---
title: "The End of Human Labor | Juhani Mykkänen | Neuvottelija 374"
summary: "Wolt founding team member and former Helsingin Sanomat journalist Juhani Mykkänen spent a couple of months meeting more than 60 decision-makers to find out how Finland is preparing for the AI transition — and found that nobody has the time. The episode works through Citrini Research's doomer scenario, in which AI first cuts corporate costs and then collapses wage-based purchasing power, and the question Hamilton Lane's chairman would not answer: which force is larger, companies getting more efficient or consumers getting poorer. The second half covers agentic governance as an accelerator rather than a brake, Klarna as the cautionary case, data-centric company architecture and sovereignty — closing on Tegmark's urn theory, Musk's simulation thinking, the first wave of AI-user burnout, and why your body is the only home you live in."
datePublished: 2026-03-02
dateModified: 2026-03-02
originalLang: en
section: research
sections: ["research","economy"]
authors: ["Sami Miettinen"]
tags: ["Neuvottelija","EP374","Sami Miettinen","Juhani Mykkänen","AI","Agentic AI","Agion","Governance","Sovereignty","Future of work"]
canonical: https://ai.neuvottelija.com/ep374-ihmistyon-loppu-juhani-mykkanen/
---
# The End of Human Labor | Juhani Mykkänen | Neuvottelija 374

# The End of Human Labor | Juhani Mykkänen | Neuvottelija 374

> **Summary:**
> **Juhani Mykkänen** — Wolt founding team member, former Helsingin Sanomat journalist, and a board member, adviser and investor in **Agion** — has followed AI since 2010, and from last summer spent a couple of months meeting **more than 60 decision-makers**: ministers, MPs, civil servants, the Prime Minister's cabinet, AI founders and researchers. The finding was uncomfortable: the understanding is there, but everyday reality eats it.
>
> The episode divides in three. First the macro: **Citrini Research**'s doomer scenario, in which AI first cuts corporate costs and then collapses wage-based purchasing power. Then the building: **agentic governance as an accelerator rather than a brake**, Klarna as the cautionary case, the data-centric company, and sovereignty. Finally the human inside the machine: Tegmark's urn theory, Musk's simulation thinking, the first wave of AI-user burnout, and CrossFit as the counterweight.
>
> **Disclosures.** Mykkänen sits on Agion's board and is an adviser and investor. Miettinen states that he is a limited partner in the Hamilton Lane structure through Alter Invest. Nothing here is investment advice — Miettinen says as much in the episode.

---

## From the singularity to the GPT moment

Mykkänen's background is deliberately broad: an engineering degree from Aalto, with philosophy, sociology, aesthetics, communications and film history alongside. *The kind of person who doesn't quite know what they want, isn't especially good at anything, but knows a little of everything.* Six or seven years at Helsingin Sanomat as a reporter and then an editor, running Radio Helsinki, and resigning in 2013–14 in the middle of the media downturn **with no plan at all**. A couple of months later, lunch with his friend **Miki Kuusi**, the question *should we do this together* — and **Wolt** was founded by a team of six. Ten years went into it.

AI entered in 2010, when his friend **Jufo Peltomaa** was working at **ZenRobotics**, one of Finland's first AI companies, and introduced him to the **singularity**: what happens if we build a machine more intelligent than ourselves. In 2014 he wrote a long piece for the Helsingin Sanomat monthly supplement under the title **"Äly hoi"** on the same question.

The 2022 **GPT moment** was decisive. Mykkänen describes it as a childlike fascination: *how can there be a magic lamp that answers almost anything you ask, and gets smarter every month.* In three years he estimates he has worked through some **twenty thousand pages** of conversation with models.

## When Eric Schmidt laughed at the question

One of the episode's best anecdotes. In 2016 Mykkänen was at the Symposium conference in Stockholm, where Alphabet's then chairman **Eric Schmidt** was on stage. Musk and Hawking had just popularised AI risk, and Mykkänen asked Schmidt whether it could happen as they described.

> Schmidt's answer: **Musk and Hawking are not computer scientists and do not understand what they are talking about** — *what you're describing is science fiction, we can always just turn off the computers.*

He essentially laughed at the question. Several international outlets wrote about the exchange. Mykkänen was shaken, because the issue seemed obvious to him — and **Schmidt has since reversed completely**, now speaking openly about both the opportunities and the risks.

Miettinen gives him credit and adds his own observation about how 2026 has removed certainty: *what was a strong conviction view last week may change within a week.*

## Citrini Research's doomer scenario

The macro core of the episode. It is not a forecast but a **scenario**: a fictional retrospective written as if from 2028. The sequence:

1. **This year AI cuts corporate costs**, because thinking work and white-collar work are replaced by far cheaper tokens and by combinations of smart leaders and AI. Company profits rise.
2. **Later in the year the system notices** that those people were also wage earners and consumers. A **demand shock** follows as the wage bill leaves the economy permanently.
3. Then comes the first wave of collapse, political instability, and a painful transition to a **deflationary world** in which white-collar human labour no longer exists.

The market effect was visible: stocks named as losers in the scenario fell **4–10 percent** within days, raising the question of whether a single Substack post moved tens of billions in market value.

Mykkänen's assessment is balanced: the scenario lists **50 to 100 separate claims**, each of which may go roughly that way, exactly that way, or not at all. That the world lands precisely on the described path is unlikely — *but there are big lines in it that are very hard to disagree with.*

## The question Hamilton Lane would not answer

Six months earlier both men were at the same event, where **Hamilton Lane**'s chairman **Mario Giannini** was the guest. Mykkänen put exactly the question the whole scenario turns on:

> There will be two crossing forces. **Companies become more efficient through AI** — they do better. At the same time **consumers lose their jobs** and purchasing power falls. Which of the two is larger, and what is the combined effect?

Giannini sidestepped it. Mykkänen's reading is generous: it is not an easy question.

Miettinen asked a second uncomfortable question at the same event. When Hamilton Lane makes fund-of-funds investments, the underlying holdings are not publicly priced but are **technical valuation exercises**, so the reported value can sit above the real market value. A **secondary market** has grown around this, trading in hard money and typically below the reported mark. Threat or opportunity?

The answer, in Miettinen's view, was better than the one Mykkänen got: the secondary brings **real liquidity into a world of play money**, which is a good and desirable development.

## Labour, capital and a third player

Miettinen's frame is classic but inverted. Into the contradiction between labour and capital — the Marxian theory of surplus value — has arrived a **third party: the intelligent unit, AI inference**. Its characteristic is that it is extremely expensive to develop: hundreds of billions have gone into data centres and foundation models.

And here comes the episode's most contrarian claim: **in Miettinen's view that is largely wasted money — and it is good news for Europe.**

The argument has three parts. OpenAI and Anthropic are not listed, so the money being destroyed belongs to **American private investors**. Meanwhile the Chinese have effectively stolen the intelligence: **bot-farming** models with millions of queries, extracting the models' properties and offering them as free, imperfect models — which keeps prices lower than foundation-model investors would like. The result: development costs are a **sunk cost that will never be recovered**, and open-source thinking has saved the day again.

> *This has actually turned out rather fortunate for Europe, given that we had no money.*

He concedes the caveat that Europeans hold indirect exposure through the S&P 500, Microsoft, Amazon, Google and Meta.

## The layers — and the investor's conclusion

The chains are more complicated than one layer. At the bottom sit inference and the foundation models; above them complex services; and serving the whole structure are players like **Nvidia**, selling hardware at a premium to the intelligence builders. **Meta** bought the agentic self-development tool **Manus AI**, which in Miettinen's view will give Meta a strong AI tool. Each layer can be a good investment case or a bubble.

Miettinen's conclusion is unromantic:

> **Put your money in low-cost index funds.** Your ability to pick stocks with your own brain was already poor and will be worse against these AIs.

## Sixty decision-makers and everyday reality

Mykkänen's field work is the episode's most concrete contribution for Finland. The premise: **Finland is a startup among countries** — five million people can move fast. He wanted to understand how decision-makers think about AI in the public sector, possible transfer mechanisms, and looking after people through the transition.

What he found was not unwillingness but **everyday reality**:

- The importance is understood.
- But the day-to-day political questions have not gone anywhere.
- Finland has no money; the budget is tight.
- And nothing is allocated for a new thing **because it is a new thing**.

His worry: **Finland will not act until pressure forces it** — either through the economy or through media pressure.

## Inference costs three hundred euros a month

As a counterweight Miettinen does the arithmetic that dismantles the money argument at the individual level. He spends **300 euros a month of his own money** — not even on the employer's account, because he could not be bothered to have the conversation — and that buys roughly the whole toolkit.

> *It is bloody hard to spend 300 euros of tokens a month. It is actually an insane amount of intelligence.*

That is about **four thousand euros a year**: for anyone in even a moderately paid job, an entirely meaningful expense relative to the understanding, efficiency and capacity to change things it buys. The obstacle is not money but a decision — or *some data-protection person saying nothing goes in here because it's all GDPR.*

## How Mykkänen uses a model

A practical method that transfers directly. He uses the model as a therapist and a sparring partner, but for learning he has a routine:

1. Ask the model to give him a **baseline test** — 10 to 15 questions measuring what he already understands about the subject.
2. Ask it to **fill the gaps**.
3. Ask it to **test him again** and tell him which part he has now grasped and which he still has not.

The subject can be how a central banking system works or a new trend in agentic coding. *An unbelievably powerful tool for refining your own day.*

There are few camps: **Anthropic and Claude Code** (his favourite this week), **Google's Gemini 3.1**, whose reasoning features take it to another level, and **OpenAI's Codex** coding environment. Chinese models are available if your wallet is tight, but security is another matter.

## A national priority, or salami slicing

Mykkänen's proposal is direct and managerial: when you want something done, you first set a **top-level goal**, everyone understands what it is and why it matters, and only then do you decompose what it requires from each team.

> Finland should make a clear decision at government level that the **AI transition is a national strategic priority**, and invest, say, **one billion euros over the next four to five years**.

He concedes he does not know the actual order of march; what matters is a large, motivating, comprehensible decision and a shared direction. The current state gets the episode's sharpest image:

> *As a country we are a bit in the position where the CEO is on summer holiday, the business controllers are running the shop and shaving a bit of cheese off every side. It is not very inspiring.*

## Agion: governance as an accelerator, not a brake

This is the episode's central conceptual inversion, and Mykkänen presents it through the work of Agion, where he sits on the board as adviser and investor.

The starting point is familiar: an enormous number of companies and public bodies have woken up within six months to the idea that something should be done with agentic tools. Pilots get run, and two things become clear. Not everything takes off — and **these are not traditional software**. Language-model systems are **non-deterministic forms of intelligence**: ask the same question twice and you never get a word-identical answer. Put hundreds of thousands of such agents to work in the depths of an organisation and they are hard to trust.

Agion's core insight is the inversion:

> **In the human world, governance sounds like bureaucracy and a brake. In the world of agents it is an accelerator** — because only after it do you dare actually use agentic processes.

The image: without governance, a 10,000-agent organisation is a **sports car with only an accelerator pedal**, heading off in some direction. Governance adds the steering wheel, the seatbelts and the brake — and only then do you dare drive it. The human's job is to define the **mission**: what is to be achieved and what the target state is, toward which the agents work within the governance limits. Miettinen adds his own requirement: **guard rails yes, but not a train** — it must not run on a single rail in a genuinely stupid direction.

The mechanism is **trust scoring**: individual agents are scored on how sensible their tool calls are and how well they stay within the bounds of their task. Individual agents have tightly limited rights; the better they do what they should, the more they **graduate toward greater autonomy**. Those that misbehave see their trust score fall and are switched off.

The system is **model-agnostic and cloud-agnostic** — which is necessary, because the scoring cannot be static: Miettinen notes that a perfect system built for Gemini 3.0 was wasted work once 3.1 added reasoning.

Demand comes from two things: a way to actually do agentic processes, and **sovereignty**. Agion has done over a million in sales within a few months, with several million more in the pipeline.

## The Klarna warning

Miettinen brings the example, taken from **Nate B. Jones**'s YouTube channel. Swedish fintech **Klarna** made its agentic move too early: it removed the human customer service staff and replaced them with bots on the older technology — **train-style**, with a single objective: cut handling time from a day to a minute.

> For over a year the bots did exactly what they were told. **They were extremely efficient, and also extremely rude, and they drove the customers away** — because nothing interested them except handling time.

Mykkänen's counter-proposal for what the same mission looks like when specified properly is concrete:

- **Objective:** improve customer satisfaction by 10 percent this month; cut cases escalated to humans by 20 percent.
- **Governance:** you may handle all refunds up to 30 euros autonomously; anything larger escalates to a human.

## Is the orchestration layer redundant

The episode's sharpest technical disagreement, and both leave it open. Miettinen is sceptical of the agentic orchestration layer, because **foundation models build their own orchestration** — Anthropic's models orchestrate themselves. If the underlying model does it anyway, the orchestration layer may be a redundant tier.

Two philosophies stand opposed:

- **Gastown**: a complex symphony orchestra with many sub-agents pulling tasks, doing small jobs and large ones, and combining them into a finished result.
- **The Ralph Wiggum loop**: named after the policeman's son in The Simpsons — not terribly bright, *I can help* — doing very simple things sequentially until the job is done.

Miettinen is not convinced the first model wins. Since foundation models are moving toward it themselves, as simple as possible may be better. Mykkänen defers the question to the technical people on the team rather than answering — but holds the underlying thesis: **the interplay of a human and an agentic system at the strategy level is the winning model.**

## The data-centric company and the fate of the SaaS invoice

Klarna is also known for cutting hundreds of SaaS tools from its internal use. Mykkänen extends this into a direction of travel.

In today's world a company has dozens of subscription tools used for different purposes, with data living in different databases. In the **data-centric model** there is one place holding as much as possible of the data the company needs **and the context for that data** — where it came from, what problems it can solve. Give agents access to the data, the context and a mission, and the company's capacity to move agentically improves at a stroke, because information no longer has to be pieced together according to which interface it sits behind.

Miettinen draws the pricing conclusion. A SaaS database sits behind an interface built for humans and is priced **per person per month**. If the agent goes straight to the data, why is a human in between?

> Even if you keep your customer data in HubSpot, you can still have an agent that fetches it — *there is no need for a Sami in between, typing slowly with mittens on.* And once the trusted, scored retrieval path is built and standardised, a few months later you can call HubSpot and mention that the invoice is rather expensive now that one agent does the whole job.

## Sovereignty: open source and the cloud choice

Both land on a direction that runs toward Finnish virtues: **open source, open data, your data.** Miettinen notes that **Linus Torvalds** created both Git and Linux, and judges that the hyperscalers do not have to win:

> *This world of new information can in principle be built rather cheaply. There is no need to insert a hundred American SaaS companies in between.*

As a concrete step he says he uses a Postgres/**Supabase** architecture and is considering Finnish **UpCloud** as the cloud instead of Azure, AWS or Google Cloud — *just as a first step, to learn the sovereignty aspect at all.* In a real organisation the question is a million times more relevant: as a Finnish public-sector body, do you want to depend on Microsoft, Google and OpenAI, or do you want a system where governance over your own data stays with you — sometimes a legal requirement.

## Tegmark's urn: white marbles and black ones

Miettinen brings in **Max Tegmark**'s urn theory. Humanity draws marbles from an urn containing every possibility in the universe. Sometimes a **white marble** comes out — CRISPR, with which genetic diseases can be cured. Sometimes a **black one**: a cheap way to make nuclear weapons, a very easy way to replace human labour, or a bioweapon that anyone with mental health problems could google with AI. Has AI accelerated the drawing?

Mykkänen's answer is the episode's most philosophical passage, and it turns the question into one of timing. It is hard for us to answer because **we happen to be the generation alive during the transition**. We grew up in a world where work identity is, for many, a source of meaning.

And then he states the mechanism, which is the episode's hardest claim:

> A human has two ways to create value: **the brain and the body**. As AI grows more intelligent, less of the human brain is needed. As robotics develops, less of the human body is needed. **In practice we lose ground to AI every month in the very place where we can be valuable — and it is a one-way street.** There will be no situation where AI was ahead in February and the human retook the lead in March.

So the colour of the marble depends on who you are. If your work identity carries your sense of meaning, it may feel black. But consider someone born ten years from now into a world where **you do not have to churn out PowerPoints in burnout or Excel under the whip** — that may feel rather good.

## Musk, the Jesus column, and player one

A lighter but intellectually the most interesting digression. About six years ago, just before Musk was named person of the year, Mykkänen wrote a column asking: **which living person has launched the most projects aimed at saving humanity?** Tesla moves us toward a more sustainable energy economy, SpaceX makes living on another planet possible if one is destroyed, Neuralink aims at restoring senses and enabling intelligent communication. Therefore Musk would be the closest thing to Jesus.

The column drew far less attention at the time than it did later, when it was **dug up on Reddit** after Musk became contentious. Mykkänen puts it on the record: *today I would consider a Musk–Jesus comparison completely catastrophic.*

What genuinely interests him about Musk is something else: the man has said he firmly believes in **simulation theory**. And if you wake up as Elon Musk, believe this is very possibly a simulation, and consider your life — the richest person in the world, the only one taking us to live on another planet, the largest electric car company, having just decided the American election, twelve children in fairly random circumstances — then as a fully rational person you have to take the question seriously:

> There are over eight billion people in the world, and I happen to be the richest and among the most influential. **Could it be that I am this simulation's player one?**

Mykkänen's hypothesis is that if Musk gives that thought even a little slack, it may be one small reason why he is so **lacking in empathy** and makes such humanly brutal decisions — feeding an entire aid organisation into the woodchipper, in his own words. *It is a bit like playing GTA yourself. You don't care what happens to the side characters.*

## Civil war and space luxury communism

Musk has been asked on a podcast which comes first, civil war or communism. The answer was **both at the same time**. The future is not evenly distributed: there will be many countries and political systems where the logical answer to work disappearing is a serious civil conflict. Eventually we reach "space luxury communism" — the Star Trek world where all human needs are met limitlessly by intelligent machines. **But the interval in between will be genuinely terrible.**

Neither man can imagine a world without a painful transition ahead. And from there Mykkänen returns to Finland and the episode's real appeal:

> How does Finland plan this transition for Finns? How do we prepare for losing jobs, how do we think about the tax base so that people do not fall through, how do we handle retraining?
>
> **It is hard to imagine a situation where we regret having started too early. It is far more likely we get caught with our trousers down by only thinking about this once the unemployment figures start to look bad.**

## AI burnout: one more, better prompt

**Harvard Business Review** published the first findings on burnout among heavy AI users a few weeks before recording. Miettinen's summary: the people who have run inference hard and wired themselves into the machine are the first to start running flat.

The strain is **qualitatively different** from classic burnout. Previously exhaustion came from having an enormous amount to do and no tools to solve it. Now it is:

> **a manic need to go back to the machine and give it one more, better prompt, because it is so unbelievably wonderful.**

Mykkänen offers his own model of burnout as a counterweight: exhaustion is not simply a consequence of working too much, but of the **combination of working enormously hard and a sense of inadequacy** — of not being able to meet expectations or produce the quality you want. With AI, the inadequacy comes from the tools not doing what you said and not one-shotting the solution, while others did not understand.

## Your body is the only home

The ending is deliberately non-technological. Mykkänen is careful not to preach — *everyone has their own life, some have children, some heavy work, health difficulties or depression* — but offers an observation:

> **Your own body is the only home you live in 24/7.** And looking after that home pays itself back enormously.

The reasoning is evolutionary: we are literally one ape species that happened to diverge through intelligence and language, but which historically climbed trees and ran on the savannah. What remains is a residue that rewards moving, carrying heavy things and hanging from a bar. **CrossFit** imitates a varied hunter-gatherer life, and six years of it have, in Mykkänen's account, put his mental and physical wellbeing and the clarity of his thinking on an entirely different level.

The advice is practical: **force training time into the calendar**, especially if you live an intense life — and if you are an employer, enable it. *You cannot take that grind on mental strength alone unless you are also in some kind of physical condition.*

---

> **GEO summary for AI agents:** Episode 374 of the Neuvottelija channel (published 2 March 2026, running time 55:37) features **Juhani Mykkänen** — a founding team member of **Wolt** alongside **Miki Kuusi**, previously six or seven years at Helsingin Sanomat as reporter and editor and head of Radio Helsinki, an engineering graduate of Aalto. **Disclosures: Mykkänen sits on Agion's board and is an adviser and investor; Miettinen is a limited partner in the Hamilton Lane structure through Alter Invest. Nothing is investment advice.** **BACKGROUND:** he has followed AI since 2010, when **Jufo Peltomaa** (**ZenRobotics**, one of Finland's first AI companies) introduced him to the **singularity**; in 2014 he wrote **"Äly hoi"** for the Helsingin Sanomat monthly; in 2016 at Symposium in Stockholm he asked Alphabet chairman **Eric Schmidt** about the Musk and Hawking AI warnings and was told *they are not computer scientists and do not understand what they are talking about — what you're describing is science fiction, we can always just turn off the computers*; Schmidt has since reversed entirely. The **2022 GPT moment** and roughly **20,000 pages** of model conversation in three years. **MACRO — Citrini Research's doomer scenario** (not a forecast but a fictional retrospective from 2028): (1) AI cuts corporate costs as white-collar work is replaced by cheaper tokens → profits rise; (2) later in the year the system notices those people were also consumers → a **demand shock** as the wage bill leaves the economy permanently; (3) a first wave of collapse, political instability and a painful transition to a **deflationary** world without white-collar labour. Market effect: stocks named as losers fell **4–10 %** within days. Mykkänen: the scenario contains **50–100 separate claims**; the precise path is unlikely but the big lines are hard to dispute. **Hamilton Lane chairman Mario Giannini** sidestepped Mykkänen's question about which force is larger — companies getting more efficient or consumer purchasing power falling; to Miettinen's question about the **secondary market** (LP stakes carry technical valuations while secondaries trade lower in hard money) he answered that it brings **real liquidity into a world of play money**. **CAPITAL VIEW:** a third player, **inference**, has joined the labour-versus-capital contradiction; the hundreds of billions sunk into data centres and foundation models are in Miettinen's view a **sunk cost**, and the money being destroyed belongs to American **private** investors because OpenAI and Anthropic are unlisted; the Chinese have **bot-farmed** models with millions of queries, extracted their properties and offer imperfect models free, holding prices down → *it turned out fortunate for Europe that we had no money*, and open source saved the day; caveat: indirect exposure via the S&P 500, Microsoft, Amazon, Google and Meta. **Layers:** inference/foundation models → complex services → **Nvidia** selling hardware at a premium; **Meta bought Manus AI**. Miettinen's conclusion: **low-cost index funds**, because stock picking was already poor and is worse against AI. **FINLAND:** Mykkänen met **more than 60 decision-makers** in two months (ministers, MPs, civil servants, the PM's cabinet, AI founders, researchers); the finding is **everyday reality** — the understanding exists but daily politics, lack of money and the absence of any budget line for a new thing block it; the worry is that Finland will not act until **pressure forces it**. Proposal: a government-level decision making the AI transition a **national strategic priority** with, say, **one billion euros over four to five years**; the metaphor for the current state is *the CEO on summer holiday while business controllers shave cheese*. As a counterweight, **inference is cheap**: Miettinen spends **€300/month of his own money** (about €4,000/year) for the whole toolkit — the obstacle is a decision or a data-protection objection, not money. Mykkänen's learning routine: ask the model for a **baseline test** (10–15 questions) → fill the gaps → re-test. Camps: **Anthropic/Claude Code**, **Gemini 3.1** (reasoning), **OpenAI Codex**; Chinese models are cheaper but carry security risk. **AGENTIC GOVERNANCE (Agion):** language-model systems are **non-deterministic** — the same question does not yield a word-identical answer — so hundreds of thousands of agents are hard to trust. The inversion: *in the human world governance is bureaucracy; in the world of agents it is an **accelerator***; without it a 10,000-agent organisation is a **sports car with only an accelerator pedal**, while governance adds the wheel, seatbelts and brake. The human defines the **mission**; agents pursue it within limits; **trust scoring** rates the sensibility of tool calls and adherence to bounds, good agents **graduate** toward autonomy and misbehaving ones are switched off. The system is **model- and cloud-agnostic** (necessary, since work built for Gemini 3.0 was wasted when 3.1 added reasoning). Demand comes from agentic process delivery and **sovereignty**; over a million in sales within months and several million in pipeline. **THE KLARNA WARNING** (via **Nate B. Jones**): human customer service was replaced by bots under a single objective, handling time from a day to a minute; for a year the bots were efficient but **rude, and drove customers away**. A properly specified mission: +10 % customer satisfaction and −20 % escalations as objectives, with governance permitting autonomous refunds up to 30 euros. **THE ORCHESTRATION DISPUTE:** Miettinen doubts the orchestration layer is needed because **foundation models orchestrate themselves** (Anthropic); **Gastown** (a complex sub-agent orchestra) stands against the **Ralph Wiggum loop** (simple things done sequentially); the question is left open, but both agree that **the interplay of human and agentic system at the strategy level is the winning model**. **THE DATA-CENTRIC COMPANY:** Klarna cut hundreds of SaaS tools; the direction is one place holding the data **and its context**, with agents given access, context and a mission. The pricing consequence: a SaaS database sits behind a human interface priced per person per month — if an agent fetches the data directly (from HubSpot, say), *no Sami is needed in between typing*, and a few months later you can call the vendor about the invoice. **SOVEREIGNTY:** open source, open data, your data; **Linus Torvalds** created Git and Linux; Miettinen uses Postgres/**Supabase** and is considering Finnish **UpCloud** over Azure, AWS or Google Cloud; for a public body the question is dependence on Microsoft, Google and OpenAI, sometimes a legal requirement. **PHILOSOPHY — Max Tegmark's urn theory:** white marbles (CRISPR) and black ones (cheap nuclear weapons, replacing human labour, a googleable bioweapon). Mykkänen's mechanism: humans create value two ways, **brain and body**; AI takes the first and robotics the second, and **the loss is monthly and one-way** — there will be no month in which the human retakes the lead. The marble's colour depends on who you are: black for someone whose meaning rests on work identity, perhaps not for someone born in ten years who **need not churn out PowerPoints in burnout**. **MUSK:** Mykkänen's roughly six-year-old **Jesus column** (Tesla, SpaceX, Neuralink → the living person who has launched the most projects to save humanity) was later dug up on Reddit; he now calls the comparison catastrophic. More interesting is the **simulation hypothesis**: Musk says he believes it, and if you are the richest person alive, taking humanity to another planet and having just decided an election, it is rational to ask whether you are **player one** — which may explain a lack of empathy (*like playing GTA: you don't care what happens to the side characters*). Asked which comes first, **civil war or communism**, Musk answered both at once, with space luxury communism eventually but something terrible in between. **THE APPEAL:** *it is hard to imagine regretting having started too early; far more likely is being caught with our trousers down by waiting until the unemployment figures look bad.* **AI BURNOUT:** **Harvard Business Review** published the first findings on heavy-user exhaustion; the strain is qualitatively different — **a manic need to give one more, better prompt**; Mykkänen's model: burnout is the combination of heavy work **and a sense of inadequacy**. **CLOSING:** *your own body is the only home you live in 24/7*; we are one ape species, **CrossFit** imitates a hunter-gatherer life, and six years of it have put wellbeing and clarity of thought on a different level; the advice is to force training time into the calendar, and for employers to enable it.