The person deciding
“Launch the small pilot. We can support the first customers manually.”
Carries the intentThe work keeps going.
Carry the understanding.
b.akery helps people, teams, and agents produce more useful work with less coordination.
An answer is only part of the work. Someone still has to explain the goal, question an assumption, check the result and bring the next participant up to speed.
See what changes when the work becomes the center
What gets harder when intelligence gets easier?
Capable people and agents can still spend much of their effort getting situated: reconstructing decisions, correcting assumptions and working out which version of the story still applies.
We call this the Coordination Tax on Intelligence. A useful discussion can be worth every minute. Rebuilding understanding we already developed is the cost we want to reduce.
Change the center.
Follow what becomes possible.
Explore a fictional pilot: a small service is ready to launch with manual support. Automation is needed before it grows. Each participant holds a different part of that understanding.
Assignments tell us who is doing what.
The reasons and conditions can still live between the assignments.
The agent building
“The implementation is complete. The pilot is ready.”
Carries the implementationThe teammate reviewing
“We still need automation before we expand.”
Carries a conditionTomorrow, someone has to put this back together.
“Is it ready—or are we blocked?” misses the arrangement that made the pilot possible.
An owner can remain accountable while the understanding is scattered across conversations, people and tools.
Make the work the organizing center.
Keep an evolving account that participants can inspect, question and correct.
The pilot, as we understand it
- Goal
- Learn from the first customers.
- Decision
- Proceed with manual support.
- Current state
- The builder reports the implementation is ready.
- Condition
- Automate support before expansion.
- Still to establish
- Confirm support capacity and verify readiness.
Keep who said what, the supporting evidence and what remains uncertain.
What does the next participant need?
You’re deciding whether to proceed.
Manual support was intentional. Confirm who can provide it and how much they can handle before approving the pilot.
The decision you need to make
The same work can support different views. Access, authority and disagreement still matter; a shared account must leave room for all three.
Prepare useful context before someone has to ask.
This is the larger research direction: help develop understanding as the work happens, then carry it into the next contribution.
- 1
Work happens
A person changes the pilot’s capacity. A builder supplies evidence. A reviewer raises a concern.
- 2
Understanding is updated
Retain the change, its reason and what it affects. Keep unresolved questions visible.
- 3
A relevant view is prepared
The next participant gets enough context for their contribution, with the sources they can access.
- 4
The next contribution improves the account
Check the current evidence, do useful work and carry corrections forward into the next cycle.
The ambition: “My agent already knew what mattered—and helped me take the next useful step.”
Illustrative example · conceptual model
“What done looks like” develops through the work.
The goal is a result someone can use—and enough understanding to continue responsibly.
A prototype may change the options. A review may expose a mistaken premise. Two people may mean different things by “ready.” Useful continuity preserves those discoveries, including the reasons behind a decision and the conditions attached to it.
The account must be correctable. Carrying an old misunderstanding into more conversations would make the problem worse.
Our ambition is for any authorized participant to join the work, understand enough to contribute, and leave it easier for the next participant to continue.
Every improvement has to earn its place.
Three connected questions guide the work. The measure is the total burden of preparation, interruption, checking, correction and continuation.
- Coordination
- Does another agent or process add enough value to justify briefing it, checking its work and repairing mistakes?
- Shared mistakes
- Can we notice when a confident person and a capable agent are reinforcing the same faulty assumption?
- Justified trust
- Which checks give us grounds to rely on a result—and which only create reassuring paperwork?
Why we’re starting with continuity.
The larger vision needs a small, observable test: can useful understanding survive a change of participant?
Begin with actual historical work.
We audited 34 candidate cases and compared preparation methods on three. After workflow repairs, each method produced the strongest-rated output in a different case. Automatic handoffs were worth pursuing; there was no overall winner.
Remembering is only the beginning.
Later tests exposed preserved instructions that an agent did not act on, and small tasks that grew into extensive redesigns. Shorter answers alone did not reliably improve usefulness.
Involve people while they know the work.
Asking someone to supervise unfamiliar history created its own burden. That pushed us toward capturing decisions and corrections during a participant’s own work.
Start with a task
you already understand.
Today’s starter lets you save a checkpoint, begin a fresh Claude Code conversation and see what you still have to explain. It’s an explicitly triggered experiment you can try in your own project.
Try the continuity starterWhat we hope to learn next
With participants’ permission, capture selected moments from real work and run separate background comparisons: what survives, what gets lost, and when human input changes the result.
That platform is proposed. Today’s starter does not record in the background or automatically send files to b.akery. One successful handoff is not proof of general time savings.