The Human–Machine Coproduction Index (HMCI): Redesigning Job Descriptions for the AI Era
Stop talking about 'upskilling'. Start measuring exactly what percentage of each role belongs to the human and what belongs to the AI agent — and design the job around the split.
- Most job descriptions in 2026 are still written as if AI agents don't exist. They do, and they're doing 20–60% of the work.
- The Human–Machine Coproduction Index (HMCI) breaks every role into tasks, then tags each task with the human/AI accountability split.
- MIT's 2024 GenAI Productivity studies show a clean split lifts both speed (+40%) and quality vs. ambiguous co-pilot setups.
- Crucial: decisions with legal, ethical, or safety implications stay 100% human. The Index makes that legally defensible.
- We provide the 7-step audit and a ready-to-use HMCI worksheet pattern.
Read your last job posting for a marketing manager, recruiter, paralegal, or junior analyst. Now look at what those people actually do in 2026. The gap is enormous. Half the tasks are now co-produced with an LLM, a co-pilot, or an automated agent. The other half — the half that requires consciousness, accountability, and judgment — is more important than ever. Most job descriptions describe neither.
What HMCI is
The Human–Machine Coproduction Index is a simple numerical breakdown of every role's task list, scored on two axes: who decides and who executes. The output is a clean percentage split — e.g. 'this recruiter role is 35% human-only, 30% human-decides AI-executes, 25% AI-decides human-reviews, 10% AI-only'.
| Quadrant | Who decides | Who executes | Example task |
|---|---|---|---|
| Pure human | Human | Human | Performance review conversation, termination call, ethical edge case |
| Human-led | Human | AI | 'Draft this offer letter using these terms' — human approves before send |
| AI-led | AI | Human | AI flags 12 high-risk transactions; human investigates and acts |
| Pure machine | AI | AI | Recurring report generation, ATS resume parsing, calendar tetris |
The 7-step HMCI audit
- Pick one role. Recruiter and marketing manager are the easiest pilots in 2026.
- List every task that person does in a typical week. Aim for 30–60 tasks. Use a week of time-tracking or self-report.
- For each task, classify into one of the four HMCI quadrants.
- Compute the percentage split. Most knowledge roles in 2026 land at 30/30/25/15 ± 10.
- Identify mismatches: tasks currently in the 'pure human' bucket that an LLM could safely do, and tasks in the 'pure machine' bucket that have ethical or relational implications and should have a human reintroduced.
- Redraft the job description around the new split. State explicitly which tasks require human accountability.
- Update the comp band — roles with more 'pure human' accountability should pay more, not less. That is the inversion many organisations are missing.
Where the human must stay 100%
Under the EU AI Act (2024), GDPR Article 22, and emerging US state laws, the following must remain human-decided: hiring, firing, promotion, discipline, performance ranking with consequences, pay decisions, accommodation decisions, and any decision involving children, healthcare, or legal status. Documenting the human in the loop is HR's legal shield.
- Hiring & firing decisions — AI can shortlist; humans must decide.
- Performance ratings with material consequences — AI can summarise; humans must rate.
- Compensation decisions — AI can benchmark; humans must set.
- Accommodation and disability decisions — AI cannot decide; humans must.
- Customer-facing decisions with safety implications — humans must own.
Designing the new role
Once you have the HMCI split, three things change about the job:
- The job description names the AI tools used and the decisions reserved for humans. This is legally defensible and a recruiting differentiator.
- The interview process tests judgment, not just execution. If the AI can execute, you are hiring for the part that decides.
- Career ladders are rewritten — progression moves you up the human-decision axis. The most senior people in any role are 80%+ pure human.
Takeaways
- Stop redrafting job descriptions around 'upskilling'. Redraft around the human/AI split.
- Pure-human tasks are now the premium tasks. Pay for them accordingly.
- The EU AI Act has made documenting the human-in-the-loop legally required, not optional.
- An HMCI audit is the most concrete way HR can prepare every role for 2027–2030.
- Brynjolfsson, Li, Raymond — Generative AI at Work — NBER Working Paper, 2024 update
- BCG x Harvard — Navigating the Jagged Technological Frontier — BCG / HBS, 2023
- Pew Research — Use of GenAI in the Workplace 2024 — Pew Research, 2024
- EU AI Act — High-Risk Workplace AI — European Parliament, 2024
Read next
All playbooksAn honest map of where AI is delivering value in HR today, where vendors are overselling, the risk and regulation landscape, and a 30/60/90 pilot framework.
Blockchain and zero-knowledge proofs are quietly handing the employee back control of their performance history. Here is what a portable, employer-signed career ledger looks like — and what HR needs to do before 2027.
When the algorithm becomes your boss, employees stop pushing back in meetings and start pushing back in code. A digital-anthropology field guide to the 2026 workplace.