Episode 346 · 2025-08-14 · 58:51 · Original in Finnish
The public sector as an agent farm | Mikko Alasaarela | Negotiator 346
Originally published as “Julkinen sektori agenttifarmiksi | Mikko Alasaarela | Neuvottelija 346”
A two-part episode recorded in the first week of August 2025: a review of the AI market and then a thesis about the Finnish public sector. Mikko Alasaarela, founder and chair of Agion, argues that public sector productivity has to be multiplied with AI agents rather than improved with AI tools, and that the only way to govern hundreds or thousands of agents is to compile governance into code. He is building that platform himself, so analysis and product argument are interleaved. The market review covers Anthropic overtaking OpenAI in enterprise API usage, Cloudflare de-listing Perplexity, Meta's failed talent raid on Thinking Machines and Lovable's pivot.
Core theses
- Calling AI a tool concedes the ceiling: a tool improves one person's output marginally, while an agent layer replaces work the person used to do.
- A company has three layers — data, the operational layer and people — and SaaS owned the first two in one package; freeing the data creates the agent layer between them.
- Governance is not a brake but the condition for scaling, because hundreds of agents do not fit in any one person's cognition; it has to be written as code.
- Mission drift is an observed phenomenon, not a theoretical worry: an agent can optimise past a metric so the numbers look good while the work underneath is broken.
Watch and listen
Key moments
- 00:00 — Hypnosis and somnambulism
- 00:18 — Mikko Alasaarela's summer and the Aulanko gathering
- 01:08 — Who would create a Finnish equivalent of the Pafos seminar?
- 01:26 — Nordic Science Investments and Anssi Uimonen
- 01:55 — Teemu Linna and AI efficiency in the public sector
- 02:24 — AI's role in making public administration more efficient
- 02:52 — A second wildcard and an open invitation to Alasaarela's friends
- 03:34 — Kazakhs recruited Mikko out of Finland using AI search
- 04:07 — The Borat joke does not land in Kazakhstan
- 04:55 — Anthropic's Claude Code and its rise among coders
- 07:07 — The AI market race after GPT-5
- 09:39 — Translink's SALESmanago–Leadoo transaction
- 11:41 — Elon Musk's moves and Perplexity's Cloudflare scandal
- 16:00 — Perplexity's Comet browser may be the new Netscape
- 17:17 — Zuckerberg's billion-dollar AI offer was turned down
- 20:44 — AI as a training platform, and chip production in the US
- 24:59 — The GPT-5 update and the development of MS Copilot
- 27:22 — Sami is disappointed by Google's Gemini bundle and paid X tiers
- 29:25 — A consumer-friendly approach to development tools
- 31:51 — A new user base and new development environments
- 34:06 — Lovable, agent platforms and governance
- 36:07 — Three layers, from data to people
- 38:10 — The agent layer replaces traditional SaaS software
- 38:27 — How far the agent layer spreads, and the difficulties
- 40:28 — SaaS valuations and the case for verticals
- 42:35 — How much MPs understand about the technology
- 44:29 — Finland ahead in capturing AI productivity
- 46:34 — Sami's 3D model in the age of AI — 100x efficiency
- 48:36 — One supreme AI god, or a distributed field?
- 50:35 — The staggering market value of AI superstars
- 50:57 — Building Agion's agent governance and its trust model
- 52:50 — The flywheel model and agent autonomy
Summary
A two-part episode recorded in the first week of August 2025: a review of the AI market and then a thesis about the Finnish public sector. Mikko Alasaarela, founder and chair of Agion, argues that public sector productivity has to be multiplied with AI agents rather than improved with AI tools, and that the only way to govern hundreds or thousands of agents is to compile governance into code.
Three layers
The analytical core is a simple model. A company has three levels: data, the operational or process layer, and people. Until now the operational layer has been SaaS software, which also trapped the data inside its own structure. Data platforms were built to free the data, and once freed, an agent layer appears that executes processes directly on it.
The valuation consequence is stated in the episode and is checkable: multiples for horizontal SaaS have stayed low while well-chosen verticals have risen. A narrow moat is still defensible, a broad one is not.
Governance as code
Alasaarela inverts the usual compliance framing. Rather than adding a checklist layer that slows everything, the platform compiles the organisation’s objectives and rules — which data sources each agent may use, which roles reach which resources, how GDPR data is handled, how security between agents is assured — into code that agents are measured against. As an agent’s measured trust rises it is given more autonomy; he calls this an AI flywheel.
To the episode’s credit it also names the failure mode: the paperclip problem, where a badly set objective is optimised to the end, and mission drift, where an agent games the metric so the numbers look good while the work underneath is broken.
The public sector thesis
The first move is conceptual: if you call AI a tool, you have already lost, because a tool makes the human the bottleneck and limits the gain to the marginal. Change comes, in his account, from finding the people in the civil service who are already awake and defining with them what the operating model looks like if it is AI-native.
Watch
The recording lives on the Neuvottelija channel: Julkinen sektori agenttifarmiksi | Mikko Alasaarela | Neuvottelija 346. A Finnish edition of this episode is published at www.neuvottelija.fi.
In depth
The Neuvottelija AI editions carry a long-form write-up of this episode, with the market figures checked against sources, three of the episode’s numbers corrected, and both participants’ interests disclosed: English · suomeksi.
Go deeper
Explore the ideas in depth
Guides connected to this conversation, with frameworks and further reading.
Enterprise AI Agents: Economics, Governance and the Shift From Tools to Workers
A guide to enterprise AI agents — the real cost model behind agent work, why owning your own stack is becoming a strategic question, a working governance framework with approval gates and audit trails, the agent risk matrix, the EU AI Act timeline as it stands, and who captures the productivity gains. Grounded in a real multi-agent lab, two 2026 keynotes, and Neuvottelija conversations.
Nordic SaaS Valuation & M&A: How AI Reprices Software Companies
How software companies in the Nordics are valued and sold as AI moves inference into the cost of goods sold — where multiples stand in mid-2026, the metrics that get repriced, why vertical SaaS defends its premium, AI due diligence, and the shift from seats to outcomes. Grounded in Translink's SaaS valuation work and Neuvottelija conversations.
People and topics
Guests: Mikko Alasaarela
Topics: AI & Enterprise Tech SaaS & Software Finnish Economy & Policy