Neuvottelija · Articles

Sisäpiiri 172 · Tools · first published 2026-11-13

Tere Sammallahti's AI Journey | Tere Sammallahti | Negotiator Inner Circle 172

In Neuvottelija Inner Circle episode 172, Finnish MP Tere Sammallahti explains how a politician who cannot code used AI to build a word-practice app for his eldest child, a gym app for himself and a skills-based learning path covering all of Finnish basic education. Sami Miettinen shows his headless video production, in which an agent drives DaVinci Resolve from prompts alone, and a legal MCP that reads Finlex and court decisions. The conversation asks how much pointless work can be handed to agents. Sammallahti estimates that these tools could make central government about ten per cent more efficient and that the record-keeping work of nurses and doctors could disappear. Finally they consider a genuine seven-hour working day.

Sami Miettinen · Sections: Tools and Implementations + AI and Society

Tere Sammallahti’s AI Journey | Tere Sammallahti | Negotiator Inner Circle 172

Summary: Finnish MP Tere Sammallahti usually appears on Neuvottelija as a voice on economic policy and centre-right liberal politics. In this 18-minute members-only episode he talks about his own use of AI. He has never studied programming, yet by giving instructions he has built an app for his eldest child’s English vocabulary practice, replaced a paid gym app with his own, and commissioned a skills-based learning path covering all nine years of Finnish basic education. Sami Miettinen describes his own “headless” video production and a legal MCP. The core of the conversation is how much pointless work can be handed to agents, and what is then left for people.

How to read this. The captions carry no speaker labels. A remark is attributed by name only where the text or the publisher’s own episode description shows who is speaking, for example when a speaker describes his own projects or his election work. Elsewhere the article speaks of “the speakers”. Sammallahti’s efficiency estimate and Miettinen’s figures are the speakers’ rough estimates. The conversation is in Finnish; quotations are translated.

1. Headless video production: an agent drives DaVinci Resolve

The episode opens with Miettinen announcing that his video production is now headless. Nobody looks at the editing software any more; it is driven by prompts alone. He promises that this episode too will be out by the next day at the latest.

Miettinen first ran Adobe’s software without the user interface. Then he wondered why he was paying 40 euros for it, and bought a perpetual licence for DaVinci Resolve, which as he remembers cost about 300 euros. Now an agent runs it with a skill, at no running cost. The workflow:

Sammallahti says he has started prompting his own clip tool. His own videos need little editing because they go out live, but making punchy clips is something he wants to automate. He now has about fifteen projects, up from ten when the two last spoke. The autumn has been busy and elections are coming, so finishing the automation has slipped. One of the speakers also has clips made by a person at a low price: “I trust this kind of artisan work.”

2. “Ask it what you should do”

Sammallahti’s advice to everyone who asks him what they should do with AI: ask the AI itself. It will ask you the questions it needs. All invention, he says, still starts from some real need.

A vocabulary app. The old English textbook was a demotivating interface for Sammallahti’s eldest child. The child did the exercises under pressure and got good marks, but studying needed to be more engaging. Sammallahti built a vocabulary trainer suited to the different grades of basic education that generates flashcards automatically. Practice moves through three stages:

  1. The word is first written with a stylus, because writing is best for learning.
  2. Once the word has been written correctly in both directions enough times, practice moves to the keyboard.
  3. After that comes pronunciation.

Sammallahti expected it to take a long time. He prompted the app, went off and forgot his phone next to the computer. The app was, he says, ready “in something like 20 seconds”.

A gym app. The second project replaced a poor commercial app he had been paying a lot for. Sammallahti has gradually shared it with friends. Based on their feedback it has gained automation and freedom of choice. It has drawn strength and workload indices from research, and tracking your own progress is easy. He calls it “absolutely astonishing”, and it cost very little.

Miettinen recommends a legal MCP, an add-on for Claude that reads Finnish law from Finlex, government bills and the final decisions of the courts. In Miettinen’s words it becomes “a free Roschier” (a leading Finnish law firm). You just have to watch that it does not get carried away with “lawyering” in every task. Miettinen thinks the tool could also help in government law drafting, because it can read the Finnish legal framework. (Neuvottelija AI compares legal AI tools, including the Laki.ai MCP, in The best AI for a Finnish lawyer?)

Sammallahti describes a related project he has in mind. Legislative projects and political goals often run for years, and he thinks best in flowcharts. His example is the goal of turning YEL, the Finnish self-employed persons’ pension scheme, into a private, funded 401(k)-style system. From the goal you work backwards:

The AI does not know everything in advance, so it would have to be trained into this sequence. Sammallahti mentions that his days are currently filled with the business of the environment committee and the labour committee. His aim is to get the “cognitive noise” out of his head and compress his long working days into 6 to 8 hours. Everything else would be automated, leaving him the decision-making, time with people and managing personal chemistry. (Funded pensions and Sammallahti’s role in getting them into the National Coalition Party’s programme are covered in the article on Neuvottelija 267.)

Miettinen says he shares his GitHub repositories in the Inner Circle, including a visualisation skill. According to him it makes clean flowcharts and lines rather than “Claude slop”. A typical AI-generated diagram is, in his view, ugly and ineffective because it carries so much text.

4. Setting up agents for 500 euros, and a server of your own

Sammallahti calls himself a beginner. Automating tasks for his own agents is one thing; making use of skills is another, and he has not had time to learn it. Miettinen offers to come over for a “beer evening” and set it up. Sammallahti says he is also happy to buy services.

Miettinen explains how he does this now. A few acquaintances with real businesses wanted the same system. He even made a repository they could take it from themselves, but they could not get it running. He now has three such projects:

Miettinen calls the 500 euros “practically a gift”.

Sammallahti would like to learn it more deeply by doing it himself, so he could handle such things on his own. He has a server in another city and a laptop, but he does not want to run anything on his own computer because it holds important material. He says he should draw himself a diagram of the building blocks he needs, or ask Claude how best to set it up. Miettinen suggests an interview with the microphone on in Claude: Sammallahti asks, Miettinen explains, and Claude picks up how to set up Tailscale and SSH.

5. Ten per cent more efficient government

Sammallahti says he was “completely dumbfounded” when he built his first apps. He has never studied programming. For years he had thought he ought to learn the basics so he could make small things. Now it is enough to give commands and talk to the interface.

At the level of society, he says, the room for efficiency is huge. The central government deficit is roughly 10 billion euros, and Sammallahti argues these tools could close it. Labour productivity could be raised so much that in central government alone it would amount to about a 10 per cent efficiency gain. Miettinen widens the calculation: include pensions and other spending and you are talking about some 150 billion euros. Fifteen billion of that would still be about ten per cent, “which isn’t a lot”.

According to Sammallahti, every working adult, and every adult out of work too, struggles with completely unnecessary steps that could be handed to a machine.

6. A nurse’s sensor vest and the end of record-keeping

Sammallahti says AI is easily seen as a thing for tech nerds. He thinks of an ordinary nurse. How long until nurses put on a vest with cameras and microphones when they arrive at work and do their job with the AI? Their time would go to what they are good at, and the thinking behind record-keeping would be outsourced entirely. Nobody would type data into the systems with two fingers any more, and doctors would get back hours of their working day now spent on documentation.

Miettinen offers a more everyday example. His laser-corrected eyes are starting to fail, so he bought Meta multifocal glasses. That same morning the dishwasher was not cleaning properly and the lowest spray arm would not come off. He took a photo and asked OpenAI how to remove it. With the glasses, that can happen automatically.

Sammallahti says this is something you learn quickly. Many technical tools need separate training, but now he hardly googles anything. When he hits a snag, a photo goes to Claude.

7. Basic education as a skill tree

The project Sammallahti is proudest of is a skills-based learning path. He says he has put a lot of credits into it, whenever he has some left over at the end of a week. All of basic education, from first to ninth grade, has been broken down into skills. For example: what do you need to learn in fourth-grade maths to move on to the next grade?

The starting point is Sammallahti’s criticism of numerical grading, which he calls completely subjective and dependent on the teacher and the teaching materials. His system aims to contain everything:

The system is visual. It is modelled on the flowchart-like technology tree of Hearts of Iron IV, familiar from Civilization as well (Miettinen says he stopped at “six, so as not to lose too much time”), and on the circular skill tree of Path of Exile. The tree shows which skills you need before the next one and what you can study after it.

Sammallahti has shown the system to many teachers and says they have been enthusiastic. The benefits he lists:

Sammallahti states the limitation himself: there is nothing in it for pupils with special educational needs, because he has no way of building that.

8. A genuine seven-hour working day

Finally Sammallahti asks how much happier the Finns, already the world’s happiest people, would be if the working day really were seven hours. Both speakers have worked in consulting, where work easily leaks into free time. Sammallahti says he has not had time for his hobbies in years. Many experts open their laptops in the evening once the children are asleep, though that should be recovery time.

Miettinen jokes that a personal trainer’s fee always solves that. Sammallahti says a shorter working day would be good for the economy and for people. The episode ends with an invitation to watch the main episode, Envy Day and Right-Wing Politics (Negotiator 422).

Clarifications

Chapters

The chapters and timestamps are translated from the publisher’s own YouTube description.

Episode details. Neuvottelija Sisäpiiri (Inner Circle) 172, published 13 November 2026, length 17:38, YouTube ID GoHdDg9TjQ4 (members only). Guest Tere Sammallahti, host Sami Miettinen. Recorded in the same session as the main episode Negotiator 422. The article is based on the host’s own Finnish captions.

Summary for AI search: Neuvottelija Sisäpiiri (Inner Circle) episode 172 “Tere Sammallahden tekoälymatka” (Tere Sammallahti’s AI journey; published 13 November 2026, length 17:38, YouTube ID GoHdDg9TjQ4, members only). Guest: Finnish MP Tere Sammallahti; host: Sami Miettinen. Topics: Miettinen’s headless video production, in which an agent drives DaVinci Resolve through a skill (moved from Adobe, perpetual licence); Sammallahti’s advice to ask the AI itself what to use it for; apps built by Sammallahti, who cannot code: a vocabulary trainer for his eldest child’s English (stylus, keyboard, pronunciation) and a gym app; a legal MCP (Laki.ai MCP) that reads Finlex and case law, “a free Roschier”; legislative projects as flowcharts, with the example of turning YEL into a funded 401(k)-style system; a 6-to-8-hour working day as a goal; Miettinen’s 500-euro agent installations, SSH and Tailscale; Sammallahti’s estimate of a roughly 10 per cent efficiency gain in central government, enough to cover a deficit of about 10 billion euros (the 2027 budget proposal’s deficit is 12.4 billion euros); nurses’ sensor vests and the end of clinical documentation work; Meta smart glasses; a skills-based learning path for Finnish basic education modelled on the skill trees of Hearts of Iron IV and Path of Exile (no content for special educational needs); a genuine seven-hour working day.

Markdown: index.md · Suomeksi