Phase 2: Collaboration
Collaboration collapses toward the mind: humans stay the brain, agents become the runtime, and business impact becomes one blended ledger.
Work at the speed of thought: a few brains holding budgets and definitions of done, surrounded by fleets that settle everything else.
The interface collapses toward the mind
Each era of software removed a layer between wanting and having. The endgame: the only latency left in work is how fast a human can decide what they want and judge what comes back.
The human stays the brain
Brain stays human, skill becomes runtime. Judgment does not leave the loop; it becomes the loop.
- What is worth doing
- What done looks like
- Which output is good (taste)
- Who is accountable
- Execution in parallel, any hour
- Across every system
- At marginal cost
- Swappable models underneath
The contract between the two sides is the unit of work: a definition of done, a budget, a single owner. Intent flows down, verified state flows back, and every cycle mints quirqs. One brain used to serve one employer, forty hours at a time; unbundled, the same brain directs ten fleets.
Business impact becomes a blend
Human and AI work settle in the same ledger, priced in the same unit.
| Horizon | Human-settled quirqs | AI-settled quirqs | Total delivered |
|---|---|---|---|
| Today | most: execution and judgment | assist | 1x |
| Near | judgment, direction, acceptance | most execution | rising |
| Far | few, dense, high-value | nearly all execution | 10x |
The human share shrinks in percentage and grows in value: fewer, denser judgment quirqs steering a much larger total. Because both sides mint into one ledger, "human + AI" stops being a slogan and becomes a P&L line: the mixed quirq ledger is the measured version of this table.
Getting there: agents that work with each other
Coordination, not capability, is the hard part. Without a shared unit, what travels between workers is conversation: context degrades at every hop and accountability blurs. With units of work, each handoff passes a verifiable state plus the contract: done, budget, identity, state, record.
Identity travels with the unit rather than the chat session, so the outcome stays attributable when no single worker saw it end to end. And because units are independently owned, budgeted, and verifiable, a fleet attacks a goal in parallel and re-integrates by checking states rather than reconciling narratives. A manager already delegates in exactly this shape (scope, budget, deadline, check the result), which is why a mixed human-agent org needs one management protocol, not two.
What to watch
Coordination has overhead: agents that loop on each other, trust that is hard to verify, accountability that blurs across many hands. The teams that win this phase keep a clear owner for every outcome even as the work spreads, and protect the apprenticeship path that grows the next generation of brains, because in this future judgment is the entire human job. A fleet without an owner is not an organization, it is noise.
Next: Phase 3, self-evolving systems, where the fleets begin improving themselves.