Episode 402 · 2026-08-21 · 01:07:11 · Original in Finnish
When There Are More Agents Than Employees | Tapio Nissilä | Negotiator 402
Originally published as “Kun agentteja on enemmän kuin työntekijöitä | Tapio Nissilä | Neuvottelija 402”
Tapio Nissilä builds organisation design around what he calls HAR thinking — the human-to-agent ratio — and the episode works out what changes when a company has more agents than employees. The load-bearing argument is that 'human in the loop' is a naive answer, because people are poor at supervision: work has to be designed to be meaningful for a human rather than left as whatever the machine could not do. The conversation runs from Jevons's paradox and freed capacity to decision compression and fatigue, from attachment theory as a better model of resistance to change than stubbornness, to span of control against span of agency. It closes on accountability, audit trails, regulation written without understanding, and the professional identity crisis.
Core theses
- 'Human in the loop' fails because humans are bad at supervision — the loop has to be designed around what a person is actually good at, not around what is left over.
- Tasks disappear and occupations rarely do, which is the distinction most automation forecasting collapses.
- Span of control and span of agency are different limits, and an organisation that manages the first while ignoring the second will break silently.
- Freed capacity does not stay free: Jevons's paradox applies to work as readily as to fuel.
Watch and listen
Key moments
- 00:00 — What HAR thinking means in practice
- 01:34 — Tapio Nissilä's background, and Lean Sales
- 04:24 — The relationship between a human and an agent
- 05:00 — How many agents one person can lead
- 05:24 — The research background, from neuroscience to systems thinking
- 05:54 — A book as a method of organisation design
- 06:36 — Human in the loop as a naive solution
- 07:17 — People are poor at supervisory tasks
- 08:03 — Work is designed to be meaningful for a human
- 08:43 — Waste, or people you can prompt
- 09:31 — Tasks disappear; occupations rarely do
- 11:09 — Optimist against realist in the agent matrix
- 12:03 — Freed capacity and Jevons's paradox
- 12:55 — How a sales team's composition changes in practice
- 13:37 — Missing routine, and cognitive load
- 14:48 — The brain needs routine work to recover
- 15:37 — Decision compression and fatigue
- 16:11 — Why a finished tool does not transfer to someone else
- 17:45 — Standardisation, documentation and word of mouth
- 18:29 — Human scraping, and the fear of being replaced
- 19:14 — Attachment theory instead of resistance to change
- 20:50 — Anthropomorphism, and how trust is built
- 22:20 — The second brain, the consigliere and the memory library
- 23:36 — Channel dependency in a network of agents
- 25:24 — Outsourcing memory, and reduced stress
- 26:39 — Two ways of approaching agency
- 27:12 — Tapio in the loop, out of the loop
- 29:00 — Small steps for large organisations
- 29:37 — Token maxing, and maximising capacity
- 31:15 — The lean school wants slack in the utilisation rate
- 32:52 — How HAR would be built into a sales funnel
- 33:55 — A deterministic process around a language model
- 35:17 — Expansion revenue, and recognising churn
- 36:17 — The SaaS apocalypse, and per-user pricing
- 37:41 — Are the users agents or people
- 38:43 — Agent compatibility as a selection criterion
- 39:08 — The system of record lasts; the layer on top does not
- 40:14 — Abusing the freemium model, and inference costs
- 42:38 — Why it is not worth coding your own software
- 44:39 — The human is the anchor of accountability
- 45:28 — Do boards know about their agents
- 47:45 — The nature of supervision as the model's third dimension
- 47:59 — Cynefin, and recognising complex work
- 49:18 — Removing the admin collapses the system
- 50:34 — The SaaS utopia, and the irony of the model's death
- 51:29 — Span of control and span of agency
- 52:53 — Whether an agent can be tagged in its own channel
- 54:23 — Who has the power to silence an agent
- 55:22 — Independent agency, and its risks
- 56:06 — Autonomy is a design question
- 57:17 — An agent's identity, and the audit trail
- 58:03 — Data protection, servers and registers
- 58:41 — Regulation written without understanding the subject
- 59:36 — Chat control and fundamental rights
- 1:00:47 — The cybersecurity view of large data holdings
- 1:01:37 — Open source as a defence
- 1:02:29 — Isolated pipes, and social engineering
- 1:03:59 — The professional identity crisis
- 1:05:11 — Doctors, and the breakdown of authority
- 1:06:01 — From coding to coding the work itself
- 1:06:40 — Closing, and continuing in Sisäpiiri
Summary
Tapio Nissilä builds organisation design around what he calls HAR thinking — the human-to-agent ratio — and the episode works out what changes when a company has more agents than employees. The load-bearing argument is that ‘human in the loop’ is a naive answer, because people are poor at supervision: work has to be designed to be meaningful for a human rather than left as whatever the machine could not do. The conversation runs from Jevons’s paradox and freed capacity to decision compression and fatigue, from attachment theory as a better model of resistance to change than stubbornness, to span of control against span of agency. It closes on accountability, audit trails, regulation written without understanding, and the professional identity crisis.
Why the loop does not work
‘Human in the loop’ is the standard reassurance about autonomous systems, and Nissilä’s objection is empirical rather than philosophical: people are demonstrably poor at watching a process that is usually correct. Designing the human role as supervision therefore produces the appearance of control and not the substance of it.
Tasks, not occupations
The distinction that most automation forecasting collapses. A job is a bundle of tasks; removing some of them changes the bundle without removing the job, and the transition looks entirely different depending on which unit you count.
Two spans, not one
Span of control — how many people a manager can lead — has a century of practice behind it. Span of agency is the same question about agents, and it has a different answer, because an agent does not notice when it is wrong. An organisation that tracks the first and not the second will find out late.
Watch
The recording lives on the Neuvottelija channel: Kun agentteja on enemmän kuin työntekijöitä | Tapio Nissilä | Neuvottelija 402. 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: 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: Tapio Nissilä
Topics: AI & Enterprise Tech Leadership & Governance SaaS & Software