The Human × AI Manifesto · Version 1.0

Human Agency at AI Speed

Four areas of actionTen commitments
Explore the manifesto

When organisations bring AI into their existing processes and structures, individual tasks speed up tremendously. The organisation as a whole doesn’t simply get faster, though. New bottlenecks appear in review, in decision-making and in coordination between people, and the people at those bottlenecks risk burning out.

This is the core change: AI can make execution very fast, so human attention, judgment and coordination become the scarce resources. Practices that used to count as good management, such as clear ownership, shared understanding and independent checks, therefore become essential. Without them, AI speed turns into rework, overload and decisions nobody really owns.

Our goal is to help build organisations where human judgment and AI capability strengthen each other: organisations that use speed to create impact at scale and to give people more room to think, learn and collaborate.

This manifesto is meant for any organisation, commercial or not. When we say customer and offer, we mean whoever an organisation serves and whatever it delivers to them (see Key terms).

How to use these commitments

The ten commitments are grouped into four areas. Start wherever the bottleneck is biggest, but note that the commitments build on each other. Clear outcome ownership (3) underpins decision rights (7). The shared specification (4) enables AI monitoring (5), independent verification (8) and shared learning (9).

I.

Protect human capacity

01Protect people’s attention and capacity to think

Delegate well-defined tasks, whether to AI agents or to colleagues, with agreed acceptance criteria, limits and escalation rules. Batch routine updates, automate routine checks, and keep interruptions for what genuinely needs a person. Judge progress by the outcomes achieved and the total human effort behind them, including supervision, review and rework, rather than by output volume.

Why it matters with AI: AI can produce work faster than people can review it in depth. If supervision and rework go up, the productivity gains are nullified, and a few people end up carrying the hidden cost.

02Decide deliberately how to reinvest the time AI frees

Before scaling AI in a team, agree with the people involved how the freed-up time will be used. Decide what share goes to learning, quality, customer contact and slack for the unexpected, and what share goes to more output. Revisit that split regularly, using the effort measures from commitment 1.

Why it matters with AI: If speed gains automatically become higher targets, the people at the bottlenecks absorb the pressure, and burnout follows.

II.

Share understanding and context

03Own each outcome end to end, across functions

Bring every function involved together, from the first promise to realised value. This may include product, marketing, sales, engineering, operations, finance, and customer success. Name one accountable owner for each outcome and make each person’s contribution clear.

Why it matters with AI: When AI speeds up each function on its own, the handoffs between functions become the bottleneck. A single owner keeps the outcome from falling between fast-moving parts.

04Give every offer a shared, living specification that serves as persistent context

Connect problem statements, value propositions, promises, capabilities, constraints, acceptance criteria and evidence in one versioned specification, maintained by the outcome owner, with all functions contributing. Keep sources, decisions, rationale and responsibilities visible, so that people and AI act on the same understanding.

A specification doesn’t have to be a technical document describing a software. For a consulting service, it might be a single page linking the client’s goals, what was promised, the agreed deliverables and how success will be judged.

Why it matters with AI: AI agents act on whatever context they are given. A shared, current specification is how people and AI stay aligned without constant meetings and reviews.

05Let AI keep shared understanding current

Ask AI to connect new customer feedback, market changes, delivery findings and incidents to the relevant specifications and decisions. When something material changes, AI flags it to the outcome owner with sources and a proposed next step.

Unless the change is urgent, these flags are batched as set out in commitment 1. The outcome owner agrees what AI may read, change and act on, together with the people affected and those responsible for data protection and security.

Why it matters with AI: Without this, the specification goes stale as soon as the work speeds up. With it, people spend their attention on what changed rather than on finding out what changed.

III.

Decide well

06Explore and challenge before committing

Use AI to develop credible alternatives and inexpensive prototypes. Test the riskiest assumptions with customers before committing scope, budget or delivery dates.

For consequential proposals, let AI question assumptions and seek counter-evidence. Then ask a colleague to examine what remains and to make the case for the strongest alternative.

Why it matters with AI: AI makes alternatives and prototypes cheap, so there is little reason to commit to the first idea. An AI challenger can handle the first round of critique, which leaves human reviewers free to focus on the questions that need judgment.

07Keep consequential decisions in accountable human hands

Tier decisions by risk. Lower-stakes, reversible decisions can be delegated to teams or AI within clear limits that may be maintained globally or specifically for the outcome, while consequential ones stay with named people.

Name who can set direction, approve commitments and stop work, starting from the outcome owner named under commitment 3. Record rationale and uncertainty in the shared specification.

Give people authority to reject AI recommendations: rejecting a recommendation should require less justification than accepting it.

Why it matters with AI: AI can recommend faster than people can deliberate and review. Without tiers, every decision queues at the same few people. Without real authority to say no, people end up deferring to the machine.

08Verify results independently

Test delivered outputs against acceptance criteria agreed with the customer, using representative data and checks beyond the builder’s own assessment.

Work produced by AI should be checked by multiple verifiers, not connected to the original builder AI. Over time, confirm that the customer actually realises the intended outcome.

Why it matters with AI: AI produces plausible-looking work at volume, but plausible is not the same as correct or valuable. Whether human or AI, the builder is not a good judge of its own work.

IV.

Learn and stay open

09Turn results and feedback into shared learning

Review successes, failures and customer feedback together, then update the shared specification and working practices. Protect the learning paths people once got from entry-level work that AI now does.

Pair less experienced people with domain and technical experts, and give them room to be critical and creative drivers of the process rather than only reviewing AI output. Make mentoring such as pair building and knowledge-sharing across roles part of the work, using the time set aside under commitment 2.

Why it matters with AI: If AI takes over the work through which people used to learn, organisations will lose tomorrow’s experts. AI creates a fast feedback loop between input and output, which can help with learning, but this opportunity is often not used properly.

10Choose open standards that keep knowledge portable

Even if it is cheap to generate bespoke formats, APIs, schemas and other artefacts, prefer open standards and interoperable interfaces, so that knowledge, specifications and data remain usable across tools, vendors and partners. Where possible, contribute reusable knowledge to open-source projects, standards bodies and professional communities.

Why it matters with AI: Over-use of custom formats and APIs reduces the potential of solutions to be valuable beyond its initial context, and may increase lock-in. It also weakens human agency over how its work is done.

Key terms

Customer:
anyone an organisation serves, such as clients, citizens, members, patients or internal colleagues.
Offer:
anything an organisation delivers to its customers, such as a product, a service, a programme or an internal capability.
Output:
what was delivered, checked against the agreed acceptance criteria.
Outcome:
the value the customer actually realises from what was delivered.
Consequential decision:
a decision that has at least one of these properties:
  • It is hard to reverse.
  • It commits the organisation to a customer.
  • It carries significant legal, financial, safety or reputational risk.
  • It significantly affects people’s roles or wellbeing.
Material change:
a change that would alter a commitment, a priority, an acceptance criterion or a risk assessment if the outcome owner knew about it.