Chapter 21
A Development Plan
The Utopia of the Agentic Enterprise, Part IV. Doing Less
Rory Sutherland sums up a pattern he saw in digital advertising in four steps. Define the function narrowly, install technology that optimises that narrow function, and declare success by metrics drawn from your own definition. Then “capture cost savings for yourself and walk away.”1 An agent rollout can follow it to the letter. One that stripped the work properly still ends at the same last step, because a saving goes to whoever is placed to take it unless somebody decides otherwise.
For the people doing the work, unemployment isn’t the only risk. The other is devaluation, where people keep working in roles that pay less, because producing the deliverable is cheap and the surplus goes to the employer and the platform. The employment statistics stay stable, so nobody has to call a press conference, while the pay and standing attached to the work decline. That is harder to see than unemployment and harder to organise against, because the people affected still have jobs.
Nobody has to plan it. An organisation that gets more efficient doesn’t automatically get more profitable, because its competitors and clients are getting more efficient on the same tools. Early adopters see a few quarters of margin above the market, and a slide about it at the investor day. Then competitors catch up and clients treat the new speed as the baseline, and by then the lower prices are in next year’s framework agreement.
B2B Pricing Collapse
A B2B service priced on bundled hours of analysis and production loses its price once the client can generate a competent first draft internally and benchmark the seller’s output against it. Procurement gets forensic. The question moves from “how many hours did this take?” towards “what decision did we make better because you were involved?” How fast prices fall depends on how easily the buyer can inspect the output. Translation went first because the output is directly comparable.2 Strategic advice may hold longest, because assessing it takes the same expertise as producing it.
When the workflow runs on a platform, the platform takes a share of the gain through usage fees. And what buyers will pay and what the work means to the person doing it don’t move together. A role can become more meaningful and worse paid.
Collective Response
Some of the machinery already exists, in unglamorous places. German works councils have long had co-determination rights over technical systems that can monitor employees, which in practice gives them a say over most workplace software.3 A 2021 amendment went further: where a works council has to assess the introduction of AI, bringing in an outside expert at the employer’s expense is deemed necessary by law.4 The works council no longer has to argue that it needs one; only the choice of expert and the fee are agreed with the employer. A group of employees gets the right to understand the system before it changes their work, and the party deploying it pays. That is one more report before go-live, but for once the paperwork protects the people the system is done to.
No individual employee could have secured that on their own. Even where works councils and sector agreements are strong, they were designed for wages and working conditions and have only begun to deal with how an AI productivity surplus gets split.
So who gets the surplus depends on the channel it travels through. Left alone, it runs downhill to whoever owns the systems and the bargaining power, not to the people who produced it.
Inside the company, the surplus question gets personal at the pay review. When economists in Denmark linked chatbot use to pay records, only about 3–7% of the time workers said they saved showed up as higher earnings.5 Produce far more with an agent’s help, ask for a raise, and you leave with a development plan.
The hardest part of the surplus negotiation is the admission. If one augmented person does the work of five, then four positions existed purely because the technology hadn’t yet made them redundant. Nobody says that at a town hall, least of all whoever approved the four positions.
Further up are the platform companies that control the models and the compute. The efficiency gain distributes across millions of organisations. The infrastructure rent concentrates in a few.
No single organisation decides any of this, and most are in the middle of the chain, buying AI capability and selling AI-augmented services to clients who bought the same capability. Individual repositioning decides where you stand when the allocation is made. It doesn’t decide the allocation.
For any claimed productivity gain, a few questions show where the surplus went. What became cheaper? Who gets the immediate benefit, and who loses income or standing? Who pays for failure and repair?
Keynes’s Grandchildren
Keynes’s 15-hour week never arrived. Graeber’s explanation fits the organisations now buying agents: a hierarchy that gets a productivity windfall converts it into more hierarchy. A 15-hour week would have required somebody with authority to decide that the surplus should go to the people doing the work, as time, and nobody with that authority had a reason to, since a manager’s standing grows with the team, not its free time.
Whether that changes this time doesn’t depend on the technology. It depends on the places inside an organisation where the surplus is allocated: the pay band and the headcount decision. Each is a decision someone makes, with a constituency that benefits from the default, and a point where a works council or a manager willing to spend standing on it can attach.
Checklist
- Which of your offerings are priced on bundled hours of production, and which on accountability for the final answer?
- How inspectable is your output? Could a client draft it in-house, benchmark yours against it, and unbundle what you sell?
- If a client asked what decision they made better because you were involved, could you answer?
- Which channel will distribute the surplus from your productivity: pricing, pay, headcount, bargaining? Who decides?
- Is your edge a window of a few quarters?
- For the last work you stopped or handed to an agent, who decided where the saving went, and was it written down?