The standard argument about artificial intelligence and labour still begins with substitution. Which tasks will software perform more cheaply than people, which occupations will survive, and which workers will be complemented rather than displaced. By mid-2026, that framing looks incomplete. Across law, design, software, finance, medicine and administration, the more immediate institutional problem is not merely replacement. It is attribution. When value is produced by a mesh of humans, models, data, prompts, edits, approvals and automated follow-through, the hard problem is no longer only who did the work, but what counts as the work.
That may sound abstract. In practice it is becoming a mundane managerial issue. A contract draft may be assembled by a model, corrected by a junior associate, approved by a partner and tailored automatically to a client’s internal policy library. A research note may combine automated retrieval, a human thesis, machine-generated tables and editorial judgement. A product launch may rely on dozens of micro-contributions, some creative, some supervisory, some embedded in prior datasets and code. Traditional payroll systems were built for hours, roles and hierarchies. They are less adept at pricing layered contribution chains.
This points to a distinct thesis about the future of work and finance. The agentic economy may alter compensation architecture before it fully alters headcount. Instead of a sudden world of mass technological unemployment, many sectors may first experience a slower but deeper shift towards traceable contribution, residual claims and royalty-like payments attached to reusable intellectual and procedural assets. That would change what workers bargain over, what firms account for, and what states may eventually tax.
From jobs to contribution graphs
Industrial labour markets were organised around posts. Even knowledge work, for all its flexibility, mostly inherited that grammar: a person occupies a role, performs a bundle of tasks and receives a salary plus perhaps a bonus. Digital platforms introduced a more granular logic by tracking clicks, rides, deliveries and engagement. The spread of agents introduces something more consequential: contribution graphs that can log who initiated, refined, verified, rejected or operationalised an output.
Such graphs are not neutral mirrors of reality. They are designed artefacts. Yet they matter because organisations increasingly need them for compliance, quality control and risk management. European AI rules require attention to transparency, human oversight and documentation for certain systems. NIST’s AI risk framework similarly emphasises governance, traceability and accountability. Once firms build the machinery to record process, the same machinery can be used to allocate rewards.
The hard problem is no longer only who did the work, but what counts as the work.
In other words, attribution is moving from a cultural nicety to an economic infrastructure. That has implications well beyond whether an employee receives praise on a performance review.
Why royalties are escaping the arts
Royalty systems have long existed in music, publishing, licensing and patents because those sectors deal in reproducible assets whose use can be counted, however imperfectly. Most salaried employment, by contrast, bundled present effort and future reuse together. An engineer designs a process improvement, the firm deploys it at scale and the engineer receives salary rather than a metered stream of payments. A consultant creates a template, a hospital administrator writes a workflow, a teacher develops materials; the institution captures most downstream value.
Software agents change the economics of reuse. A playbook, prompt library, decision tree, compliance rule-set, synthetic training corpus or client-specific orchestration can now be deployed repeatedly at low marginal cost across teams and geographies. Once that happens, pressures arise to separate one-off labour from reusable asset creation. In some professions this will look like internal royalties, with bonus pools linked to the frequency or profitability of adopted assets. In others it may resemble licensing arrangements inside employment contracts, especially for senior specialists whose methods can be operationalised by agents.
The hard problem is no longer only who did the work, but what counts as the work.
This does not mean every worker becomes a rights-holder. Firms will resist complexity, and employment law in many jurisdictions gives employers broad claims over work produced in the course of employment. But the pressure is structural. The more production depends on reusable cognitive artefacts, the more organisations need a language for recognising who created them and how benefits should be shared if they are used at scale.
The return of metering
For years, the digital economy normalised measurement at the consumer edge while leaving much of professional value creation comparatively coarse. Salary bands, annual reviews and team budgets remained the principal tools. Agentic systems are bringing metering inward. They generate logs of prompts, interventions, exceptions, approvals, model confidence, error rates and revision histories. Some of this is built for safety. Some is needed for audit. But metering rarely stays confined to its original purpose.
The economic temptation is obvious. If firms can identify which templates reduce legal review time, which analysts’ checklists lower error rates, or which clinicians’ protocols improve triage efficiency, they can connect those effects to compensation. The result is not necessarily a freer or fairer workplace. Metering can become a surveillance technology as readily as a basis for reward. The same visibility that supports residual payments can also justify harsher benchmarking and tighter managerial control.
That is why the politics of attribution will matter. Workers may reasonably ask whether they are being measured only when it lowers costs, but not when it creates claims on future value. Employers may argue that collective production makes individual royalties impractical. Both sides will often be right.
What this means for professions once thought immune
The familiar claim that routine work is most vulnerable to automation remains broadly plausible, as research from the OECD and the ILO has suggested. But the compensation effects may hit high-skill professions first because those professions produce codifiable methods with large reuse value. In law, for instance, precedent, drafting logic and negotiation heuristics can be embedded in agentic workflows. In medicine, diagnostic support and administrative pathways can capture specialist judgement indirectly, even where final responsibility remains human. In finance, analyst frameworks, risk flags and portfolio memos can be distilled into institutional playbooks. In academia, teaching materials, peer review scaffolds and grant-drafting routines can all be operationalised.
Once professional judgement is partly converted into reusable system behaviour, a new question appears. Is the professional being paid for live performance, or for seeding an asset that will continue to generate value after the immediate task is complete. Much white-collar anxiety about AI has focused on replacement. A quieter concern is expropriation: not losing today’s role, but losing the future rents from one’s accumulated know-how.
In an agentic economy, attribution is becoming an economic primitive.
The accounting problem hiding underneath
National accounts and company accounts are not well designed for this transition. The System of National Accounts distinguishes between labour compensation and returns to capital, yet digital production often blurs the line. If a worker’s expertise is converted into an enduring process asset, is subsequent value creation still labour income, or has it become a capital-like return attached to an intangible? Existing accounting practices already struggle with internally generated intangibles. The growth of agentic systems makes that ambiguity more pervasive.
In an agentic economy, attribution is becoming an economic primitive.
For firms, this affects incentives. If compensation remains overwhelmingly tied to present labour while the firm captures most gains from reusable intangibles, employees have reasons to withhold their best methods, or to reserve them for outside work where they retain rights. If, by contrast, firms promise residual participation, they may secure greater cooperation in codifying expertise. This is less a philosophical issue than a practical one about organisational learning. Institutions that cannot persuade skilled workers to externalise tacit knowledge may find their automation efforts shallower than expected.
New currencies without new money
Talk of novel labour currencies often drifts towards speculative token schemes. The more plausible near-term development is duller and more significant: internal units of account for contribution, provenance and downstream usage. Organisations already operate with shadow currencies of billable hours, sales credits, citations, performance points and carry. Agentic production may extend this repertoire with systems that assign weighted claims to people who create prompts, taxonomies, rule-sets, validation data, exception handling and quality assurance.
These are not currencies in the monetary sense. They are accounting conventions that determine who gets paid, promoted or credited when outputs are co-produced by humans and machines. Because they are conventions, they can be contested. A hospital may treat the nurse who identified edge cases as central to value creation; another may reward only the physician who signed off. A bank may pay for successful deployment of internal agents; another may pay only for nominal ownership of the client relationship. The distributional consequences could be large even if the technology is similar.
Why social contracts may shift inside firms before they shift at the state
Much public debate still centres on grand national remedies, above all universal basic income, robot taxes or sweeping reductions in working time. Those debates will continue, and some may eventually mature into policy. But the first durable social contracts of the agentic era may emerge at a lower level: employment contracts, collective bargaining agreements, professional codes and sector-specific standards for data rights, audit rights and residual compensation.
There is a precedent here. Platform work was initially framed as a technological novelty, yet many of the real disputes turned on classification, transparency, ratings and access to information. Something similar is happening with agents, only among salaried professionals rather than gig workers. Questions likely to become ordinary include whether workers can inspect logs used in performance assessment, whether they can contest attribution models, whether contributions to reusable systems count in promotion decisions, and whether internal knowledge capture triggers any right to residual payment.
These are not glamorous issues. They are, however, the sort of institutional details that determine who benefits from productivity gains.
The role of law, standards and professional norms
No single body of law resolves this. Intellectual-property regimes matter, but they are neither universal nor sufficient. Copyright is ill-suited to many process innovations. Trade secret law often favours firms. Patent protection is expensive and narrow. Employment law governs ownership and bargaining power but varies widely. Data protection can constrain monitoring, yet usually does not settle remuneration. This legal patchwork is one reason standards and professional norms may matter as much as statutes.
UNESCO’s ethics recommendation and NIST’s framework both point, in different ways, towards transparency, accountability and human oversight. The EU’s AI regulation has given compliance teams reasons to document system behaviour. None of these instruments creates a general right to AI royalties. Yet they push organisations to build provenance and governance capabilities. Once those exist, labour-market actors can use them to negotiate over compensation and recognition.
The next social contract may hinge less on universal basic income than on universal auditability.
Professional bodies may also intervene informally by redefining what merits credit. In some fields, supervision, validation and ethical judgement may gain value precisely because agents can produce plausible but unreliable outputs at scale. In others, the prestige hierarchy may invert, with those who design robust workflows and exception protocols becoming more important than those who merely generate first drafts.
The risk of a new hidden inequality
If attribution becomes central, inequality may increasingly depend on who can convert skill into recognised, reusable assets. Workers with bargaining power, scarce expertise or mobility across firms may secure residual claims. Others may see their know-how absorbed into systems with little compensation beyond wages. The resulting divide would not map neatly onto old distinctions between manual and cognitive labour. It would separate those who retain rights in codified expertise from those whose contributions are logged but not economically recognised.
That possibility should temper both utopian and catastrophic accounts of AI. Productivity gains can coexist with stagnant median bargaining power. Equally, continued employment can mask a transfer of future value away from labour. Historical arguments over mechanisation often focused on whether workers kept their jobs. In the late 2020s, a subtler issue may be whether they keep any claim on the reusable fragments of intelligence they help create.
What finance will notice first
Investors and lenders tend to care less about philosophical debates than about revenue durability, margins and control over intangible assets. From that perspective, firms that can systematically capture employee know-how into agentic systems may appear more scalable and resilient. But there is a countervailing risk. If the legal or cultural basis for that capture is weak, the asset may be fragile. Key staff may leave, challenge ownership, or simply stop sharing high-value methods. The quality of a firm’s attribution and compensation regime could therefore become a hidden determinant of enterprise value.
This is one reason the future of work and the future of finance are converging. Questions once treated as human-resources minutiae may increasingly matter to balance sheets. How reusable are internal decision assets. Who controls them. Are the people who maintain them motivated to improve them. Can their provenance be audited. In industries heavy with regulated judgement, these questions may prove more material than headline staff reductions.
What to watch by the end of the decade
By 2030, the most telling developments may not be spectacular humanoid replacements or universal income schemes. More prosaic indicators will reveal the direction of travel: whether employment contracts begin to specify rights over prompts, workflows and synthetic data; whether collective agreements demand access to attribution logs; whether companies report more systematically on internally generated intangibles; whether professional status comes to attach to validation and governance rather than only origination.
The next social contract may hinge less on universal basic income than on universal auditability. If workers cannot see how agentic systems use, score and extend their contributions, bargaining over fair reward will be almost impossible. If they can, new forms of compensation may become thinkable even within conventional firms. The quiet revolution in work, then, may not be a clean break from wages to some futuristic token economy. It may be the much more consequential shift from opaque salaried effort to contested, metered claims over machine-amplified value.
That would still leave familiar political choices. States may decide to tax intangible rents differently, recognise new categories of labour income, or strengthen portability of reputation and contribution records across employers. Firms may choose between extraction and partnership. Professions may defend old boundaries or redesign them. But the central economic novelty would remain the same. In an age of agents, the question is not only whether labour is replaced. It is whether labour can trace, and bargain over, the value it helps to encode.
