Faster Alone, Slower Together

On what AI is doing to teams, and what teams have to change.

The fastest person on your team may be slowing it down.

That beautiful, AI-generated deliverable sitting in the Teams chat wasn't reviewed. The forty dashboard variants you whipped out in the morning probably didn't save the team time how you had imagined. Maybe it cost the team a meeting and then another one, just to wade through them. Production used to be the constraint. The cost of an artifact was weighed and measured. It isn't anymore. Has your team noticed that the bottleneck has moved?

I used to take pride in being a “fast designer.” I could produce reasonably high quality artifacts more efficiently than others and that was a boon to me and my team. Now, that efficiency is table stakes for everyone.

When everyone is enabled to produce, the problem becomes about coordination. Solving it requires a shift in focus and skill.

The cost curve flipped, and the work changed shape

For most of the history of design and product work, producing an artifact was expensive. Exploration always had built-in limitations due to the time and cost. A wireframe may have taken hours. A prototype days. A research synthesis took a week. The cost of production set the rhythm of the work: you converged early because you couldn't afford to explore widely, and the artifacts you did make were taken seriously because each one represented real labor.

AI collapsed that cost to something near zero. Forty wireframes in an afternoon. A fifteen-page strategic memo in twenty minutes. Three working prototypes by lunch. The constraint that shaped how teams worked for decades has been lifted, and the work itself must change in response. Most teams haven't figured out what that means yet.

Now, in an AI-first exploratory phase, the work is increasingly subtractive rather than additive. The old question was whether you could draw enough cards from the deck to find a good one. The new question is whether you can discard them fast enough to find the right one. Production isn't where the value lies anymore. Discrimination is. Curation is. Direction is.

This is genuinely liberating. Strategy and design work can now consider possibilities that were previously too expensive to even sketch. We can build entire throwaway tools, custom one-use widgets, prototype-grade products that exist for a single conversation and then are placed in the trash can. Artifacts can be effervescent. Their value no longer lies in their permanence; we can throw them away now, and we should. Most of us haven't learned to do it without feeling guilty.

That said, the same dynamic that liberates the individual is breaking the team. While production got cheap, comprehension didn't. Reading a fifteen-page report still takes time. Synthesizing forty wireframes into a decision still takes a meeting. Aligning a group of people around a direction can still take months. The team's capacity to absorb, evaluate, and converge has not changed, yet it's being asked to absorb ten times as much.

That gap is where the new modes of failure live.

The new modes of failure

Visual verisimilitude. AI output looks finished. The visual signals that used to tell a team “hey, this is just a rough draft, please push back on it” versus “this is the real deal, react to the substance” have been flattened. Everything can arrive polished. Reviewers lose the cue that used to trigger critical reading, and the artifact gets approved on the strength of its appearance rather than its substance. The polish is communicating intent the author didn't consider, but the viewer can't tell the difference.

Fan-out without reconciliation. Every team member with AI access can now fork the team's working artifact in parallel and produce their own version of it. This feels like productivity. “Look how much we made!” But it mostly produces version control problems and reconciliation meetings. The team spends its time navigating an avalanche of variants instead of collaboratively producing one artifact it actually wants.

The pace-setter problem. The fastest producer on the team will set the team's consumption load. If you can generate six prototypes in an afternoon and the team has to walk through all six to decide, you have just spent the team's afternoon for them. Your speed has become their burden. This is the hardest one to see when you're the fast producer, because it feels like contribution.

Workshop confusion. Whiteboards, drafts, working sessions, and shared sketching weren't inefficient ways of producing artifacts. They were a technology for alignment that happened to produce artifacts as a byproduct. When teams replace the byproduct with AI-generated output, they get the artifact and lose the alignment. They are left to wonder why the team feels disconnected from work that, on paper, they all reviewed.

What must change for the individual

The shift required is from production to a curation of intent. From “what can I make?” to “what does this need to be in order to enable a decision?”

Start from the decision, not the artifact. Before producing anything, name the decision the artifact is meant to unlock and the person who has to make it. If you can't answer those two questions, the artifact will fail no matter how well you make it. This is, in theory, what we've always done. But how we do work now is not, and the undercurrent of change is not always perceptible at first. AI makes the failure mode much easier because you can produce the comprehensive artifact in the time it used to take to scope one.

Produce in bulk, present with restraint. Show one, hold many. The skill isn't generating forty variants — that's the easy part now. The skill is choosing which one to bring to the table, and holding the other thirty-nine in reserve as preloaded next steps. “If you like this direction, I have three iterations on it ready.” That uses production capacity in service of the reviewer rather than at their expense. It treats your speed as something you spend on the team, not a firehose of production you fire at them.

I'm actively living this one. On a project I'm working on now I can spin up prototypes and flows in a few hours, and my instinct is to bring it all to the table because the work is there, why not show it? “Aren't these all cool?!” But showing all of it doesn't accelerate the team. It transfers my afternoon's worth of cognitive production directly onto six others' capacity for consumption. I have multiplied the total cost of my production by six with no guarantee for a cohesive, reconciled path forward. The discipline I'm trying to build is producing widely and presenting narrowly. It feels like Superman working out at a Planet Fitness. It feels like restraint. It's actually respect.

What must change for the team

Individual discipline isn't enough though. Soon, when every team member is AI-augmented, the team itself has to develop new infrastructure for convergence. Old ways of working assumed production was a bottleneck, and now it isn't. Not in the same way at least.

Designate a synthesizer. Someone has to know the truth. Someone has to be explicitly responsible for reconciliation: pulling parallel forks into a shared artifact, identifying agreement and divergence, surfacing decisions, and serving as the integrity check against AI drift. When the hallucinated fact, the reframed conclusion, the nuance which doesn't come through in the AI summary isn't caught, it degrades the team over time. This is probably not the same person as the highest-producer, and it is probably not a hat someone wears. It requires a real, dedicated person, and on AI-augmented teams it might soon be the most important one.

Run convergence rituals on a cadence. Most teams have well-developed practices for divergence like brainstorm sessions, design sprints, and exploration phases, but almost none for convergence. AI massively amplifies the divergence side, which makes the absence of convergence practices catastrophic. We must treat convergence reviews the way we treat sprint reviews: regular, scheduled, non-optional. The artifacts that come out of them — the validated, signed-off, or team-aligned outputs — become the canonical inputs for the next round of work. They get dedicated space in the project repo. They are what gets fed back into the AI to prevent drift. They are the team's grounding context documents, the cores of truth that everything else builds on.

Set artifact budgets. Put a limit on what everyone can bring to the table. Cap what gets brought to a review. Number of variants, page lengths, time limits. Not because more isn't possible but because the team's job is to decide, and excess input is now the noise. Consumption capacity is a real budget. Don't burn it on slop.

Keep a graveyard. When the team kills a direction, write down briefly what it was and why it was killed. This is not a wall of shame. It is a negative space which outlines the work: a horse is everything it is not. The explored, and then rejected, territory is important knowledge, and it solves a real problem on AI-augmented teams; the same dead ends can keep getting rediscovered in parallel because individuals can't see each other's discards. The graveyard makes discard a productive activity rather than waste.

None of this is about slowing down, or how teams are going away. They're not, and the point is the opposite. The teams that get this right will not produce less, they will produce more while actually making meaningful decisions, staying aligned, and not drowning in their own output. Velocity is enhanced by spending it on the right things.

Principles for AI-first work

Here's what I'm trying to keep in mind now:

The teams that win in this phase will not be the ones producing the most. They will be the ones discarding the most, while keeping shared understanding intact. That is hard. But I also think it's more interesting work than what we were doing before.