From One Hire to a Whole Economy: How AI Agents Transact, Form Markets, and What It Means for People
In my Agentic Economy framework, software hires AI agents one task at a time and pays for the result. So far I have written about the parts. This article zooms out in three steps: one hire, a market of hires, and a whole economy built on them. At each step I ask the same question: what happens to the people?
In my Agentic Economy framework, software hires AI agents one task at a time. This article follows that idea from one hire, to a market of hires, to a whole economy, and asks at each step what happens to the people affected and what questions the law raises.
- One hire is a full transaction. Find, hire, measure, release. The manifest is the offer, the hire-token is the contract, and the signed receipt is both payment and proof.
- Many hires make a market. Registries become marketplaces, track record becomes the currency of trust, and capabilities cross industries in chains.
- A whole economy changes shape. Technician work is hired from capabilities; engineer work, the judgement about what to hire and who answers for it, becomes more valuable.
- People keep their rights at every level. The customer, the staff and the workers affected still need to be told, to be shown why, and to reach a person who can change the result.
Bottom line. Agents do the hiring. People still live with the result, so every hire has to leave a record a person can be shown.
Below: the full article, for technical teams ↓"Agents do the hiring. People still live with the result."
Start here
If the Agentic Economy is new to you, read the five-minute explainer first. It covers the idea and the six terms I use below: capability, manifest, hire-token, receipt, registry and OLA. If you prefer video, watch Hire the AI Agent: SaaC, the End of SaaS.
The short version: software used to be sold by the seat. In the Agentic Economy, a software application, or an AI agent working inside it, hires a piece of work for one task. That piece of work is a capability. The application uses it, gets a signed record of the result (a receipt), and moves on. Think of a staffing agency, not a software licence.
How the Agentic Economy works: one agent, hired again and again
- One agent is listed once in the registry: a timesheet check, with a manifest saying what it does, what it never does and its price.
- It is hired again and again. A builder checks a subcontractor’s site hours, an engineering firm checks a contractor’s invoice, a staffing agency checks timesheets before billing. Three industries, three separate hires.
- Each hire is its own contract, with its own purpose and limits, and leaves its own signed receipt. Each is paid for per job.
- Every accepted job adds to one track record, and the track record decides who hires it next. That is how single hires become an economy.
Built once, listed once, hired by anyone who needs that job done, each time under its own contract and with its own receipt.
Why I wrote this
After the EU AI Act article, the next questions fall naturally into three groups.
- How does one hire actually work? Who agrees to what, who pays, and what happens when it goes wrong.
- What happens when there are millions of hires? How a market forms, who sets the prices, and who wins.
- What does this do to people? The person on the other end of a decision, the staff inside the companies, and the people whose jobs change.
Economists often use three levels, micro, meso and macro: one transaction, one ecosystem, one economy. I use the same three levels here. Regulation does not get its own chapter. It runs through every level, because it shows up differently at each one.
Two things before we start. I am a technical architect, not a lawyer or an economist. Where I touch the law, I describe duties, not dates, and I point to the EU AI Act article for the detail rather than repeating it. Where I look ahead, I say what I think, and where my research goes next.
The three levels, at a glance
From one hire to a whole economy: zoom out
Legend
- A company’s software: it hires
- An AI agent: it is hired
- A signed receipt for every job
- The registry: listed, found, rated
- Hires, or returns the result
- Listed in, and rated by, the registry
- Zoom out: where it sits in the bigger picture
- Who is affected at that level
Start at the top with one job. Each dashed cone zooms out: the hire becomes one of many in a market, and the market becomes one of many across the economy.
Level 1, micro: one hire
1Micro level at a glance
- What happens
- One piece of software hires an AI agent for one job, gets the result with a signed record, and lets it go.
- Who is affected
- The person behind the paperwork, such as the contractor whose invoice it is.
- EU AI Act and related rules
- Who is responsible? Does the check stay a check? Can the outcome be explained?
Find, hire, work, release: the same four steps every time.
Micro level: how one hire works
Take one example. ExampleCo is an engineering firm that works with contractors. Before its finance software pays a contractor's invoice, the invoice needs checking against what was agreed, so an agent inside that software hires a timesheet check supplied by another company. The check compares each line of the invoice with the approved timesheets, the contract rate and the purchase order. The whole hire has four steps.
- Find. The agent looks in a registry and reads the capability's manifest: what it does, what it will never do, what it costs, how well it has performed, and who stands behind it.
- Hire. It issues a hire-token: the contract for this one job. Who is hiring, for what purpose, under which rules, in which country.
- Measure. The capability returns a result, either "matches" or a list of differences, and a signed receipt: what went in, what came out, which rules applied, which versions ran. The work is checked against the service level the manifest promised.
- Release. The hire ends. Nothing is owed beyond this one task. The same agent can be hired again a minute later, by ExampleCo or by someone else entirely; if the next timesheet check goes to a different supplier, that is the next hire.
Zoom 1: one hire, in four steps
- 1FindReads the agent’s CV in the registry: what it does, never does, costs.
manifest - 2HireAgrees one job, with the purpose and the limits written down.
hire-token - 3WorkDoes the job and sends back the answer with a signed record.
receipt - 4ReleaseThe hire ends. Nothing more is owed; the record is kept.
release
The same four steps every time, whoever is hiring and whatever the job.
I call this loop hire, measure, release. It is the same loop a company runs with a contractor, but it runs in less than a second and it runs for every task.
Three things make the transaction trustworthy without a person in the middle:
- The manifest is the offer. The terms are written down before the hire, in a form software can read.
- The hire-token is the contract. Purpose and limits travel with the job, so the supplier knows what it was hired for.
- The receipt is the settlement. Payment, proof and audit trail are the same signed record. Both sides keep a copy. Nobody can quietly change it.
Micro level: the person on the other end
The timesheet check is about paperwork, but there is a person behind it: the contractor who sent the invoice. They never saw the hire. They do not know a second company was involved. What they experience is an invoice paid on time, queried or held up.
This is where the Agentic Economy can be better or worse than what came before.
- Worse, if the hire is invisible. A query assembled from three suppliers' checks, none of them named, is harder to answer than one raised by a single company.
- Better, if the receipt exists. A signed record of exactly what was checked, by whom, under which rules, is more than most contractors can get today when they ask why a payment was held up.
The design choice is simple to state: every hire that affects a person should leave a record that a person can be shown, in words they understand, and a route to a human who can change the outcome.
Micro level: impact of the EU AI Act
At the level of one transaction, three questions come up, and the hire should be able to answer all of them.
Who is responsible? Under the EU AI Act, as I read it, the company that supplies the AI (the provider) and the company that uses it (the deployer) carry separate duties, and those duties are heaviest where the use is on the Act's high-risk list. Automation does not remove either role. ExampleCo still decides whether to pay; the timesheet-check supplier is still responsible for what it supplies. The hire-token is where that split can be recorded.
Does the check stay a check? The Act sorts AI by what it is used for, not by what it is. Its high-risk list names categories of use. In the world of work, they include recruitment, decisions about work contracts and promotion, allocating tasks, and monitoring or evaluating the performance and behaviour of workers. The Act also treats a system on that list that profiles people as high-risk in every case. A check that compares an invoice with a contract is designed to stay outside those categories: it looks at documents, not at people, builds no profile of anyone, never rates the contractor, and leaves every decision to a person. Whether a given use stays outside is a question for legal advisers; the limits in the manifest are where the design makes its case. Data protection law applies either way, because timesheets name people.
Can the outcome be explained? When a payment is held up, the contractor will ask why. European law already gives people rights when decisions with serious effects are made about them by machines alone, and for some high-risk uses the AI Act adds human oversight and a right to an explanation. Here a person decides, but the same principle holds: the receipt is the raw material for the explanation. A log line is not.
The detail of how each rule maps onto a hire is in the EU AI Act article, which works through a different example, an insurer hiring a fraud check. The point here is that, at the micro level, regulation is not a separate system. It is three fields in the transaction: who, why, and what happened.
Level 2, meso: a market
2Meso level at a glance
- What happens
- Hires add up to a market. Agents are listed in registries, hirers choose on track record, and agents hire other agents across companies and industries.
- Who is affected
- Staff in the companies that hire, supply and check the work.
- EU AI Act and related rules
- Who carries the duties along a chain? May the data travel? Is the market fair?
List, compare, chain, rate: when single hires become a market.
Meso level: from hires to a market
One hire is a transaction. A million hires a day is a market. Three things appear that were not there before.
Registries become marketplaces. A registry starts as a list of capabilities. Once hires flow through it, it becomes the place where suppliers compete. Agents compare manifests, prices and, above all, track records.
Track record becomes the currency of trust. An agent cannot take a sales call. It chooses on evidence: how a capability has performed, measured receipt by receipt, over time. A supplier with a long, signed record of meeting its promises wins hires that a newer, cheaper rival does not. I wrote about this in The Four Moats.
Capabilities cross industries. A timesheet check built for engineering contractors can be hired by a builder or an IT consultancy. A route forecast built for weather can be hired by a logistics company. A document check built for one sector can serve ten. Software stops being sold within an industry and starts being hired across them. The supplier who serves many industries sees more cases, gets better, and gets hired more.
The same agent is hired again and again. Take the timesheet check from ExampleCo's invoice. In the same week it might be hired by ExampleCo for a contractor's invoice, by a builder to check a subcontractor's site hours before paying, and by a staffing agency to check timesheets before it bills its own clients. One agent, one manifest, three hirers. Each hire is separate: its own purpose, its own limits, its own receipt, its own payment. Nothing from the builder's job leaks into the agency's. What they share is the track record: every accepted job, whoever it was for, adds to the evidence the next hirer reads.
Zoom 2: a market, in four steps
- 1ListSuppliers publish each agent’s manifest in a registry.
registry - 2CompareHirers choose on track record, not on a sales call.
track record - 3ChainOne agent hires another, across companies and industries.
chained receipts - 4RateEvery accepted job adds to the agent’s record.
receipts
Many hires through one registry: that is when single transactions become a market.
Meso level: chains of hires
Markets also create chains. ExampleCo's finance agent hires a timesheet check. The timesheet check, to do its job, hires a rate-card lookup from the staffing supplier, which hires a VAT-number check from a fourth company. One contractor's invoice now touches four companies in under a second.
Chains are where the Agentic Economy is most useful and most fragile.
- Useful, because each company does the one thing it is good at, and nobody has to build everything.
- Fragile, because a fault at the end of the chain shows up at the start. If the VAT-number check is wrong, the contractor's payment is held up, and the contractor only knows about ExampleCo.
Two rules keep chains safe, and both are design rules, not legal ones:
- Purpose travels with the job. Each hire in the chain carries the original purpose and limits. A supplier cannot use the contractor's data for anything beyond the purpose ExampleCo stated.
- Receipts chain together. Each receipt points to the one before it, so the whole chain can be reconstructed, link by link, when someone asks what happened.
Meso level: the people inside the companies
At this level, the people most affected are not the contractors or customers on the other end. They are the staff of the companies that hire and supply.
Procurement starts to look like staffing. Buying software used to mean a tender, a contract and a renewal date. Now someone has to decide which capabilities the company's agents may hire, under what limits, from which suppliers, and when to release them. That is closer to running a bench of contractors than signing a licence.
New jobs appear inside familiar teams. Someone writes and maintains the manifests for the capabilities the company supplies. Someone reviews receipts and decides when a supplier is falling short. Someone owns the limits: what the company's agents may never hire, and what its own capabilities will never do. These are not exotic roles. They are product owners, risk managers and architects, doing their existing jobs in a new place.
Finance sees work, not seats. The cost line stops being "120 licences" and becomes "this many timesheet checks, at this price, with this record". That is a better conversation for a finance team, and a harder one for a supplier who was selling seats nobody used.
Meso level: impact of the EU AI Act and related rules
At the meso level, the Act and the laws around it ask different questions.
Who is liable along the chain? When four companies contribute to one decision, the question of who pays when it goes wrong gets harder. Europe's revised Product Liability Directive, adopted in 2024, brings software, including AI, within the meaning of a product. The AI Act sets out when a company that rebrands, substantially changes or repurposes a high-risk AI system takes on the provider's duties. As far as I can tell, neither rule was written with chains of agents in mind. The chained receipt helps a court, a regulator or an insurer find where the fault actually was.
Is the data allowed to travel? Every hire that crosses a company, a sector or a border carries data with it. Data protection law applies to the personal data, whichever software handles it. That is why purpose and jurisdiction have to be fields in the hire-token, checked at the moment of hire, not settled once in a contract and forgotten.
Is the market fair? When agents choose suppliers on track record, the leaders can pull away fast. When pricing is set by software watching other software, prices can move together without anyone agreeing to it. Competition authorities have already raised concerns about pricing algorithms in other markets. I expect the same questions to reach agent registries: who runs them, how ranking works, and whether a supplier can be shut out.
Level 3, macro: a whole economy
3Macro level at a glance
- What happens
- Capabilities become a unit of trade, receipts the unit of settlement, and a network of registries the market, across every industry.
- Who is affected
- Workers and professions: routine work is hired, and judgement becomes more valuable.
- EU AI Act and related rules
- Are workers told? Do people keep their rights? What happens if a widely hired capability fails?
What is traded, how it is settled, where it is traded, and what changes for people.
Macro level: what the economy trades in
Zoom out once more. If enough work is hired this way, the economy itself starts to change shape. Three things shift.
- The unit of trade becomes the capability. Not the licence, not the seat, not the project. One well-described piece of work, hired for one task.
- The unit of settlement becomes the receipt. Payment and proof travel together. A company's record of receipts is its record of work bought and sold.
- The market becomes a network of registries. No single marketplace, but many, each with its own rules, connected so that agents can hire across them.
There is a useful comparison in the shipping container. It did not invent trade. It standardised the box, and once the box was standard, ports, ships, trucks and whole supply chains reorganised around it. A described, hireable, receipted capability is a standard box for work. The question is what reorganises around it.
Zoom 3: an economy, in four steps
- 1What is tradedCapabilities: one well-described piece of work, hired by the task.
capability - 2How it is settledPayment and proof travel together, one receipt per job.
receipt - 3Where it is tradedA network of registries across every industry.
registries - 4What changes for peopleRoutine work is hired; judgement becomes more valuable.
engineers and technicians
The unit of trade, the unit of settlement and the market: the same three pieces, at the scale of an economy.
Macro level: what happens to jobs
The question I am asked most is whether AI agents will replace people. I covered it in the episode What AI Will Replace: Myth vs Reality. The short answer is that the question mixes up two kinds of work.
- Technician work follows a known procedure: turn a clear requirement into code, write the tests for stated criteria, apply a standard fix, run the runbook. This is the kind of work that becomes a capability most easily, because it can be described, measured and checked.
- Engineer work decides which procedure applies: where one system ends and the next begins, what to measure, how things fail, which trade-off to make, and who takes responsibility when it goes wrong. This is much harder to package, because it depends on judgement in situations nobody has written down.
The distinction is not seniority. Some very senior people do mostly technician work; some juniors are already making real decisions. It is about the nature of the work, and it is not limited to software. Claims handling, document review, scheduling and parts of accounting all contain both kinds.
In an economy where capabilities are hired by the task, I expect technician work to be hired more and more from capabilities, and engineer work to become more valuable, because someone has to decide what to hire, set the limits, read the receipts and own the result. That is my reading of the direction, not a forecast with numbers attached.
There is a hard problem underneath this. Most experienced engineers learned their judgement by doing years of technician work first. If that work moves to capabilities, the training path that produced today's seniors gets thinner. Companies that still hire juniors, and teach them judgement through review, real decisions and broad exposure, will have the experienced people others lack. Companies that stop hiring juniors are borrowing from their future.
Macro level: trust at the scale of an economy
At the macro level, trust stops being a question between two companies and becomes a property of the whole system.
Evidence becomes infrastructure. When every hire leaves a receipt, the economy keeps a record of its own work. Auditors, regulators and courts can check what happened rather than reconstruct it. That only works if receipts follow shared standards, in the way invoices and accounts do today.
Concentration becomes a risk. If many companies' agents hire the same few capabilities, a fault in one of them spreads everywhere at once. European financial rules, under the Digital Operational Resilience Act, already require banks and other financial firms to manage the risk that comes from outside technology suppliers. I expect the same thinking to reach any capability that a large part of an economy depends on.
People need a way back in. An economy where decisions are assembled from many hires must still give every person a place to ask why, and a person who can change the answer. If that is missing, trust in the whole system falls, however good each capability is.
Macro level: impact of the EU AI Act
At the macro level, the Act moves from transactions to society.
Workers have to be told. As I read the EU AI Act, an employer that uses a high-risk AI system at work has to inform the workers affected, and their representatives, before it is put into use. When capabilities are hired into the workplace, that duty does not go away because the hire was made by software.
People keep their rights. Rights over automated decisions and over personal data belong to the person, not to the company or the agent, and for some high-risk uses the AI Act adds a right to an explanation. They apply however many hires sit between the person and the decision.
Rule-makers will watch the system, not just the parts. I expect attention to move from single AI systems to the chains and registries that connect them: who runs a registry, who can be shut out, what happens when a widely hired capability fails. I am deliberately not putting dates on any of this. The direction is clearer than the timetable.
Across all three levels: what it means for people
The three levels come back to the same people.
| Who | What changes | What to ask |
|---|---|---|
| The person a decision is about | Decisions are assembled from several hires they never see | Can I see what was decided, why, and reach a person who can change it? |
| Staff in companies that hire capabilities | Buying software looks like managing contractors | Who decides what our agents may hire, and who reads the receipts? |
| Staff in companies that supply capabilities | The product is a described, measured piece of work | Is our manifest honest, and is our track record something we can prove? |
| People doing technician work | Repeatable tasks are hired from capabilities | Which parts of my work need judgement, and how do I do more of those? |
| Leaders and boards | Cost lines turn into hires, and systems can earn revenue | Where are we dependent on a few suppliers, and what is our evidence worth? |
None of this asks people to compete with agents. It asks them to do what agents cannot: decide what is worth doing, set the limits, and take responsibility for the result.
Where my research goes next
These are the questions my next pieces of research will take on. Stay tuned.
- Who runs the registries. A registry that ranks suppliers holds real power. I will look at the models on the table, open, industry-run, regulated or owned by a platform, and what each one means for fairness.
- How long a chain can safely be. Every hire adds a link. I will work out where a chain becomes too long to explain to the person at the end of it, and how linked receipts keep it explainable.
- How prices behave. Per-task pricing between agents could make prices clearer, or make them move together in ways nobody intended. I will look at what keeps it the first and not the second.
- What replaces the junior apprenticeship. I have described the remedies I believe in. Next, I will test them against how teams actually grow judgement.
- How the law treats a chain of agents. The rules mostly picture a supplier and a user. I will follow how they are applied as hires multiply.
The small print
I am a technical architect, not a lawyer or an economist. This article is my reading of how the Agentic Economy works at three levels, and of the questions European rules raise for it, as they stood on 18 October 2026. It is not legal advice, and other readers of the law may reach different conclusions. I have deliberately not quoted deadlines, because they move. Check the duties and the current timetable against the official texts before relying on them, and take advice on your own situation. ExampleCo is an illustration, not a client, and the article contains no client figures.
Important notice. This article is part of my ongoing research into the Agentic Economy and reflects my thinking at the time of writing. It is shared in good faith, for information and discussion. It is not legal, regulatory, technical, financial or other professional advice, and it is not a verdict on how the EU AI Act or any other law applies to any organisation, system or situation. Laws, guidance, technology and my own research all move on, and others may reach different conclusions. The examples, designs, diagrams and technology choices show one possible approach. Before relying on, adapting or applying any part of it in your own situation, you must test and verify it, design it for your own architecture, and take independent expert advice, including from qualified lawyers and from your own IT, security and data protection specialists. Any decision to act on it, and its consequences, rest with you. To the fullest extent permitted by law, the author accepts no responsibility or liability for any loss, damage or consequence arising from any use of, or reliance on, this article, which is provided without any warranty of accuracy, completeness or fitness for a particular purpose. Reading it does not create a professional or advisory relationship with the author. The views expressed are the author’s own.
Coming next
This article stayed at the level of the idea. Part 5, a companion case study, puts it to work in a university, with a different agent: one that keeps the university's systems in step, hired again and again by admissions, student records, finance and a partner university. It also shows how it would run on AWS.
After that, the series builds an agent, step by step:
- Part 6: On AWS Bedrock, building a capability and its manifest.
- Part 7: On AWS Bedrock, the hire and the signed receipt.
- Part 8: On AWS Bedrock, measure, track record and release.
- Part 9: When an agent hires an agent. Chains of hires across organisations, with purpose and receipts carried all the way down.
- Part 10: Building the marketplace. Listing, finding and paying for agents at scale, and the rules a marketplace itself has to follow.
If you build AI agents and would like to try one against the registry as it takes shape, message me on LinkedIn.