether companies will be needed, and which three new knowledge-work roles are emerging. The article sets out the book's content and separates the participants' views from checked facts."
datePublished: 2026-10-13
dateModified: 2026-10-13
originalLang: en
section: society
sections: ["society","economy","tools"]
authors: ["Sami Miettinen"]
tags: ["Neuvottelija","EP414","Tekoäly","Työn murros","Tietotyö","Agentit","Organisaatiot","Vedenjakaja","Kari Angeria","Janne Ikola"]
canonical: https://www.neuvottelija.com/ai/ep414-tekoaly-tekee-keskinkertaisesta-ilmaista-angeria-ikola/
---
# AI Makes the Mediocre Free | Kari Angeria, Janne Ikola | Negotiator 414

# AI Makes the Mediocre Free | Kari Angeria, Janne Ikola | Negotiator 414

> **Summary:**
> Kari Angeria and Janne Ikola join Sami Miettinen to discuss their new book Vedenjakaja (The Watershed: AI and the Redistribution of Work). The book's basic claim is that occupations are made of tasks that change at different speeds; the episode's sharpest is that AI makes the mediocre free: once you must write down for an agent what a role is meant to achieve, it turns out nobody ever defined it for the person either. The episode weighs whether this time is different, why organisations change as a slow avalanche, whether people want orders from machines, whether companies will be needed, and which three new knowledge-work roles are emerging. The article sets out the book's content and separates the participants' views from checked facts.

## Who the participants are

**[Kari Angeria](https://www.neuvottelija.com/podcast/guests/kari-angeria/)** spent 14 years at a
regional media house in Upper Savonia, Finland, working on media technology and as editor-in-chief.
For the past five years he has been an AI entrepreneur; in his own words his company is "an AI
partner for businesses". He appeared on the show earlier with futurist Risto Linturi
([EP394](https://www.neuvottelija.com/ai/ep394-euroopasta-tulee-paasiaissaari-linturi-angeria/)),
and he runs Neuvottelija's AI working group (Tekoälytyö) together with host Sami Miettinen.

**[Janne Ikola](https://www.neuvottelija.com/podcast/guests/janne-ikola/)** runs AI-Sanomat, a Finnish AI media channel, and a venture studio that builds and
operates AI businesses. He says he started building on language models between GPT-2 and GPT-3, with
a "Marketing Agency in your pocket" product, and has worked in marketing, PR, sales and strategy
since 1998. He is trained as both an engineer and a business graduate.

Together they wrote **Vedenjakaja – Tekoäly ja työn uusjako** (*The Watershed: AI and the
Redistribution of Work*), and the episode is in effect the book's launch conversation. Both guests
work with AI and help organisations adopt it, so they also speak from their own professional
experience. The host, Sami Miettinen, talks openly about his own "androidisation" and his agent
Samantha, and he shares his own views on society during the episode, which are marked below as his
views.

## The book in brief: what *Vedenjakaja* argues

*Vedenjakaja – Tekoäly ja työn uusjako* ("The Watershed: AI and the Redistribution of Work",
Basam Books 2026; 168 pages plus sources, in Finnish) was read in full for this article.
The book asks three questions: what in work survives, what changes, and how to prepare. It sets
firm limits on itself. It is not a technology book, and it is neither a "fear book nor a rapture
book". It also does not try to settle questions about taxation, welfare-state funding or democracy.
The authors say they wrote in one voice. Each author's own stories are set apart and signed, and
they sit among more than ten first-hand accounts from Finnish employees and entrepreneurs
interviewed for the book. One sentence runs through the whole book: *"things will be better in the
future than in the past"*. The epilogue finally admits that this is not a forecast but a choice.

The publisher's [press release](https://www.sttinfo.fi/tiedote/72353939/miten-tyolle-kay-tekoalyn-aikakaudella-uutuuskirja-nostaa-esiin-kolme-muutosta-jotka-koskevat-jokaista-tyopaikkaa?lang=fi)
(29 September 2026) condenses the book into three shifts: work is redistributed task by task, value
moves from production to judgement, and the benefit requires work to be redesigned. That summary is
accurate, but the book itself covers more ground. It is built in four parts.

**Part I: history, three machines and three laws.** The book opens with the Captain Swing riots.
In August 1830 day labourers in East Kent smashed a threshing machine, and the riots that followed
ended in 252 death sentences. The book asks two questions: who pays the price of the transition, and
why the rioters' grandsons won in the end anyway. Three transitions follow:

- **The field.** The machine took over physical labour, and Finland's great migration of the 1960s
  and 70s was, in the book's words, "Finland's own watershed".
- **The camera.** Delaroche's "painting is dead", and the three paths open to miniature
  painters: resistance, adaptation and reinvention.
- **The office.** Solow's productivity paradox; the electric motor, whose benefit showed only after
  about 40 years; the ATM, which at first increased the number of bank tellers; and Estonia, which
  lightened bureaucracy by rebuilding its administration.

From these the book derives **three laws of disruption**:

1. *Work does not disappear; it is reorganised* (Acemoglu and Restrepo's displacement and
   reinstatement effects).
2. *Value moves from the visible to the invisible*: from performance to judgement, interpretation
   and responsibility.
3. *Poor organisation breeds bureaucracy*: a machine glued onto an old process adds work.

As the shadow of these laws, the book keeps the **Engels' pause** in view. In Britain between 1780
and 1840, output per head rose by almost half while wages rose by only a little over a tenth. A law
can hold on average while a generation still pays the price. On "is this time different" the book
takes a clear position: AI is the laws' *strongest example, not their exception*. The host disagrees
in the episode (see section 1).

**Part II: how work comes apart.** Five chapters describe the present:

- **Occupations break into tasks (chapter 5).** Frey and Osborne's famous 47 per cent automation
  risk shrinks to about nine per cent when risk is counted task by task instead of occupation by
  occupation.
- **The mediocre becomes cheap (chapter 6).** The machine raises the baseline for everyone, and
  beginners and the weakest performers gain most. The book adds two warnings.
  - *The jagged frontier*: the machine is dazzling at one thing and helpless at the next, and it
    does not tell you which is which.
  - *The illusion of trust*: experienced developers believed they had sped up but were measurably
    slower.
- **The junior problem (chapter 7).** The bottom rung of the career ladder is crumbling, but a
  beginner who gets in grows faster with the machine than ever before. The book's answer is the
  medical model, in which learning happens in real, supervised work.
- **Value moves from making to choosing (chapter 8).** Every knowledge worker becomes a bit of a
  curator. The book also notes that taste and judgement are not evenly distributed.
- **The organisation is the bottleneck (chapter 9).** IBM Watson at MD Anderson cost about $62
  million without treating a single patient. According to McKinsey, only about a fifth of companies
  using AI have redesigned their workflows.

**Part III: four futures that happen at the same time.**

- **The slow avalanche.** AI shows up "everywhere but in the statistics", the productivity J-curve
  is only beginning, and the slowness is both a gift and a trap.
- **Selective collapse.** Translators, illustrators and copywriters are the "canaries", and it is
  now the middle of knowledge work that is being hollowed out. The book uses the China shock as its
  warning.
- **Turbocharged expertise.** At the individual level AI often narrows gaps, but at the level of
  firms, countries and capital it widens them. This chapter also warns about a retracted MIT paper.
- **The AI version of bureaucracy.** Workslop, the irony of automation and blurred responsibility.
  The counter-example is Kela's aim to remove the customer's need to deal with the agency at all.

Chapter 14 pulls these together into the book's title image: an **uneven watershed** that runs
between countries, regions and company sizes, and even within a single profession. In the European
Investment Bank's survey Finland leads the EU in generative-AI use, yet inside the country it splits
by industry and company size.

**Part IV: what to do.** The book climbs four steps:

- **The individual** builds five skills: problem framing, evaluation, a feel for context, ethical
  responsibility and meta-thinking. They can also rely on three lasting traits: laziness, curiosity
  and the need to belong.
- **The team** chooses to work as a centaur (a clear division of work with the machine) or a
  cyborg (close back-and-forth), and avoids the self-automator, who hands work to the machine and
  learns nothing.
- **The leader** stops "paving the cow paths" and answers three questions: what do we give up, what
  do we build in its place, and which unrealised idea do we finally take on? The last is the book's
  **latent business potential**.
- **Society** does what Finland's comprehensive school did in 1972: build an institution that carries
  people across the watershed. The book also reports that researchers in Finland did not find a dip
  in youth employment in AI-exposed occupations. It credits the Nordic labour-market model and
  employment protection.

The epilogue asks what people are paid for, and whether that is the same as what makes a person
valuable. Its answer is that it is not. In the end, responsibility is the one thing that cannot be
handed to a machine.

### What the book does well, and where it is careful

The book is more careful with its evidence than most AI writing. For each analogy it spells out a
"where the comparison may fail" section. It also brings up the evidence that cuts the other way:

- The retracted MIT materials-science paper.
- A note that the Anthropic CEO's forecasts "should be taken with caution because the person making
  them sells AI".
- Acemoglu's modest productivity estimate, set beside the optimists.

The book's sources are listed chapter by chapter, and the book says its web addresses were checked
in July 2026. I checked a few key figures:

- The 16 per cent relative decline in employment among 22–25-year-olds in Stanford's *Canaries in
  the Coal Mine* matches the study's
  [November 2025 version](https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine).
  The August version gave 13 per cent.
- The mileages at Parting of the Waters match the sign at the site.

The book's success stories include Angeria's own client case: the Sotkamo water company Vaarojen
Vesi and the strategy agent he built for it. The interviewees' accounts lean towards positive
experiences. The book is hopeful by design, and it says so itself.

### What the episode takes from the book, and what it does not

In the episode the authors refer to the book several times. The central reference is the
episode's title idea: **"One chapter in our book is about AI making the mediocre free."** In the
book the chapter is titled *The mediocre becomes cheap* (chapter 6), and there mediocre means
*baseline performance*: the machine gives almost anyone access to passable text, translation and
code. The episode applies the idea to *organisations* and their undefined work (see section 3). The
two perspectives complement each other.

The episode's other references can be found in the book: the counter-forces to change (chapters 3,
9 and 13), the slow avalanche (chapter 10), the junior problem (chapter 7) and diverging realities
(chapter 14). Several of the episode's topics are not in the book: tokenmaxing, agents as legal
persons in Argentina, and whether companies are needed at all. There the conversation takes
the book's ideas further.

### Where the name comes from

Sami asks directly in the episode what a watershed means. The authors answer that the image comes
from Wyoming. Chapter 14 of the book describes the place more precisely. In the Teton Wilderness,
on the Continental Divide, is
[Parting of the Waters](https://en.wikipedia.org/wiki/Parting_of_the_Waters), where North Two Ocean
Creek splits in two. Pacific Creek flows about 1,353 miles (some 2,200 km) to the Pacific, and
Atlantic Creek about 3,488 miles (some 5,600 km) to the Atlantic. The book also draws a hopeful
side from the image. The divide is so shallow that during the spring flood a fish can swim across
it, and Yellowstone Lake's cutthroat trout are thought to have reached the lake this way. The book's
conclusion is that a watershed is not a wall but a ridge, and it can be crossed at the right place
with the right help.

### A note on attribution

The captions carry no speaker labels. The guests are named below only where the speaker is clear,
for example when one refers to the other by name or Sami addresses a guest. Elsewhere the text
refers to "the guests" or "the authors".

## 1. Is this time different?

The episode opens with the oldest question every technology wave raises. Sami recalls that "this
time is different" is the most famous wrong sentence in economic history, and says he nevertheless
feels strongly that this time it is true. His reason is generality: earlier machines replaced the
physical dimensions of human work, whereas now both intelligence and physical capability are being
replaced.

The guests share the view that the change is powerful, but they put other things alongside it.
Ikola brings in **friction**: those who adopt the technology are pulling ever further ahead
of "the large mass trailing behind", and the gap generates resistance, regulation and conflict. In
his description a company is a legal and organisational contraption built to stay standing even as
people, names and customers change. When a technology that changes everything is brought into it,
the structure pushes back.

**A real difference remains in this exchange.** Sami stresses that the change is unprecedented; the
authors stress that it is slow and uneven. Neither position is a checkable fact, and the episode
leaves the question open.

The book takes a position in chapter 4, under the heading "Is AI the exception?". It weighs the
arguments both ways. In favour of the exception: AI improves itself, scales to billions through a
single model, and reaches layers of work once thought safe from machines. Against it: the same
productivity paradox seen with electricity and computers, Acemoglu's estimate of a modest
productivity effect, and the slow resistance of organisations. The book concludes that AI is the
laws of disruption's *strongest example, not their exception*. It does not dismiss Sami's worry,
though. Already in chapter 1 it sets out three "honest objections" to the farm analogy:

- **Timing.** The Engels' pause lasted a generation and a half.
- **Structure.** Carl Benedikt Frey asks where new work will appear if the machine takes both the
  lowest and the highest layer.
- **Threshold.** In Korinek and Suh's model, labour and capital become substitutes once automation
  passes a certain level.

So the book considers Sami's scenario possible, though not the most likely one.

The conversation also draws a parallel with the Luddites and the mechanisation of textile work. In
Sami's summary, those who resisted the machines have never won before, and the guests say the book
treats the same phenomenon through historical examples.

## 2. Tasks are replaced, work is not, at least not directly

Angeria says what muddles the debate is that **work is talked about as if it were the same thing as
tasks.** He estimates that "probably 90 per cent" of the individual tasks in knowledge work will be
replaced, but work contains more than that: an understanding of the whole and of the physical world,
which machines in his view do not yet have. The figure is Angeria's own
estimate.

The shape of the change is, in his words, the **slow avalanche** the authors write about in the
book: in one organisation knowledge-work processes can be rebuilt completely while in some municipal
office people are still "shuffling papers". At first change is slow and expensive; then it goes
vertical, old companies fall over and the pace accelerates. Angeria also warns against
underestimating people's capacity to resist change. He refers to stories in *Helsingin Sanomat*, the
first predicting that public-sector work would soon be replaced by AI and a later one reporting that
the change is not progressing. His explanation is human: if you are told your job will be replaced,
why would you push for it?

In the book, the slow avalanche has a chapter of its own (chapter 10). There it is described as the
productivity **J-curve**: redesigning work costs money now and pays off only
years later. The electric motor's benefit showed up in factory figures with a lag of about 40 years.
The book says two things about slowness. It is a *gift*, because it buys time to adapt. It is also a
*trap*, because by the time the change shows up in the statistics it has already happened. The book
also warns against averages: for the translator whose income halved, the avalanche "came like a
cliff".

### Programmers show the way

Sami offers software development as the model case: git, code review and the processes of team
development existed to divide work among people, and now one person can take a change to production
directly with agents. One of the guests says that eighteen months to two years ago he believed
programming as a profession was finished. In leading companies the opposite happened: demand for
programmers who work with AI has grown, and the established ones have not yet lost their jobs. The
work shifted and changed, and new work appeared.

Sami sums it up: programmers used to code code; now they "code the company". The trend runs bottom
up: the more operational, keyboard-level your work, the more your role rises to that of an
**orchestrator**. The question for companies is who can make the move to understanding the wider
process and context of the work. Those who can are in high demand right now. This is the core of the
book's second claim: value moves from production to judgement.

## 3. AI makes the mediocre free

This is the episode's sharpest claim, and it is worth reading carefully, because it is aimed not at
individual employees but at organisations. It builds on the book's chapter 6, *The mediocre becomes
cheap*, which the episode applies to organisations (see below).

According to the authors, a large share of the work done in companies is mediocre because **there
are no ways to demand more and no ways to measure what is being done.** Especially in medium-sized
and large companies, size breeds layers of middle management whose main function is to organise
meetings for others, route reporting towards themselves and coordinate internally by email. The
value this work adds for the company is low, which is why AI takes it first, particularly because
customers do not pay for the extra cost.

### An agent's goal reveals that roles were never defined

The episode's most interesting observation is a practical one. Now that agents can work
independently towards a given goal, people talk about "loop engineering", "goal engineering" and
"reward hacking". The example is concrete: give an agent the goal of optimising customer acquisition
cost, and it may decide to offer each new customer a thousand-euro sign-up bonus. The goal is met;
the company loses. Put an agent in charge of social media marketing, and it soon starts doing the
wrong things, too much, for the wrong audience.

The authors' point follows: you have to write down for the agent what the role is really meant to
achieve, and **that is exactly what companies typically have not thought through.** The job
description exists, but what a person actually does may be something quite different. Sami puts it
bluntly: the "specification" of human work is usually very poor, and when a machine does the same
work at a higher clock speed, you are forced for the first time to ask why you pay someone €70,000 a
year to do two right things and three wrong things a day.

The guests point to the other side: even if new work appears, it does not appear for those who
cannot move from one level of abstraction to another. It is easy to say that judgement and seeing
the whole are needed, but these are genuinely difficult things.

### What the book says about the mediocre

In the book's chapter 6, mediocre means **baseline performance**. In 2023 a machine passed both the US bar exam and the US medical licensing exam, the
very tests that separate a professional from a layperson. The ability to produce passable text,
translation or code therefore no longer separates the expert from the beginner as much as it did.
The book adds at once that the machine no longer produces only the mediocre; in many areas it
produces the excellent, but unevenly. That is exactly why human judgement becomes more valuable.

The episode's version applies the same idea at the level of the organisation. Once baseline
performance gets cheap, it becomes clear how much of the work was baseline performance with no
defined goal. The book's chapter 13 names the flip side. In workslop, the machine produces content
that looks like work, and the cost of checking it falls on the person who receives it.

### The public sector

Sami singles out the public sector: in his view it never had performance metrics, only inputs,
tasks and official liability, and the risk function of official liability rewards inaction. He
refers to roughly €15 billion of annual state borrowing and calls it irresponsible to postpone
prioritisation decisions. He asks whether a top-down cutting programme of the kind seen in the
United States or Argentina is needed, or whether change can come from the grassroots. The borrowing
figure is Sami's round estimate, and the comparison of cutting models is his view.

### Sami's view of capital and the left

During the episode Sami states several views of his own: that one economist's understanding of
capital as a factor of production is "dangerously wrong"; that real capital is the money of owners
and creditors and active ownership; and that people on the left who treat work as an end in itself
cannot design at the level of abstraction an AI world requires. Later he says "the leftists are the
Luddites". Sami presents these provocations as his own opinions. The guests approach the subject from
another angle and ask what society actually wants to optimise: the maximum amount of work, or
something else?

## 4. Tokenmaxing, individual productivity and personal licences

Sami urges every sceptic to actually try AI: it "makes the mediocre free", and if you yourself are
mediocre the experience is eye-opening. Costs are discussed openly. One guest calls roughly €2,000 a
month on tools a cheap price and says that even hesitating over a €20 monthly subscription shows the
value has not been understood. Sami says he spends about €700 a month. Later the other guest
calculates that he spends about €500 a month and estimates the tools do around €7,000 worth of work
for him. The subtitles do not make clear which guest gave which figure. The numbers are the
participants' own estimates.

**Tokenmaxing**, maximising token consumption, is becoming a way for companies to demonstrate AI
use. The guests say it reflects a world in which AI is seen as an IT tool that makes an individual
employee more efficient and simply needs to be used more, rather than a reason to rethink the
company's processes from scratch. This is the book's third claim: the benefit requires redesigning
work.

Sami proposes a path from Stephen Covey's *The 7 Habits of Highly Effective People* (1989): the first
three habits concern personal effectiveness, the next three cooperation. In his view AI works the
same way: you need to master individual effectiveness before you can say anything relevant about
team effectiveness. Otherwise you only see "magic boxes".

The guests add a complication: an individual can experiment with a shotgun, trying every tool and
switching freely, but **a team or an organisation cannot.** It has to choose a direction and a
budget, which slows the change. Some people also play zero-sum games and hoard the best licences for
themselves, just as they hoard the best teams and resources. Sami's answer is to pay for it
yourself. The guests note that this builds capability in the individual rather than the company, so
it is not necessarily in the company's interest.

## 5. AI-native uses have not been found yet

One guest uses an analogy he says he has repeated in training sessions: when the first television
broadcasts arrived, they showed stage plays and newspaper ads. Nobody yet grasped that an entirely
new mass medium was being born. He says he made the same mistake himself with mobile in 2007: he saw
the phone as a small browser and did not realise that GPS would make services like Uber possible.
His estimate is that with AI we are at the internet's 1996–97: everyone can see a new world coming,
but the equivalent of e-commerce is not yet understood.

Sami sees agentic delegation as the most obvious first native use, and cites Cloudflare's chief
executive, who said agent traffic overtook human traffic on the web well ahead of forecast. **This
checks out:** Cloudflare's Matthew Prince said in early summer 2026 that bot traffic had
[overtaken human traffic](https://ppc.land/cloudflare-ceo-expects-bot-traffic-to-hit-1-000-times-human-within-5-years/)
on Cloudflare's network in May 2026, although in November 2025 the company had predicted the crossover
for the second half of 2027. The
figure counts HTTP requests on Cloudflare's network, which covers roughly a fifth of websites, not
what people read.

## 6. Do people want orders from a machine?

The episode has also carried the YouTube title "People want orders from the machine", and this is
the most surprising passage in the conversation. Angeria says he had assumed people would resent
taking orders from a machine. His wife, a psychologist, told him over breakfast that the opposite
is true: the machine is patient and plays no favourites, so "of course people want orders from a
machine". In Angeria's view the next step is AI managers that hand out tasks, and that too will
arrive as a slow avalanche, like robotisation, which is already on production lines but not yet in
Finland's physical everyday life.

Sami offers the counter-image: in an Amazon warehouse the machine orders you to walk to shelf 37 and
scolds you for toilet breaks, until the human is replaced by a humanoid robot. Nobody wants a boss
like that. One of the guests is not convinced: if the machine makes better decisions and leads
better, why should it not lead, when Finns have complained about poor management for decades? **This
disagreement remains open**, and it is one of the most interesting in the episode, because it does
not map onto the usual optimist–pessimist divide.

The book takes this up in chapter 16 and lands between the two views. It says many people would
welcome a machine boss because "the machine has no favourites", and that in studies, employees who
expected unfair treatment from a human boss trusted the machine's assessment more. Yet for Uber
drivers and in Amazon's warehouses the same technology is a "tireless whip". The book says the
difference does not come from the technology but from **autonomy, transparency and power**. The line
often falls here: a machine is accepted as a source of feedback and as the one handing out tasks,
but decisions about pay or dismissal are wanted from a human who can be asked why.

### The human job is to bring entropy

The guests offer people a task that they admit sounds like jargon: **bringing entropy into the
system.** Agents working towards a goal tend to end up going in circles, however intelligent the
system, and chaos has to be introduced into a closed system from outside. Sami describes an
observation from the Tekoälytyö group: when six agents tackle the same problem with the same best
practices, they end up in a conformist "yes-circle". Sami believes this too can be solved by the
machine, for example by instructing an agent to increase its variance, and asks whether bringing
entropy is a permanent human task at all. The question stays open.

## 7. Intuition, tacit knowledge and the limits of language models

Ikola has an intuition he calls **shadow data**: far more knowledge exists in people's experience and
subconscious than on the internet, and it reaches us as intuition. A language model can solve new
mathematical problems, but raw computational intelligence is not necessarily what moves thinking
forward.

Angeria builds on a text he read called *Natural General Intelligence*: mathematical problems are,
in the end, simple within their own logical frame, whereas the complexity of ecosystems, weather and
the whole biological reality is on another level entirely. He suspects that the importance of
brute-force intelligence for the physical world is being overestimated, at least in the short and
medium term. (A book of that name by Christopher Summerfield was published in 2022; the episode does
not make clear which text Angeria means.)

The conversation answers technically: the problem lies in the current transformer architecture,
introduced by Google researchers in 2017, which emits tokens. Attempts at a remedy include the
**world models** championed by Yann LeCun, which try to model how the world works, and
**neurosymbolic AI**, which tries to link the meanings of concepts rather than just the
co-occurrence of words. The prediction is that in five years today's language models will look like
floppy disks in the SSD era.

Sami brings in Daniel Kahneman's two systems (*Thinking, Fast and Slow*, 2011): fast intuition and
slow reasoning. In his account, fast, intuition-like classification is starting to be built with
JEPA-type architectures, and once that is combined with reasoning models the machine will go "way
past humans". His thought experiment: if he could ponder every answer for a year and still reply in
a second, he would seem godlike, and that is where machines are heading.

### Facts, opinions and value choices

Ikola connects this to a distinction: on mathematical and physical matters you can say something is
A or B, but in taxation and social choices there are more views than facts, and a "superbrain" can
be made to argue either side. So the final value choices remain with humans. Sami disagrees: in his
view people's value choices are the product of poor cognitive heuristics, and opinions can be pushed
towards facts "quite hard". Ikola replies that society's direction has emerged from the consensus of
millions of people, not from one person's correct choices. **This is the episode's most fundamental
disagreement, and it is not resolved.**

## 8. Will we need companies at all?

One of the authors throws out a radical question he says he first raised in a group chat: will
companies be needed at all in the future? The structure of the firm fits a particular kind of world,
and he himself works as a sole trader. Could structures emerge that are neither companies nor
individuals, structures we do not yet have a name for?

Sami connects the question to economics and Ronald Coase's theory of the firm (*The Nature of the
Firm*, 1937; called the "Coase theorem" in the episode): the firm exists because contracting in the
market for every task is costly, and inside the firm people can assign each other work without
contracts. Sami's conclusion is that once AI drives coordination costs down drastically, the
protective shell is no longer needed, and economic activity will be bought from combinations of
micro-firms, agents and AI. He recalls proposing to then prime minister Jyrki Katainen that everyone
be given a business ID alongside their personal identity code, so that entrepreneurs and employees
would be equal. He thinks the idea's time has come.

Ikola links this to **agentic commerce**: it would already be technically possible for someone who
has spent twenty years in middle management and expert work to codify their **tacit knowledge**
behind an AI agent and sell it in micropayments. What is missing is the marketplace and the
transactions.

### The audience decides

Sami says he has tried Discord, Facebook and Slack for communities, but Finns are on WhatsApp. The
guests explain it from media experience: old media houses still have their audiences even though the
print business should have died long ago, and the audience leaves only when the readers die. Bots
would work much better on Telegram, but Meta has the audiences. Sami compares it to VHS beating
Betamax: network effects can decide in favour of the technically weaker option.

### Diverging realities and agents as legal persons

The authors return to the book's title: people, companies and countries may end up in completely
different realities. In the book (chapter 14) this is a measured phenomenon. The book cites the
OECD: AI diffusion is accelerating more because the leaders are pulling away than because the
laggards are catching up. The episode's examples of agent companies and AI-free zones go further
than the book. The example is Argentina, where it is being considered whether an agent could
found a company and be a legal person. **This checks out:** President Javier Milei's government has
sent Congress a bill amending the companies law to create a new type of company operated by AI
agents or robots
([Buenos Aires Herald](https://buenosairesherald.com/business/tech/milei-harari-clash-over-non-human-companies-as-argentina-pushes-ai-legal-framework)).
According to [The Next Web](https://thenextweb.com/news/argentina-ai-non-human-corporation-still-needs-humans),
the draft still requires a human legal representative for certain acts. The bill's legislative status
had not been verified at the time of writing.

As a counter-example, Italy is mentioned for banning ChatGPT. Italy's data protection authority
blocked the service at the end of March 2023, and the ban was lifted at the end of April. The authors
foresee "AI-free zones" and "agent-company zones", and ask whether agents will become a little like
children, for whom parents are responsible only until they come of age.

Sami continues politically: on one side of the watershed there could be an androidised
ultra-capitalism in which humans disappear because they no longer add value; on the other, a "Star
Trek utopia" where machines do all the work for free. He gives the latter "maybe a 0.1 per cent
chance". The guests reply that when resources are concentrated in one place the outcome is not good,
because people are corruptible.

## 9. The young, the junior problem and three new roles

At the end Sami asks for concrete job advice. The authors start from the **junior problem**, which
the book also discusses: the jobs through which people used to enter working life are disappearing.
One of them has a practical observation from recruitment: many young applicants now bring their own
agents with them. Set 20–30-year career veterans against a young person who arrives with a
five-agent team, and the young person represents the new world. Sami jokes about "agent nepotism":
the parents will have to pay for the Claude Code subscriptions.

The same guest names **three new knowledge-work roles**, admitting he has probably borrowed the
framework from somewhere:

- The **operator** is the orchestrator who runs the agents, as Sami does with Samantha.
- The **curator** makes sure the agents work: skills, connectors, goals and data access. In some
  companies that use AI agents this is already a role of its own.
- The **designer** brings human creativity, insight, direction and strategy. This is the
  "entropy supplier".

Sami pushes back: if concrete tasks cannot be named, he will discount it all as wishful thinking. One
of the guests answers with his own sons. The older one is building a data-driven coaching agent with
the runner Santtu Heikkinen and has found the work much harder than mere coding: he has to learn
architecture, complex thinking and judgement, exactly the things the book talks about. The book has a first-hand account of the same experience
(chapter 7). Angeria describes building his own CRM and agents with AI, and finding that he spots
the machine's errors "significantly worse" than an experienced programmer would. Coding may be
the literacy of the future. He also says that as a father he regrets today's end-of-the-world talk:
young people are anxious, and poorly reasoned doom-mongering is irresponsible. Better to teach them
that the world changes and they will find their own path.

Sami predicts work in home automation and robotics: his Home Assistant system runs on a laptop under
his bed, and he expects home robotics and energy management to be commercialised at scale during
2027. The home is a sensitive place, so there will be plenty of work in safety and regulation. Ikola
closes on scarcity: as knowledge work and robots scale, **human contact is the resource that does
not scale**. When agents handle the rest, he says, he will go to lunch with Kari or come on a podcast
to have his ideas challenged, and that is why he expects events and businesses that bring people
together to rise.

The book offers two counterparts to the episode's roles. In chapter 8 every knowledge worker becomes
"a bit of a curator". Chapter 16 describes the **agent boss**, who builds, directs and supervises a
group of agents like a small team. On the junior problem, the book reports a Finnish finding that the
episode does not mention. In the US, youth employment has fallen in the occupations most exposed to
AI: by about 16 per cent for 22–25-year-olds relative to comparison groups, according to the
Stanford study. According to the book, researchers in Finland did not find a similar dip, and it
credits the labour-market model and employment protection. The adjustment happens through
retirement and internal retraining.

## Sources and views

**Checked against sources:**

- The book's content: *Vedenjakaja* has been read in full. Its chapters, structure and claims are
  described from the book itself, not just the press release.
- The book's title, publisher, release date and three central claims:
  [Basam Books press release, 29 September 2026](https://www.sttinfo.fi/tiedote/72353939/miten-tyolle-kay-tekoalyn-aikakaudella-uutuuskirja-nostaa-esiin-kolme-muutosta-jotka-koskevat-jokaista-tyopaikkaa?lang=fi).
- Bot traffic overtook human traffic on Cloudflare's network in May 2026, earlier than forecast.
- Argentina's bill for a company type operated by AI agents.
- The watershed image: Parting of the Waters, Wyoming.
- Italy's ChatGPT ban in spring 2023.

**Clarifications:**

- The book chapter mentioned in the episode is titled *The mediocre becomes cheap* (chapter 6) and
  deals with baseline performance; the episode applies the idea to organisations.
- What Sami calls the "Coase theorem" refers in substance to Coase's theory of the firm (1937).

**The participants' own estimates and views:**

- Roughly 90 per cent of tasks replaced (Angeria).
- Monthly tool costs and estimated returns.
- The slow-avalanche model.
- The three new roles.
- Sami's political views on capital, the left and public-sector cuts.
- The forecast of a home-robotics breakthrough in 2027.

## What to take away

- **Measure tasks, not occupations.** The book's basic claim is its most practical tool: break your
  own work down into tasks and see which change first.
- **An agent's goal is a mirror.** If you cannot write down for an agent what a role is meant to
  achieve, you probably never wrote it down for a person either. That is what the authors mean in the
  episode when they say AI makes the mediocre free. The book's own version is more measured:
  baseline performance gets cheaper and judgement gets more valuable.
- **Usage is not productivity.** Tokenmaxing measures input. The benefit only arrives when the flow
  of work and the right to decide are redesigned.
- **An individual can experiment; an organisation has to choose.** That is why organisations change
  more slowly than their front-runners, and why the gap keeps growing.
- **Read the book through its three laws.** When someone claims AI will make an occupation
  obsolete, the book suggests three questions: where is the work reorganised, where does human value
  move, and does the organisation dare to rebuild its work?
- **Left open:** whether people want machines to lead them, and whether value choices belong to
  humans. The episode resolves neither, nor should it.

## Sponsorship and affiliations

The episode includes a paid partnership with Posti, the Finnish postal and logistics company (a
Posticast advertisement at 2:20). The advertisement is unrelated to the episode's subject and has no
bearing on this article. Kari Angeria runs Neuvottelija's AI working group with Sami Miettinen, and
Sami promotes Ikola's AI-Sanomat during the episode. Both guests provide AI-related services for a living. The book's success stories include Angeria's own client case
(Vaarojen Vesi and the strategy agent he built).

## Episode details

- **Episode:** [Tekoäly tekee keskinkertaisesta ilmaista | Kari Angeria Janne Ikola | Neuvottelija 414](https://www.youtube.com/watch?v=wFv1Ml_gm1c) (in Finnish), published 13 October 2026, 1:09:57
- **Guests:** [Kari Angeria](https://www.neuvottelija.com/podcast/guests/kari-angeria/) and [Janne Ikola](https://www.neuvottelija.com/podcast/guests/janne-ikola/)
- **Book:** Janne Ikola and Kari Angeria, *Vedenjakaja – Tekoäly ja työn uusjako*, Basam Books 2026 (in Finnish)
- **Earlier episode:** [Europe as Easter Island | Risto Linturi, Kari Angeria | Negotiator 394](https://www.neuvottelija.com/ai/ep394-euroopasta-tulee-paasiaissaari-linturi-angeria/)
- **Source:** the episode's own Finnish subtitle track and the publisher's chapter marks

## How the episode runs

- **00:00** — Episode highlights
- **01:05** — Guests introduce themselves, Janne Ikola first
- **02:20** — Paid partnership: Posticast
- **03:27** — Back to the episode, Kari Angeria introduces himself
- **04:35** — The book Vedenjakaja and the redistribution of work
- **05:25** — The Luddites and earlier technological upheavals
- **07:18** — Is this time really different?
- **07:59** — Organisational friction slows change
- **09:33** — Tasks are replaced, work is a wider whole
- **12:16** — Programmers' experience predicts other knowledge work
- **14:16** — Coders become the company's orchestrators
- **15:49** — Agents' goals expose unclear roles
- **17:05** — Human work is poorly specified
- **18:20** — Paradigms and the concept of capital
- **20:35** — AI makes the mediocre free
- **21:38** — The public sector and the risk function of official liability
- **23:55** — Tokenmaxing and the real value of tools
- **25:12** — Individual effectiveness before team effectiveness
- **27:33** — Hoarding resources and personal licences
- **29:37** — AI-native uses not yet found
- **31:14** — Agent traffic overtook humans on the web
- **33:40** — Robotisation advances as a slow avalanche
- **34:15** — Do people want orders from a machine?
- **37:43** — The human job is to bring entropy
- **40:02** — Tacit knowledge and intuition beyond language models
- **42:14** — World models and neurosymbolic AI
- **43:31** — Kahneman's systems and machine intuition
- **45:45** — Facts, opinions and value choices
- **48:45** — Will companies be needed in the future?
- **49:26** — Coase and the coordination problem
- **50:54** — Selling tacit knowledge through agents
- **53:14** — The audience decides: why WhatsApp wins
- **54:12** — Agents as legal persons and diverging realities
- **57:40** — Job tips for the young and the junior problem
- **59:10** — Three new roles: operator, curator, designer
- **1:00:31** — Our own children and responsible talk about the future
- **1:03:02** — Robotics and home automation as new work
- **1:04:46** — Human contact is the scarce resource of the future
- **1:07:12** — A culture of borrowing best practices, and your own vision
- **1:08:06** — Wrap-up, and next topic: the Sotkamo data centre