Intent sets the destination; demonstrated work decides which pieces count.
Fromcurriculum→Tocapability pathway
In the proposed system the learner starts with an intent rather than a course. A personal agent assembles the pathway from pieces that different providers supply: knowledge, simulation, practice, assessment and work. The evidence supports tying learning to employer demand and real work, and it supports breaking capabilities into skills. It doesn’t support assuming that any collection of micro-skills adds up to something an employer will pay for. Pieces count when they lead to demonstrated work, and knowledge providers are paid, eventually, for what their pieces contribute.
Intent sets the destination
A conventional curriculum begins with what an institution can teach. The proposed pathway begins with what the person wants to be able to do: I want to become capable of X [CN]. The agent works backwards from that intent to the competencies involved, what the person can already demonstrate, and where each missing piece can be learned, practised and proven.
The concept’s own example is an emerging form of agentic AI work. It might break down into data preparation, API integration, tool calling, retrieval, evaluation, workflow orchestration and testing, with different providers contributing different pieces [CN]. The network doesn’t prescribe the sequence. It makes the options discoverable, and the person’s agent chooses among them.
Open infrastructure provides the choices. The person’s intent sets the destination.
Learning alone is not the product
The evidence doesn’t support treating course completion as capability. A U.S. Department of Labor review of 54 studies of work-based learning found that sector-focused training programs improved education and training outcomes in every well-designed study that measured them, while their effects on employment and earnings were mixed [1].
The programs that did change earnings share a recognisable design. Across four randomised evaluations of sector-focused programs, earnings rose 14–38% in the year after training, and they were still 12–34% higher in the latest year measured, as long as eleven years later [2]. Those programs combine up-front screening, occupational and soft-skills training toward credentials employers recognise, strong employer relationships, placement and support services. The gains came mainly from getting people into better-paid jobs in the targeted sectors, not simply into more work [2].
Open online learning looks different. In one study of highly motivated, active participants in career-related online courses, participation helped workers stay employed but did not raise their wages. The authors caution that the result doesn’t generalise to typical learners [3].
Learning alone is not the economic product. Learning connected to demand, practice and placement is.
Micro-skills add up only inside a frame
Breaking capabilities into skills is well established. National and regional skills frameworks catalogue thousands of occupations and skills, and several public employment services use them to suggest skills missing from a profile or a vacancy [4]. The harder claim is that small pieces recombine into something valuable.
Here the evidence is thin and mixed. A study of stackable credentials, using national, survey and college-system-level data, found weakly positive and inconsistent earnings gains from combining credentials, generally no different from the returns to holding a single postsecondary credential [7]. In Europe, some industry certifications in manufacturing and retail stack into ladders from entry-level jobs toward technician and engineering roles, but industry bodies that employers trust define those ladders [5]. The sector programs that worked also trained toward certifiable, transferable skills that employers already recognise [2].
The pattern suggests that composition works when someone employers trust defines how the pieces fit together. So the capability graph shouldn’t start as a universal map of micro-skills. It should start inside one capability family, defined with that family’s employers (Shift 1), and grow only where it improves a real transaction.
- Micro-skill
- Capability
- Task
- Work output
- Economic outcome
A micro-skill matters when it is part of something an employer will pay to have done.
Go deeper
Distributed formation
Once the pieces can be trusted, a pathway can be assembled across the economy. It might combine knowledge from one provider, a simulation from another, a micro-skill from a practitioner, an apprenticeship with a small firm, an assessment from an independent evaluator and projects with several employers [CN]. Supply widens. A practitioner who knows one valuable skill can contribute without becoming a school, and a small firm with real project work can contribute without running a training department.
A caution comes with it. European research on short credentials finds provision fragmented and learners’ tools for navigating it immature. It also warns that very narrow credentials can splinter occupational skills, while moving between jobs depends on broader, transferable skills [5]. A pathway built from micro-skills alone risks producing someone who can do fragments of a job. The agent has to keep the broader skills in view.
Knowledge providers: from enrolment to contribution
Today a university, course creator or platform is usually paid for access, enrolment or completion. In the concept, knowledge becomes a component of capability formation. Learners or trainers select content with human-capital credits, and a provider is paid when its content contributes to capability demonstrated in an assessment [CN].
The evidence urges patience. Outcome-linked payment to providers is only weakly supported, attribution is difficult, and paying on downstream outcomes invites gaming and the quiet selection of easier learners. A person’s eventual performance depends on the content, the trainer, the apprenticeship, the employer, the task and the person, so the network can’t simply pay whichever course came last.
A better first step is contribution measurement. Providers publish versioned competencies and the prerequisites they assume, and the network observes which resources tend to come before demonstrated capability, and in which contexts. Payment linked to outcomes can follow once there is something reliable to pay on.
Illustrative: two providers teach the same micro-skill. Provider A has high completion, but its learners rarely go on to demonstrate the skill in paid work. Provider B has lower completion, but its learners reliably do. A market organised around enrolment rewards A. A market organised around demonstrated capability would eventually reward B, but only if the network can tell the two apart, and that is itself the test.
Open does not mean unaccountable
Distributed formation also distributes responsibility. Who answers when a pathway fails, when an apprenticeship turns out to be poor, or when a micro-skill was never really acquired? The programs with the strongest results don’t leave this to chance. Screening, training, placement and follow-up sit with one organisation that answers for the outcome [2]. In an open network, that role belongs to an accountable intermediary in each capability family. It doesn’t belong to the network as a whole, and it doesn’t belong to the learner (Shifts 4 and 9).
Open does not mean unaccountable.
Not our invention
Competency-based education, stackable credentials, skills frameworks and sector-focused training all exist. Some public employment services already show job seekers which alternative occupations their skills fit and which gaps they would need to close [6]. The contribution here is a pathway that crosses independent providers, is orchestrated on the person’s behalf, and is judged by demonstrated work rather than completion.
What could go wrong
- Illusory composition.The agent may assemble pieces that never add up to capability an employer will hire.
- Choice overload.More providers bring more choice and a heavier curation burden.
- Selection effects.Motivated learners may succeed whatever the pathway, which would make formation look better than it is [3].
- Narrowness.A pathway made of micro-skills may neglect the transferable skills that carry people between jobs [5].
- Gaming under outcome funding.Providers may teach to the assessment or avoid learners who are harder to place.
How we’d test it
| The test | What we would measure | What would end the idea |
|---|---|---|
| Training-to-work conversion (Experiment 3) | For one job family, curriculum only against curriculum plus assessment plus a paid project: time to first paid work, earnings at 6 and 12 months, retention, total cost | Completion rises but employment, earnings or productivity don’t. The review’s planning threshold is 15% faster entry to productive work or 10% higher 12-month earnings, at no more than 15% higher cost |
| Composition | Distributed pathways against a single-provider pathway to the same capability: demonstrated performance, cost, completion | Distributed pathways do worse once coordination costs are counted |
| Provider contribution | Whether later demonstrated performance separates providers that teach the same skill | No separation, so outcome-linked payment has nothing to pay on |
How strong is the evidence?
- Established
- Sector-focused programs that combine training, employer links, screening and support raise earnings for years [2]; course completion alone is a weak proxy for capability
- Evidence-supported
- Small credentials compose best inside ladders that an industry defines [5]; composition in general is weakly supported [7]
- Hypothesis
- Pathways assembled across independent providers can produce hireable capability at lower cost
- Unknown
- Whether outcome-linked payment improves provider incentives once gaming is accounted for
Sources
- [1]U.S. Department of Labor, Clearinghouse for Labor Evaluation and Research, synthesis of evidence on registered apprenticeship and work-based learningChecked
- [2]Katz, Roth, Hendra and Schaberg, Why Do Sectoral Employment Programs Work? Lessons from WorkAdvance, Journal of Labor Economics (2022)Checked
- [3]Castaño-Muñoz and Rodrigues, study of MOOC participation and labour-market outcomes, Computers & Education (2021)Checked
- [4]World Bank, Solutions for Youth Employment, draft note on digital job-matching platforms in public agencies (2023)Checked (draft discussion note)
- [5]Cedefop, Microcredentials for labour market education and training: the added value for end users (2023)Checked (executive summary and chapters 1–3)
- [6]Broecke, Artificial Intelligence and Labour Market Matching, OECD Social, Employment and Migration Working Papers (2023)Checked
- [7]Community College Research Center, study of the labour-market value of stackable credentialsChecked (summary page)
- [CN]Open Commerce for Human Capital: concept document consolidated from the discussion (internal)Provided
Spotted a source we missed or a rule we misread? Tell us what we got wrong.