The capability as an entity

Contents

1. 1. The claim · Proposal
2. 2. Where the idea comes from · Mark’s account
3. 3. What a capability is · Proposal
4. 4. No fixed bottom: capabilities made of capabilities · Proposal
5. 5. Two sides of one table · Proposal
6. 6. How capabilities bond: three terms and a trial · Proposal
7. 7. Who holds the data · Proposal
8. 8. Why now: the economics of the edge of the firm · Proposal
9. 9. Same shape as the twin · Proposal
10. 10. Where this could fail · Counterargument
11. 11. Who answers for it · Proposal
12. 12. What we will test · Hypothesis
13. References
14. How this paper was written
15. Conflict of interest

1. The claim · Proposal

Most organisations still describe their work the way they describe their people: as a chart of roles, each reporting to the one above. A job title says who someone is and where they sit. It says very little about what they actually produce.

This paper argues that the unit that matters is changing. We think work is breaking into smaller, more specialised units, each of which does one thing: it takes something in, transforms it, and hands something on. We call that unit a capability. A capability can be done by a person, by software, by an AI system, or by a mix. What defines it is not who does it but what goes in, what happens and what comes out.

Our position has four parts:

1. Work moves from roles to capabilities. The question stops being "whose job is this?" and becomes "what goes in, what comes out, and who or what does the transformation best?"
2. Capabilities become entities. They have an identity, a record and someone who answers for them. They advertise themselves, and they may serve one organisation, several, or join others to form something larger.
3. Capabilities include their tools. A capability is not only a person. It is the person, the tools they use (some of which they may have invented), what they need to start, what they deliver, how much and how fast, and at what price.
4. Organisations need to see themselves first. No organisation can safely rebuild itself out of capabilities without first seeing how its work really runs. That is the job of a digital twin, which is the subject of a later paper.

The author put the first part this way, in his notes for this programme (October 2026): work will change "from the normal hierarchical responsibilities to much more atomic elements that do a very specialist thing, with or without AI and have an input, a transformation and an output."

This is a proposal, not a finding. Sections 10 and 11 set out how it could be wrong and what we will measure.

2. Where the idea comes from · Mark’s account

The argument draws on the author’s 25 years of experience analysing and building systems for organisations. That is experience, not 25 years of research, and we say so deliberately. What follows is his own account, from a recorded interview (142 pages) and written answers, made during October 2026. It has been matched to the transcript text but not independently checked against other records.

Three things recur in that account.

First, organisations do not see themselves whole. "Organisations didn't have a full picture of themselves. The real workings of the business were spread across systems, processes, approvals, workarounds and people's heads, and nobody held all of it." (author’s notes, October 2026). Describing an early software project, he said the team "didn't map out properly the whole company so the software was built in isolation" (interview, p. 25).

Second, the parts connect in ways the chart does not show. "I never really realised the potential of breaking all those moving parts down and reconstructing them but I did start to realise that everything is interconnected and parts from one part of the business can be fed from other parts of the business" (p. 27). One shift system had two rotas: the planned one and the one the organiser actually used. "The second rota was the reality, shifts not covered, sickness, holiday etc" (p. 74).

Third, taking work apart is old; putting it back together is new. "I had been taking systems apart for years… but AI takes this to a completely new dimension, where before it was 100s of components/dimensions, now we have 100s of thousands" (p. 100). The turning point was not decomposition but the arrival of AI agents that could do pieces of work on demand: "not taking what was already there and just speeding up the slow parts, but completely rebuilding the whole process using re-usable workers, registering them and then using on demand" (p. 109).

It is from that point that the definition used in this paper comes. Each reusable piece "can be called on from anywhere can be used permanently, or used only once, we can determine if it's suitable for the job to be done by 3 simple pieces of information, what goes in, what comes out and how the input changes to the output" (p. 108). And on finding them: "it was essential that another worker could search for what it needed and then find it by a simple input - transformation - output" (p. 114).

We are careful about one thing. The interview shows that the author’s thinking changed over time. He did not hold this theory for 25 years; he says himself that, in his earlier work, the potential of reconstruction was something he "never really realised" (p. 27), and that the turn came with agentic AI (p. 109). We record the change rather than rewrite the history.

3. What a capability is · Proposal

A capability is a piece of work with a clear input, transformation and output, and someone who answers for it.

That definition has three parts that are borrowed and one that is ours. Input → transformation → output is the basic model of operations management, and we credit it rather than claim it (we state this as general knowledge; no textbook is cited yet). Describing an organisation by what it does rather than by how it is arranged is the idea behind business capability maps in enterprise architecture (The Open Group, TOGAF Business Capabilities guide). What we add is that the capability is treated as an entity: something with an identity and a record, which can be listed, found, tried, engaged, compared, replaced and taken back.

To describe one, we use a single card, the same for a person, a team, a piece of software or an AI agent. We have published a first draft of its format as an open standard, the Element card (see the Periodic Table of Work and the format). The vocabulary is borrowed from chemistry, partly because it fits and partly because it makes people smile.

Download this table (CSV), free to reuse with a credit.

Two fields deserve a word.

Catalysts. The author’s notes put tools at the centre: "in future we will all have our capabilities that will include our tools (perhaps some specialist tools we have invented), ourselves and our input/output and rates and output rates". A capability is not a CV. It is a person or a system with its kit. When the kit was invented by the capability itself, that matters for price, for lock-in and for what leaves when the engagement ends.

The record. Most adverts are written to sell. We propose that every claim on a card carries one of three labels: verified (checked against evidence), declared (stated by the capability, not yet checked) or projected (expected, not yet seen). And the card keeps its failures. A record that can only carry good news is not evidence; it is marketing.

The idea of a structured, honest card has a respected lineage that we follow rather than invent. Datasheets for Datasets (Gebru et al., 2018) proposed that every dataset should come with a sheet stating how it was made and when not to use it, borrowing from the datasheets that come with electronic components. Model Cards (Mitchell et al., 2019) did the same for machine-learning models. The Element card applies that idea to work.

4. No fixed bottom: capabilities made of capabilities · Proposal

How small is a capability? An earlier form of this argument, in the author’s own written answers, drew a line: a capability was the smallest unit "the business would notice if it stopped", and anything smaller was a task. Later that month the author changed that position, and we record the change.

His reasoning was an example. Scraping data for prospects looks like one task. But it is really several: "connect to the app, choose the page, copy the data, paste into a database, add some metadata. Each of those could also be broken down further (choose the page needs a list of criteria, a way to check if that criteria exists on the page and select the top choice out of 5) and each of those moving parts could be reused."

That gives four points.

1. There is no smallest size. Anything with a clear input, transformation and output can be a capability, at any level.
2. Capabilities are made of capabilities. A large one, such as "find prospects", is a chain of smaller ones, which are chains of smaller ones still. In the Periodic Table we call a chain a Molecule, and a whole organisation a Compound.
3. Different buyers work at different depths. A director buys "find prospects". A sales team buys "qualify these leads". An AI agent assembling a workflow buys "pick the best of these five pages". Each is a real buyer of a real capability.
4. Every piece can be reused. The step that picks the best page for prospect research could pick the best page for supplier research.

Here is the example as it now appears in the Periodic Table, at two depths. The second line is what "prospect research" looks like when you open it.

Download this table (CSV), free to reuse with a credit.

The examples in the table are illustrations we invented to show the format. They are not real capabilities or measured results.

This view has a precedent we should credit. The MIT Process Handbook (Malone, Crowston and colleagues) represented business processes at many levels, with specialisations and dependencies, so that people could find "sensible alternatives" for how a process might be done (Malone et al., 1999). In banking, BIAN partitions the business into non-overlapping "service domains" that exchange services with one another. Both show that work can be described at many depths. What they did not do is treat each piece as an entity that advertises itself and can be bought.

"No fixed bottom" is not the same as "split everything". There is an obvious cost to splitting: every split adds a hand-off, and every hand-off can fail. Where splitting stops paying is the question in section 10, and the first of our experiments.

5. Two sides of one table · Proposal

If capabilities are entities, there are two sides to a market in them.

On one side, capabilities offer. A person, a small team, a software service or an AI agent publishes its card: what it takes, what it makes, how much, how fast, at what rate, and its record.

On the other side, organisations ask. An organisation (or a larger capability assembling a chain) posts what it needs. In the Periodic Table this is a Formula: the product needed, the reactants it can supply, the volume and the timing.

The match is made on what goes in and what comes out. That is the important difference from most marketplaces today, which match on who someone is: job titles, skills, years of experience. When the author reviewed an early list of these ideas, his first question was about the input side: "and the input you put in??" A buyer needs to know not only what they will get, but what they must provide for it to work.

Two things already exist that look like parts of this.

• Freelance marketplaces list people and let buyers filter by engagement type and skill (a general observation; we cite no single marketplace). They list people, mostly, and describe them by skill, not by input and output.
• AI agent registries list software. The A2A protocol’s Agent Card is a small JSON file that states an agent’s name, version, skills and the input and output formats it accepts. The official MCP Registry is an open catalogue of tool servers. These list software and describe technical interfaces, not performance evidence or accountability.

What we propose is one card format for all of them: a person, a team, software and AI listed side by side, searched by input and output, with evidence and failures shown and a named person behind each. A buyer should be able to compare a human researcher and an AI research agent on the same terms, and choose a mix.

The author’s description of the use case is wide: people "will go to sites and be able to register their capability, organisations or other entities or whatever who need these capabilities will then be able to search on these sites for a specific capability, perhaps on a short term or once only basis or perhaps on a year contract that renews".

6. How capabilities bond: three terms and a trial · Proposal

The way a capability is engaged matters as much as what it does. Following the author’s three cases (once only, short term, a year that renews), the Periodic Table names three kinds of Bond:

• Spark: once only. One job, one product.
• Half-life: a set period. Weeks or months, then it ends unless renewed.
• Stable: a year, renewing.

Before any bond, a buyer should be able to test a capability cheaply. The author suggested two ways: "perhaps offer a trial for 10 transactions to see if it works or on a time basis". We call this Titration, after the chemist’s method of adding a little at a time until a reaction shows. The capability runs on a small number of real transactions, or for a short time, and its results are compared with the buyer’s current way of doing the work. The pass mark is agreed before the trial starts, so neither side can move it afterwards.

Buying work by its output rather than its hours is not new either. Outcome-based contracting, where a buyer pays for a result (hours flown, availability) rather than for inputs, has a long history in engineering services. Research on it has also found that pure outcome contracts are rarer than claimed: a review of reported cases found that nearly all pay for availability rather than a true result, and that what counts as an outcome is hard to define (Grubic and Jennions, 2018). We take that as a warning for section 10.

7. Who holds the data · Proposal

When a capability works for an organisation, it makes and uses data. Who owns it?

Today the usual answer is the organisation. The author raised another possibility: the data "always stays with the capability and it 'rents' to the company". That opens a range of models, which the Periodic Table now asks every card to state.

Download this table (CSV), free to reuse with a credit.

The stewarded model borrows from data trusts. The Open Data Institute says a data trust "provides independent, fiduciary stewardship of data" (ODI, 2020). That replaced its 2018 working definition, "a legal structure that provides independent stewardship of data", which the ODI said missed the fiduciary duty. Where any of these models involves personal data, UK data-protection law decides who controls and who processes it; we have not analysed that and will not state an answer without legal advice.

8. Why now: the economics of the edge of the firm · Proposal

If this is a good way to organise work, why has it not happened already?

The classic answer is Ronald Coase (1937). Firms exist because using the market has costs: chiefly finding out what the prices are and negotiating a contract for each exchange. A firm grows until organising one more transaction inside costs as much as buying it outside. To Coase’s costs we add a third, checking the work, which later writers developed; it is our extension, not his. Oliver Williamson added that the decision turns on the features of each transaction, above all how specific to the firm the work is. In his Nobel lecture (2009), simple transactions with no specific assets go to the market, while those that rest on firm-specific investment and face large disturbances do better inside the firm. Work that only makes sense inside one organisation stays inside.

On this reading, the question is whether the cost of finding, agreeing and checking is falling. We think it is, for two reasons. AI agents can do much of the searching and contracting. And a card with a verified record lowers the cost of checking. A recent NBER working paper, "The Coasean Singularity?" (Shahidi, Rusak, Manning, Fradkin and Horton, 2025), makes the first argument in detail: AI agents can cut search, communication and contracting costs, so more can be bought. It also names the risks: congestion and price obfuscation in agent-run markets, and new problems for regulators.

An earlier note in our programme archive made the same link to Coase; we treat it as interpretation that still needs a full literature review, not as a finding. The point we take from Williamson is the caution: not everything will recompose. Firm-specific work will stay inside, and the theory predicts which.

9. Same shape as the twin · Proposal

The fourth part of our position is that an organisation must see itself before it rebuilds itself. That is why the card has the shape it has.

A digital twin of an organisation describes its work as objects with inputs, outputs, an owner, dependencies, risks, evidence, unknowns and cost. The Element card holds the same things under other names.

Download this table (CSV), free to reuse with a credit.

Because the shapes match, a capability found inside an organisation can be compared like for like with one offered outside. The journey we propose is: build the twin; take a process apart into capabilities; find one where an outside capability might do better; let the twin write the Formula from what it already knows (today’s cost and yield become the baseline); search the table; try the candidate on the copy, not the company; and only then engage, measure, and keep the option to take the work back.

We should be exact about what exists. The Periodic Table and its card format are published as drafts. The registration forms are previews: nothing is stored yet. The steps that compare and swap capabilities on a twin are design, not working software. A separate paper (Working Paper 7) will deal with the twin.

10. Where this could fail · Counterargument

We take the case against seriously. These are the strongest objections we know.

Coordination costs. Every split adds a hand-off. An earlier review in our own programme put the hostile case bluntly: a market of capabilities may cost more to coordinate than hierarchy, intensify measurement and control, weaken careers and overstate how substitutable work is. Williamson predicts that firm-specific work stays inside. If the costs of the hand-offs exceed the gains, the theory fails where it matters.

Gaming. Buy work by its output and the output will be gamed. Goodhart’s law applies directly. Goodhart’s own point (1975) was about statistics used for monetary control; the popular wording, "when a measure becomes a target, it ceases to be a good measure", is a later summary, commonly credited to Strathern (1997). A capability paid per qualified lead will find ways to call a lead qualified. Our answers (measure the next step, sample the product, keep failures on the record) are untested.

Weaker agents lose without noticing. In Project Deal, a pilot Anthropic published on 24 April 2026, Claude agents bought and sold real items for 69 employees and struck 186 deals. People represented by the stronger model got better deals, yet those with weaker models "didn't notice their disadvantage". Anthropic notes the pilot was small and self-selected. A market of capabilities run partly by agents could quietly favour whoever has the best agent.

Work that resists description. Some work is judgement, relationship and know-how nobody wrote down. It may not reduce to input, transformation and output without losing what matters.

A catalogue of vague labels. An earlier review warned that a weak vocabulary could turn a precise-looking registry into a list of ambiguous labels. That applies to us.

Nothing new. Others have taken jobs apart before. Jesuthasan and Boudreau (Work Without Jobs, 2022, and Reinventing Jobs, 2018) deconstruct jobs into tasks and reconstruct them, with automation. Lawler argued in 1994 for organisations built on competencies rather than jobs. Critical work, such as Rogiers and Collings (2024), asks what organisations and people lose when jobs go. Our claim to novelty is narrow: the join of one card for people, software and AI, search by input and output, trust from history, and the twin. If that join adds nothing that these ideas lack, the paper is just different vocabulary.

People. Atomic work could weaken bargaining power, careers and professional identity, and cut the apprenticeships through which people learn. Whether a person listed as a capability is an employee, a contractor or a business is a legal question; we will not state an answer without legal advice.

11. Who answers for it · Proposal

Every capability on the table has a Nucleus: a named person who answers for it. That holds when the capability is software or AI. The author’s position is that software and AI can do the work, but a person keeps the deciding.

This is also the law in one place we have checked. Under the UK Companies Act 2006, section 155, "a company must have at least one director who is a natural person". The section does not mention AI; our reading is that an AI system is not a natural person, so it cannot be the director this section requires. legislation.gov.uk also lists a change not yet in force: the Small Business, Enterprise and Employment Act 2015, s.87, would omit section 155 and require every director to be a natural person (new s.156A, with exceptions that regulations may set under s.156B). We have read both on legislation.gov.uk; a solicitor should confirm how it applies before anyone relies on it.

When a chain of capabilities fails, who is accountable: each Nucleus, a lead Nucleus for the Molecule, or the organisation that assembled it? We do not know yet. It is one of the scenarios we will test.

12. What we will test · Hypothesis

A proposal is only worth what it predicts. We will run three experiments first, in the Periodic Table’s lab, on invented organisations before any real one.

1. Titration. Does a short trial (10 transactions, or a set time) predict how a capability performs over months? We will compare trial results with later results, against a pass mark set in advance.
2. Split or bond? Is one larger capability better or worse than a chain of smaller ones doing the same work? We will compare total cost, time, hand-off failures, time to fix, and how easy it is to swap one part. This is the test of "no fixed bottom": it tells us where splitting stops paying.
3. Who holds the data. What happens at the end of a bond under each of the five models: what is lost, what it costs to switch, and whether the capability can learn across clients.

Over time we will also measure the gap between declared and verified claims on cards. If declared yields drift away from verified ones, the trust model is failing.

What result would show the central claim is wrong? We have not yet fixed that, and we would rather say so than pretend. The candidates we are considering are: chains of small capabilities that consistently cost more and fail more than the roles they replace; buyers who, given the choice, keep choosing job titles over input and output; and records that cannot be kept honest. The author will choose, and the choice will be dated here before the first results are in.

We will publish negative results as well as positive ones.

References

Status: checked means we read the source (or its official page) and confirmed the detail we use; register means the reference is from our source register and has not been re-read for this paper.

• A2A Project (2026). Agent2Agent (A2A) Protocol Specification, version 1.0.0: Agent Card. a2a-protocol.org/latest/specification. Checked (no version date on the page). The site is copyright The Linux Foundation and announces that A2A has joined the Agentic AI Foundation.
• Anthropic (2026). Project Deal: our Claude-run marketplace experiment. Posted 24 April 2026. anthropic.com/features/project-deal. Checked.
• BIAN (Banking Industry Architecture Network). Service Landscape 14.0 (February 2026). bian.org. Checked (the portal needs a free sign-up).
• Coase, R. H. (1937). The nature of the firm. Economica, 4(16), 386–405. doi:10.1111/j.1468-0335.1937.tb00002.x. Checked (publisher copy paywalled; read through a text index).
• Gebru, T., et al. (2018). Datasheets for datasets. arXiv:1803.09010. Checked.
• Goodhart, C. A. E. (1975, publ. 1976). Problems of monetary management: the U.K. experience. Papers in Monetary Economics, Vol. I, Reserve Bank of Australia (conference July 1975). Reprinted in Goodhart, C. A. E. (1984), Monetary Theory and Practice, Macmillan, pp. 91–121, doi:10.1007/978-1-349-17295-5_4. Original not yet read; the common wording is a later summary.
• Grubic, T., & Jennions, I. K. (2018). Do outcome-based contracts exist? The investigation of power-by-the-hour and similar result-oriented cases. International Journal of Production Economics, 206, 209–219. doi:10.1016/j.ijpe.2018.10.004. Checked (accepted manuscript).
• Hardinges, J. (2020). Data trusts in 2020. Open Data Institute, theodi.org. Checked. Gives the current definition and the October 2018 one it replaced.
• Jesuthasan, R., & Boudreau, J. W. (2018). Reinventing Jobs: A 4-Step Approach for Applying Automation to Work. Harvard Business Review Press. Register.
• Jesuthasan, R., & Boudreau, J. W. (2022). Work Without Jobs: How to Reboot Your Organization's Work Operating System. MIT Press. Checked (the publisher’s description, not the book).
• Lawler, E. E. (1994). From job-based to competency-based organizations. Journal of Organizational Behavior, 15(1), 3–15. doi:10.1002/job.4030150103. Checked (abstract).
• Malone, T. W., Crowston, K., & Herman, G. A. (eds.) (2003). Organizing Business Knowledge: The MIT Process Handbook. MIT Press. Register (book not re-read).
• Malone, T. W., Crowston, K., Lee, J., Pentland, B., et al. (1999). Tools for inventing organizations: toward a handbook of organizational processes. Management Science, 45(3), 425–443. doi:10.1287/mnsc.45.3.425. Checked (source of the "sensible alternatives" quotation).
• Mitchell, M., et al. (2019). Model cards for model reporting. ACM Conference on Fairness, Accountability, and Transparency (2019). doi:10.1145/3287560.3287596; arXiv:1810.03993. Checked.
• Model Context Protocol (2025). The official MCP Registry, preview, 8 September 2025. Checked.
• Rogiers, P., & Collings, D. G. (2024). The end of jobs? Paradoxes of job deconstruction in organizations. Academy of Management Perspectives, 38(2), 177–196. doi:10.5465/amp.2022.0236. Checked (accepted manuscript).
• Shahidi, P., Rusak, G., Manning, B. S., Fradkin, A., & Horton, J. J. (2025). The Coasean singularity? Demand, supply, and market design with AI agents. NBER Working Paper 34468. Checked.
• Strathern, M. (1997). ‘Improving ratings’: audit in the British University system. European Review, 5(3), 305–321. Register (publisher’s abstract only; the popular wording of Goodhart’s law is commonly credited to this paper, not confirmed in its text).
• The Open Group. TOGAF Series Guide: Business Capabilities. Checked (guide date not confirmed; the page now needs an Open Group login and was not re-checked on 29 October 2026).
• UK Companies Act 2006, section 155. legislation.gov.uk. Checked; legal confirmation pending. A change not yet in force (Small Business, Enterprise and Employment Act 2015, s.87) would omit it.
• Williamson, O. E. (1975). Markets and Hierarchies. Free Press; (1985). The Economic Institutions of Capitalism. Free Press. Register (books not re-read).
• Williamson, O. E. (2009). Transaction cost economics: the natural progression. Prize Lecture, 8 December 2009. nobelprize.org. Checked.
• Maynard, M. (2026). Recorded founder interview (142 pages) and written answers, October 2026. Private; quoted by page number. Mark’s account.
• Maynard, M. (2026). Author's notes, October 2026. Private; quoted by page number.

How this paper was written

This paper starts with the author: 25 years inside organisations, a recorded interview and his notes, where the ideas are set out. We use AI (mainly Claude, by Anthropic) as a writing and research assistant, and we say so openly.

• The ideas are the author's. They come from his experience, his interview and his notes, and we quote them by page.
• AI helps write them up: it turns that material into clear, well-organised text, makes sure no important point is missed, finds and checks supporting sources, and prepared the Spanish version.
• The author has the last word: he reviews, corrects and approves every section, and answers for every claim.
• The paper has not been submitted for a degree or to a journal. If that changes, we will follow that institution's rules on AI.
• The notes behind it are private, because they mention clients and colleagues. We quote them by page number, so every quotation can be held to its source.

Conflict of interest

The author is a founding member of Stratapault, a separate business that builds digital twins of organisations. Stratapault does not yet sell or operate the capability model in this paper, and nothing here describes what its product does today: it works at the second stage, seeing the organisation as it really is. This research is independent, free and exists to share the work; it is also where many of our ideas start, and we hope organisations will one day use Stratapault to carry out their own digital twinning. We mention Stratapault from time to time because it is the author's working model: the business where the idea of a twin of the organisation is put into practice.

Respond

Add your view.

Academics, practitioners, sceptics: tell us where this is right or wrong. Mark reads every response, and the best are published here with your name and role, with our reply.

Your name Email (never published)

Role and organisation Link to your profile or work (optional)

Why you are commenting: your field or experience, in a line or two Your response You may publish my response with my name and role, on this site and in future editions of the book. Remember my details on this device Send my response

We use your email only to reply. Privacy.