Does real-time collaborative work (co-editing, pair work) still need people in the same session when AI can coordinate contributions asynchronously?

The shift

Merging divergent contributions, reconciling conflicts, and threading separate people's work into one coherent artifact goes from expensive human labor that had to happen in a shared session to something an agent does continuously in the background — it holds every contributor's version, resolves the conflicts, and keeps the whole coherent without anyone being present at the same time. What stays scarce is the shared real-time thinking that produces an idea neither person had alone, and the trust and tacit skill that only accrue while people are actually working together.

The axioms

• Two people producing at the same moment generate ideas neither reaches alone — synchronous co-thinking is where the non-obvious answer comes from, and it's scarce.
• Merging separate people's contributions and reconciling where they conflict is expensive, error-prone human labor, so it's cheaper to keep everyone in one session than to stitch their work together afterward.
• Working side-by-side is how tacit skill transfers — the junior person learns by watching the senior one make micro-decisions in real time, most of which never get written down.
• Trust and a working relationship accrue through shared real-time stakes — struggling through the same problem together is what makes people able to disagree bluntly and rely on each other later.
• Joint ownership of a shared artifact means both people are answerable for it, because both had their hands on it as it was made.
• The friction of real-time work — the interruption, the "wait, why did you do that," the immediate pushback — is what catches errors early and sharpens a rough idea while it's still cheap to change.

Invalid axioms

1. It's cheaper to keep everyone in one session than to stitch their separate work together afterward. The reconciliation cost was the reason to co-locate the work in time. An agent can now hold every contributor's version, detect where they conflict, propose a merge with the rationale, and keep a growing artifact coherent across people who never overlap — so the coordination that justified the shared session is close to free and continuous. Habit-trap: teams still book a two-hour co-working block to "get on the same page" on work an agent could have merged and surfaced the open conflicts on before anyone sat down.
2. Someone has to actively track and reconcile who changed what, or the versions drift apart. Conflict-tracking and version-reconciliation across contributors is now ambient rather than a discrete task someone owns. Habit-trap: still treating "integration" or "merge day" as a scheduled, dreaded piece of work with its own turnaround, when the merge is available the moment each contribution lands and the only remaining question is which conflicts are load-bearing.

Unchanged axioms

1. Synchronous co-thinking still produces ideas neither person reaches alone, and it's still scarce. The thing that happens when two people build on each other's half-formed thoughts in real time — one person's tangent triggering the other's insight — is a live, mutual act, not a set of contributions to be merged. An agent reconciling two finished drafts produces a coherent combination; it does not produce the third idea that only exists because two humans were thinking against each other at the same moment. Coordination got abundant; co-creation didn't.
2. Tacit skill still transfers by working alongside someone, not by reading the merged output. Most of what a junior person learns from a senior one is the micro-decisions — what got rejected and why, what wasn't worth doing, when to stop — almost none of which survives into the artifact an agent merges. An async pipeline can deliver a better combined result while quietly deleting the channel through which the next person learned to produce it at all.
3. Trust and the working relationship still accrue through shared real-time stakes. The kind of trust that lets two people disagree hard and still rely on each other comes from having been in the same hard moment together, not from an agent showing that their contributions were compatible. A clean async merge can make a group's output look aligned while none of the underlying relationship that makes future hard calls possible has been built.
4. A human still has to own whether the merged whole is right and worth shipping. An agent can reconcile contributions into something internally consistent; internal consistency is not correctness. Someone accountable still has to decide the combined artifact reflects reality and is worth shipping — and joint ownership only holds when the humans, not the agent that stitched their work together, are the ones answerable for it.

New axioms

1. Async AI-coordination can strip out the generative friction that made real-time work valuable, invisibly. The interruption, the immediate "wait, why," the objection raised while an idea is still cheap to change — that friction was doing work: catching errors early and sharpening thinking. When an agent smoothly merges finished contributions, the friction never happens, and nothing signals that it's missing. The output looks more coherent precisely because the productive conflict got engineered out.
2. Replacing pairing with async merge removes the mentorship channel without anyone deciding to. Pairing was never only about the code or doc produced — it was the main way tacit judgment passed to the next person. Route the work through async coordination and the artifact improves while the teaching quietly stops, and the cost shows up a cohort later as people who can operate the tools but never absorbed the judgment. Nobody chose this trade; it fell out of optimizing for throughput.
3. We now have to decide, per task, whether coordination or co-creation was the point — because AI made only the first one cheap. Some collaboration was mostly a merge problem wearing a meeting's clothes, and async is strictly better for it. Some was co-creation that happened to require coordination as overhead. The two look identical on a calendar, and the pull is to send both async because the coordination cost collapsed — leaving no forcing function to protect the synchronous work that was the actual source of value.

Where it breaks

Teams route collaboration to async merge because coordination is now effectively free (invalid) at the exact point where the thing that justified working in the same session was never the merge — it was the generative friction and the tacit transfer, both of which async structurally removes (new). The pipeline gets more efficient and the output more coherent, which reads as success, while the two things that actually made synchronous work worth its cost stop happening and nothing on the dashboard shows their absence. (This call hinges on model capability that's moving fast: as agents get better at surfacing conflicts and prompting for missing perspectives, they'll reconstruct more of the friction — but reconstructing friction as a prompt is not the same as two people thinking against each other live, and that gap is the part worth watching.)

ai-axioms.vercel.app/axioms/society/collaboration-realtime-coordination-ai-axioms