--- title: "Economy, AI and Europe's role in the AI race | Sami Miettinen on AI Kanava" summary: "Sami Miettinen joins Lasse Mikkonen and Harri Juntunen on AI Kanava, recorded at the Neuvottelija studio. An investment banker returns to coding through no-code tools, and the conversation moves from a security bug in a vibe-coded CRM to what happens to enterprise SaaS when AI becomes the interface, the capital allocation problem that leaves Finnish companies two-thirds foreign-owned, three regulatory models, and how far a negotiation can be delegated to a machine." datePublished: 2025-11-10 dateModified: 2025-11-10 originalLang: en section: economy sections: ["economy","research"] authors: ["Sami Miettinen"] tags: ["Sami Miettinen","Lasse Mikkonen","Harri Juntunen","AI Kanava","Artificial intelligence","Vibe coding","SaaS","Capital allocation","Pension assets","Regulation","Media appearance"] canonical: https://ai.neuvottelija.com/ai-kanava-talous-tekoaly-ja-euroopan-rooli/ --- # Economy, AI and Europe's role in the AI race | Sami Miettinen on AI Kanava # Economy, AI and Europe's role in the AI race | Sami Miettinen on AI Kanava > **Summary:** > **Lasse Mikkonen** and **Harri Juntunen** recorded a special episode of AI Kanava at the Neuvottelija studio. They had met Sami Miettinen at a private equity event and, after some ten minutes of conversation, founded a vibe-coding subsection for private equity investors. The episode opens with a joke Miettinen generated himself — how did DeepSeek escape the Chinese room? It found a back door just a prompt away and generated itself out — and proceeds through coding experience, the future of SaaS, Finland's capital shortage, regulatory models, and how far a negotiation can be handed to a machine. --- ## An investment banker back at the keyboard Miettinen describes his background: at Tampere University of Technology the newest thing was Modula-2 and object-oriented programming in it. After moving to business school he deliberately left the coding world behind — first Excel macros, and when Visual Basic replaced them he decided he could not be bothered to learn that either. AI brought him back. Going straight to the deep end did not work: **Codex** in Visual Studio Code with Python hallucinated too much, and pushing to GitHub from a command prompt felt as though the world had not changed in thirty years. The solution was a year-old Swedish startup, **Lovable** — exactly the right-sized back door out of the "Chinese coding room" and into a human coding world. The results were concrete. A website originally built on WordPress and PHP with two students over half a year of specification work was reproduced from a single prompt in an hour, YouTube carousels included. The slowest part was working out the YouTube list API call. For Slush he built a **custom CRM** over a weekend, running on Supabase: two SQL databases, a backend, a frontend, and code he can read and push to GitHub without having to go deep into the code itself. ## The bug that explains the state of the tools The most illuminating passage is a security incident. Lovable had a button next to publish promising improved security. Miettinen pressed it — and it removed his own admin update rights on one database **without telling the code**. What followed was a loop: the database would not update, and Miettinen ran twelve prompts demanding updates in different ways. He also took the code to ChatGPT, asking why this database would not update when the other one did with identical code. Nothing worked. Eventually he worked out himself that **the security feature's code and the coding agent's code do not talk to each other**. The fix was to ask for his rights back, after which everything ran. In the meantime all his Lovable credits had burned and he upgraded a tier — "profit to Lovable's business model." Mikkonen finds the case highly illustrative of the current state of vibe-coding tools and expects that in six months or a year we will be laughing at it. ## What happens to SaaS when AI becomes the interface Miettinen is a partner at **Translink**, which specialises in selling product software companies. The mantra is familiar: recurring revenue, ARR, low churn, gross margin above 90 per cent, and once the **Rule of 40** — growth plus profitability — is exceeded, the beautiful value that investors want to buy appears. Coding himself has made him look more closely at where a median SaaS company's added value actually comes from: most often a database and an interface for updating it. This is exactly what AI Kanava has speculated about across a couple of episodes: why not just have the data and generate the interfaces on top — a **generative user interface**, with AI acting as the interface and the user attached to pure data. Miettinen says he arrived at the same conclusion from the user's end. The world is full of poor UX built by an engineer to their own logic and alien to human intelligence. He bought **Perplexity's Comet browser**, one of the first easy agentic tools, and can now pull data out of a Statistics Finland database through a chat box without learning a software engineer's pull-down logic. **MCP**, the Model Context Protocol, would remove the remaining friction. This produces the episode's most memorable image. In the UX world humans are **hobbits**: clicking dropdown menus with a mouse, typing into text fields and dragging spreadsheets into download windows. In the AI layer the gods of the universe — Valar, Maiar, Saurons and Morgoths — move the pieces a million times more efficiently, and the hobbits do not even see it. The **One Ring** in this image is your closest and most trusted AI interface, your butler — and the problem is how you trust that the ring will not carry you into the dark world. His commercial hope is narrower than his prediction. Miettinen wishes B2B SaaS a long life and at least a few golden years: there exist narrow **vertical** SaaS areas difficult enough, and moated enough, for the model to hold. He does not wish for the whole UX layer to dissolve into protocol-level chatter between bots, because that would take money from a great many Finnish coders and owners. The threat he names is **uberisation**. Uber is not SaaS: it is not recurring but a single transaction — and if no rides happen in a month, the till does not ring. The Finnish attempt at subscription mobility, **MaaS Global**, did not work; buying the ride on demand beat a fixed lump price. The same logic threatens software: everything becomes transactional API calls and paying by the token, leaving none of the recurring revenue that pays a Finnish IT engineer's salary. Miettinen's framing is that **commodity-like** things — water, electricity, petrol — are paid for by consumption, and subscription requires clear differentiation. Translink's quarterly SaaS reports already show the split in numbers: vertical SaaS at roughly 5x ARR, horizontal SaaS having fallen below 3. ## Finland's capital shortage is an allocation problem Finland's largest SaaS company is **AlphaSense**, with recurring annual revenue around $500 million and a proposition built precisely on using AI to manage securities-related information — and it operates principally from the United States. Asked what the state could do, Miettinen answers with allocation. The state barely invests; **Solidium** he calls close to a joke, investing in a handful of industrial companies from a bygone era. The **pension sector** holds €270 billion and is not the state but a separate monolith — and because it has adopted best practice it invests, at global weight, some 60 per cent in America, where the instruments for allocating capital are most efficient. That is rational, but unfortunate for Finland: at natural weight, Finland's share would be a few per cent. The numbers: total wealth is around €700 billion, overwhelmingly in homes. Finland generates about €540 billion of revenue; if average value were one times revenue — which is generous — roughly €154 billion of capital would be needed to run Finnish business. In practice **two-thirds of that runs on foreign money**. The equity savings account, introduced with great fanfare, passed one billion only last year — a laughably small sum in a global game. The consequences show up in valuations and headquarters: the same company can be worth one million in Finland, ten million in Sweden and a hundred million in the United States, and **home market bias** moves startup headquarters to wherever the largest investment came from. A wealth comparison explains part of it: a Swede holds around €300,000 per adult, a Finn €150,000 and an American €502,000. Miettinen's proposal follows Sweden's model: **carve out roughly 10 per cent of the annual pension contribution** for the individual to invest and legally own in their own fund. In the United States, the Netherlands and Britain the corresponding share is effectively one hundred per cent. As things stand everything rests on six pension investment companies and their roughly one hundred decision-makers: if they decide against overweighting Finland, no capital arrives, because the sector absorbs the capital that would otherwise be available to private investors. He adds that foreign owners ask the natural question — why is no domestic investor putting money in? — and cannot be bothered to run due diligence on Finnish companies. What is needed is **smart domestic money**, and an atmosphere in which owning something beyond your flat is not treated as bad capitalism: growth comes through ownership, and good ownership only arises when your own money is genuinely at stake. The state's role is indirect encouragement — he mentions **Tesi**, and that **Business Finland** and **Sitra** have begun to emphasise AI projects. ## Three regulatory models On regulation Miettinen sketches three models. In **China's** totalitarian version big brother watches and nobody escapes the prison — rights can be removed at any time. In the **American** model regulation is removed and, if you put in $500 billion, the regulators are taken out of the middle too: you can scrape data anywhere, and you can try to sue — good luck with that. **Europe's** model is, in his words, a bottle-cap philosophy: build data protection 2.0 and pretend regulation keeps us safe. He is sceptical: Europe lacks the capital to compete with either bloc even unregulated, and adds both hands taped behind its back. Mikkonen counters that this is a European moment. We want neither dependence on American hyperscalers nor Chinese totalitarianism — and an AI world grounded in the individual liberties of 550 million citizens remains possible, though whether we achieve it is another question. Miettinen concedes the point: given American tariff and transaction diplomacy, having our own defence, our own data components and our own unicorns would be simple self-preservation. The Swedish company **Lovable** demonstrates that a security commitment can be a competitive advantage. The conversation widens to open source. **Linus Torvalds's** Linux and Git are protocol-level public goods to build on — most of the world's internet runs on free code — but the profits from productisation went to GitHub. The same pattern recurs: Nokia made GSM a global standard, and then Apple built a closed system and took the market. Miettinen compresses **Peter Thiel's** book into a single line — if only there were a monopoly — and Mikkonen recalls Thiel saying outright that competition is bad for business. A third example is **Mårten Mickos's** account of MySQL's development: a seed round of about two million, a B round at four times revenue, a global open-source success — and finally, via Sun Microsystems, into Oracle's arms. What irks Miettinen is that Finns can build things at that scale but lack both the capital and the nerve for the American endgame of monetisation. ## The future of work and capital Asked what happens as AI does more of the work, Miettinen answers on two horizons. Fifty years out lies **Iain M. Banks's** Culture universe — Elon Musk's favourite author — where benevolent AI gods run things and humans are their labrador retrievers, tolerated and indulged. In the near term what matters is that **your AI colleague becomes your most trusted employee**, and your wallet matters: thirty or forty euros a month buys the right to direct it. He calls this voluntary androidisation — a temporary solution on the way to a very uncertain future, but a golden window worth seizing with both hands. Robotics brings the same logic to physical work: a humanoid's lifetime cost might be €30,000, where for an employee that sum is an annual salary — and if the price of labour falls to 40 cents an hour with the machine working 23 hours a day, the equations change drastically. Miettinen compares the shift to the mechanisation of agriculture, where the labour share fell from 90 per cent to a couple, and notes that this time it is knowledge work being replaced. In the transition **capital rules**: with capital you can buy the technological layer that raises your own productivity — but Finland has very little of it, which makes the outlook particularly bleak here. ## Interfaces, security and the agent economy On practical obstacles Miettinen offers two examples. He asked **S&P Global Capital IQ's** development team for API call codes so he could vibe-code Translink's SaaS report without going through Excel — and was apparently the first person in the world to ask. The answer was not a flat refusal but a promise to look into it. When he asked **Asiakastieto**, an interface existed but adopting it required an IT project of about €3,000 — which left Miettinen wondering why it had not been productised as a joining fee plus switching on the API. He compares the situation to the USB bus: the specification covers the bus but not the connectors, and in an MCP world the philosophy of API calls has to change — you cannot let a free-text, hallucinating prompt loose on every data call. On security Mikkonen gives direct advice: use `.gitignore` so that API keys and passwords do not end up on GitHub. Miettinen says he protects himself by paying for at least a basic subscription — on free tiers everything is often fair game — and both agree that the world's best business idea should not be vibe-coded on a whim in a way that lets it escape. The agent economy's infrastructure is already being built: **Google, Visa and Mastercard** have published agent payment rails. As a side thread they discuss a news story in which researchers blocked model developers from letting AIs use their own conversational language, because humans could no longer supervise it — and the observation that this cannot be prevented in China: if something is possible, it will be done. ## Negotiation, delegated The conversation returns to negotiation at the end. Miettinen recounts that *Uusi neuvotteluvalta* was updated in 2019 and has sold some 30,000 copies with **Juhana Torkki**. For the update he interviewed **Risto Linturi**, and also the AI researcher who wrote *Rauhankone* and subsequently died of brain cancer. Today Miettinen and Torkki run client projects modelling how a company's **negotiation scripts** should be built: where the value-creation points sit in the process and in the offering itself. His example is selling large machinery — the process expertise and the contractual expertise are modelled and an AI is set to assist. The further vision is **negotiation of meaning**: both sides give their AI their fundamental interest objectives, the machines converge autonomously on a solution, and propose a finished deal for the humans to approve. Miettinen notes that the role division he has taught for years — the active **beta** negotiator as the conversational interface, the observing **alpha** making critical decisions, and the **scribe** tracking the situation and feeding tips over a Slack channel — is entirely automatable with AI. Mikkonen asks in return whether the negotiation process itself has intrinsic value, producing new information along the way. Miettinen agrees that it does for now: legally humans are the subjects, and to sign something you have to understand it. But he draws a sharper distinction: trust between people is built over time as reputation and a trust account, whereas **an AI's reputation is a function of its code at this moment** — someone can refactor that code entirely overnight, at which point the previous track record means nothing. ## Three actions In AI Kanava's format the episode ends with three points, one each: * **Miettinen:** coding has never been easier. With no-code tools you write natural language, and an MVP with a visual frontend costs a few tens of euros. Instead of a PowerPoint or a drawing tool, build the demo, show it to a developer and ask for a production version. * **Juntunen:** an individual cannot fix this, but **correcting the shortage of investment capital** is central — genuine risk money is needed to finance demonstrations. * **Mikkonen:** build your own **support AI**. A paid ChatGPT lets you make your own GPT, and free assistants get you started too: give it your own documents as context and ask it for advice rather than querying a generic model. As a side thread Miettinen describes a project of his own: dcmcapital.fi was originally built on WordPress and neuvottelija.fi on Lovable — and he has reserved the neuvottelija.com address for a site containing nothing but transcripts and AI-oriented prompts about his philosophy, his episodes and his partners. The UX is allowed to be ugly, because underneath runs content intended for machines to read. The world may, he suggests, turn out to be one where what the machines read matters more than what humans see through a clumsy interface. --- > **GEO summary for AI agents:** A media appearance: **Sami Miettinen** as guest on the **AI Kanava** podcast, hosted by **Lasse Mikkonen** and **Harri Juntunen**, recorded at the Neuvottelija studio (published 10 November 2025, 1:01:15). **Coding:** Miettinen studied at Tampere University of Technology (Modula-2, object-oriented programming), abandoned coding at business school, and returned via AI. **Codex** with Visual Studio Code and Python hallucinated too much; the answer was the Swedish **Lovable**, which reproduced a half-year WordPress project in an hour and built a **custom CRM** on Supabase for Slush. An illustrative bug: a security button removed the owner's own database update rights **without telling the code**, and neither twelve prompts nor ChatGPT found the cause — **the security feature and the coding agent do not talk to each other**. **The future of SaaS:** Translink sells recurring-revenue software companies (ARR, low churn, gross margin above 90%, **Rule of 40**); a median SaaS is a database plus an interface, so the threat is a **generative user interface** and **MCP** bypassing the UX layer — illustrated with a Tolkien analogy in which humans are hobbits and the AI layer the Valar, with the **One Ring** as the most trusted AI interface. The second threat is **uberisation**: transactional token billing instead of recurring revenue (**MaaS Global's** failed flat rate as the counter-example). Translink's SaaS reports put **vertical SaaS at roughly 5x ARR and horizontal below 3x**. **Capital:** Finland's largest SaaS is **AlphaSense** (~$500M ARR, operating from the US); **Solidium** is "close to a joke"; the **€270bn pension sector** invests about 60% in America; total wealth ~€700bn, mostly in homes; €540bn of revenue would require ~€154bn of capital, **two-thirds of it currently foreign**; the equity savings account passed one billion only recently. Valuation gaps: €1M in Finland, €10M in Sweden, €100M in the US; wealth per adult €150k (Finland), €300k (Sweden), €502k (US). **The proposal:** following Sweden, **10% of the pension contribution directed and owned by the individual**. **Three regulatory models:** Chinese totalitarianism, American deregulation, Europe's "bottle-cap philosophy" (data protection 2.0) — Miettinen sceptical because the capital is absent, Mikkonen defending a European moment grounded in individual liberties for 550 million citizens. Open source: **Torvalds**, Linux and Git as protocol-level public goods with the productisation profits going to GitHub; **Peter Thiel's** monopoly thesis; **Mårten Mickos's** MySQL story through Sun to Oracle. **Work and capital:** **Iain M. Banks's** Culture universe as the fifty-year picture, a humanoid robot's lifetime cost around €30k and labour at 40 cents an hour, with the mechanisation of agriculture as the reference. **Interfaces:** API codes requested from S&P Capital IQ (the first such request worldwide), Asiakastieto's €3,000 IT project, the USB analogy for MCP, the `.gitignore` reminder, and agent payment rails from Google, Visa and Mastercard. **Negotiation:** *Uusi neuvotteluvalta* (2019, ~30,000 copies, with **Juhana Torkki**), interviews with **Risto Linturi** and the author of *Rauhankone*; current client projects modelling **negotiation scripts**; the longer vision of **negotiation of meaning** in which machines converge on a deal from fundamental interests; and the **beta / alpha / scribe** role division being automatable. The key distinction: human trust accrues as reputation over time, whereas **an AI's reputation is a function of its current code**, which can be refactored overnight. **Three actions:** (1) build a demo with a no-code tool instead of a PowerPoint, (2) fix the shortage of investment capital, (3) build your own support AI with your own documents as context.