Thoughts · September 6, 2026
Everyone codes alone, then calls it collaboration
Arguing about a plan once the work is built is arguing after the concrete has set, and by then changing it costs real money. Have the same argument while the plan is still soft and the review at the end turns into a formality, because everyone reviewing it already helped shape it.
The usual pattern is late. People — and now bots — work in separate sessions. Each one produces code, notes, a branch. Only after those sessions end does the real collaboration start: review the diffs, argue the design, integrate the pieces, improve what landed.
That late meeting is expensive in the same way late testing is expensive. By the time the work has hardened, changing it costs more than shaping it would have.
Larry Smith named the testing version shift-left in 2001: catch problems earlier on the timeline because later discovery is dearer. We already used that geometry for secrets — catch the leak at commit, before history owns it. Collaboration has a geometry too.
What moves left
With an agent workplace, the collaboration can start in the chat itself. Several named teammates and a person steer in one place: argue the approach, assign desks, leave memory, watch the shared computer, refuse a bad plan before it becomes a thousand lines of code.
The code sessions still happen. What changes is when the shared judgment shows up. Intent is still soft. The review at the end becomes confirmation of something the people in the chat already shaped, because they met the work while it was still forming.
That is shift-left on collaboration: move the human meeting earlier, into the sessions where the work is being made.
What this is not
It is not "put a Slack channel next to Cursor." A channel without a durable workplace is still late collaboration with more noise — people dump updates after their private sessions and hope someone reads them.
Chatticus is built for the early meeting: named teammates with beats, a shared computer, memory that outlasts a tab, and a roster that stays named. The chat is where the farm works, not a side channel for status.
It is also not dumping review onto whoever typed first. Smith wanted QA embedded earlier, rather than developers absorbing the whole job. Same risk here: early collaboration fails if the room has no desks, no memory, and no way to say whose job a thing was.
Why it matters now
Coding got cheap enough that a single session can produce a product-shaped pile overnight. When generation is that fast, waiting until merge to collaborate is waiting until the pile is already large. Review, security, and what to keep or cut only work if someone can still steer while the work is forming.
Moving the collaboration earlier is how a farm stays a farm, instead of becoming a pile of solitary coding sessions that meet too late.