Implications
Four practical consequences for teams adopting agents today.
The unit of work is the right abstraction; collaboration is what it unlocks; the environment is what makes it true and, over time, what makes it cheap.
Four practical consequences follow from this note.
Write definitions of done, not better prompts. The leverage is in specifying the end state and making it checkable by the environment. Prompt phrasing is a rounding error by comparison.
Standardize the handoff package. If every unit carries the same contract (done, budget, identity, state, record), humans and agents can exchange work freely, and the org chart can mix them without inventing new process.
Run repeat work in the same environment. The cost curve only exists if memory, records, and definitions of done accumulate in one place. Spreading similar units across fresh contexts resets the agent to unit-1 economics every time.
Audit environments the way you audit code. Gameable checks, unmetered actions, and self-reported completion are environment bugs, and they will be found by agents, under optimization pressure, reliably. The intervention-rate signal from the Phase 1 thesis applies here too: when humans stop having to re-verify what the environment already verified, the environment is doing its job.
How environments help businesses
Persistent environments make intent cheaper to infer: the same unit of work costs ~42% less by run 50.
How the Environment Affects Agent Performance and Token Cost
A pilot measuring how a workspace's project context shapes a coding agent's performance and token cost: richer context is nearly free to run, it does not hurt task success, and it lowers the cost of getting oriented.