What the Friction Was Hiding

Eighteen months in, the company feels different. Good different.

The product teams have stopped fighting about what to build. That grinding quarterly prioritization process, the one where every meeting ended with someone’s roadmap item cut and someone else’s ego bruised, has loosened its grip. AI has given the teams room to dream bigger. What used to be a painful conversation about tradeoffs has become a conversation about ambition. The PMs are energized. The engineers are shipping. The backlog that used to feel like a weight on everyone’s chest is finally moving.

The things that used to kill momentum are gone.

That custom integration a mid-market customer had been requesting for two years? The one the CS team had been diplomatically deflecting because it would have taken six weeks of engineering time nobody had? Done in a sprint. The onboarding experiment the growth team had been wanting to run since last year, the one that kept getting bumped by higher-priority work, shipped in three weeks. The enterprise admin features that used to require painful scoping conversations and hard tradeoffs now just get built. The answer to customer requests has shifted from “we’ll put it on the roadmap” to “we can do that.”

Each team is doing exactly what it was hired to do. The growth team is growing. The platform team is enabling. The enterprise team is retaining. Every KR is green.

The signal that something is wrong doesn’t arrive cleanly.

It starts small, in places easy to dismiss. A support ticket about two features that work correctly in isolation but behave unexpectedly when used together. A CS rep flagging that a customer’s onboarding flow, customized six months ago, now conflicts with a new default that the growth team shipped last quarter. A sales engineer on an enterprise call who can’t cleanly answer how the admin tool maps to the buyer’s existing workflow, because three teams have extended it in three different directions.

Each signal has an explanation. Each explanation is reasonable. None of them surface together in a single dashboard.

Then the volume picks up. Support tickets aren’t spiking in one category, they’re rising across all of them, a slow increasing trend that doesn’t trigger any alert. QBR conversations with mid-market customers start taking longer, not because the customers are unhappy with any specific feature, but because they can’t quite articulate what the product is for anymore. The sales team starts losing deals they expected to win, not on price, not on features, but on something harder to name. The product feels complicated in a way that’s difficult to explain and impossible to demo around.

Six months later, the lagging metrics arrive. Activation rates slip. Expansion revenue stalls. Net revenue retention starts its slow move in the wrong direction. A few enterprise renewals wobble. Margin tightens as support load climbs and CS time per account increases.

The post-mortem will search for the decision that caused this. It won’t find one.

Every team made reasonable choices. Every initiative traced back to a real customer signal. Every feature shipped was something someone asked for. The incentive system worked exactly as designed.

That’s the problem.

The friction was doing the work

Here is what nobody planned for.

The friction was doing real work.

Those grinding prioritization battles weren’t just organizational overhead. The six-week scoping conversations weren’t just inefficiency. The painful tradeoffs about what not to build, the ones that left someone’s roadmap item on the cutting room floor every quarter, were where something crucial was happening. Teams were being forced to negotiate shared intent. To argue about what the product was for. To make the implicit explicit.

Not because anyone designed it that way. But because they had no choice.

The time to build it was the constraint. And constraint forced coherence.

Coherence is not the same thing as alignment. Alignment is about direction and how well teams stay true to that direction. Coherence is about integrity, the product feeling like one thing, the strategy holding together under pressure, the customer experiencing a company that knows what it is. You can have perfectly aligned teams producing an incoherent result. I have seen it happen exactly this way.

When you can only ship three things this quarter, you have to agree on which three things. That agreement, though messy, at times political, and sometimes arbitrary, was the organization’s coherence mechanism. The debates weren’t just a tax on execution. They were the process by which strategy became shared. By the time something shipped, it had been argued over, scoped down, negotiated across functions. The friction was frustrating. It was also doing the cutting.

AI didn’t just speed up execution. It took away the scissors.

When teams can ship ten things instead of three, the negotiation doesn’t scale with the output. There’s no forcing function to make intent explicit.

Consider how most product organizations are structured. The growth team is rewarded for activation. The platform team for stability. The enterprise team for retention. These are not the same objectives and they were never meant to be. Specialization was the point. In a world where execution was slow, those focused incentives made each team sharper. The friction at the boundaries was what kept three different objective functions from pulling the product apart. Remove the friction, and the specialization that made each team effective becomes the mechanism that fragments the whole.

This is the inversion most organizations miss. They experience the removal of friction as pure gain. Faster shipping, more experiments, happier teams, better metrics. And it is gain, for a while. What they don’t see is everything that the friction was containing.

Speed doesn’t cause drift. It reveals the drift that was always there.

AI didn’t create this. It was present long before the first model was deployed. It lived in the gaps between team objectives, in the interpretive space between strategy and execution, in the distance between what leadership intended and what incentives actually rewarded. Friction kept it from expressing itself. Remove the friction, and everything it was containing becomes operational.

This is why the post-mortem finds no bad decisions. The decisions were fine. The coherence mechanism was gone.

The cost that doesn’t show up in the sprint

When execution gets cheap, the natural response is to do more.

More experiments. More features. More integrations. More variants. The growth team runs five onboarding experiments instead of one. The platform team enables integrations that would have required a painful scoping conversation six months ago. The enterprise team builds the admin features that customers have been requesting for two years. None of it requires the negotiation it used to. The blockers are gone. The answer keeps being yes.

Nobody is misbehaving. This is the system working.

At some point a customer opens the product and doesn’t know where to start. Not because they haven’t been trained, but because the product has grown in too many directions at once. Every team added something real. Every addition made sense in isolation. But the customer doesn’t experience features in isolation. They experience the whole thing.

That accumulated complexity has a name: surface area. Everything a customer has to understand, navigate, and hold in their head to use your product. Every feature adds to it. Every integration extends it. Every variant multiplies it. It doesn’t compound linearly, each addition interacts with everything already there. A product with ten features isn’t twice as complex as a product with five. It’s the ten features plus every relationship between them.

Building a feature is a one-time cost. Maintaining its consistency with everything else is permanent. Every addition creates a new surface that has to be kept coherent; with the UI, with the documentation, with the support team’s mental model, with the customer’s understanding of what the product does. The conversations that used to happen before a feature got approved, the painful scoping debates, the integrations that got declined were, in their own friction-laden way, forcing someone to ask: do we really want to carry this indefinitely? The friction wasn’t just slowing things down. It was extracting a commitment.

Now those conversations don’t happen. The integrations get built. The features ship. Each individual addition is completely justifiable; a real customer signal, a real metric, a real business reason. Nobody is building the wrong thing. Everyone is building the right thing for their part of the product.

Here is what makes this structurally dangerous: most organizations reward exactly this behavior. Shipping is visible. Experiments are countable. KRs tied to feature output, experiment volume, and activation metrics all point in the same direction; build more, ship faster, hit the number. The incentive structure was designed for a world where execution was the constraint. In that world, rewarding output made sense. Output was hard to produce.

That world is gone. The incentive structure hasn’t noticed.

What it never accounted for, because it never had to, is the coherence cost of each addition. The teams aren’t drifting because they’re misaligned. They’re drifting because they’re aligned to objectives that were designed for a slower world, that never needed to account for coherence because friction was accounting for it automatically. The fragmentation isn’t a failure of discipline. It’s a rational response to a system that was never updated to reflect the new constraint.

The constraint is no longer execution. It’s coherence. And most organizations are still building for a world that no longer exists.

The bottleneck nobody planned for

There is a trap that comes with cheap execution that almost nobody sees coming.

When AI makes building easier, the natural instinct for a leader is to stay close to it. The things that used to require delegation, the prototype, the analysis, the draft business case, are now accessible. A leader who couldn’t reasonably spend time in the details six months ago can now produce sophisticated output in an afternoon. The “could do” bucket has exploded. And it feels productive. It feels like leverage.

It isn’t.

Leadership time doesn’t scale while everything else in the organization just got faster. Coherence requires sustained leadership attention. Not occasional attention. Not attention when the metrics move. Sustained, deliberate, continuous attention to what the organization is building toward and whether everything in motion is still pointing there.

Every hour a leader spends in execution, however valuable that execution feels, is an hour not spent on the thing only they can do.

This is not a time management problem. It is a structural one. AI has simultaneously made execution more accessible to leaders and made coherence more dependent on them. The pull toward execution is stronger than it has ever been. The cost of following that pull is higher than it has ever been. Both things are true at the same time.

The leaders who feel this most acutely are often the best ones. The ones who care about quality, who know the product deeply, who could in theory contribute meaningfully to almost any decision in their organization. In a slower era that instinct was manageable, there wasn’t enough execution capacity to act on it constantly. Now there is. That constraint is gone.

What only a leader can do has not changed. Setting constraints. Maintaining coherence across teams moving at different speeds in different directions. Resolving the ambiguities that require someone who can see the next eighteen months, the competitive horizon, and the organization’s current reality at the same time. Noticing when the surface area is expanding faster than the shared intent that should be governing it.

That work does not show up in a sprint. It does not produce a visible output by Friday. This is the work that determines whether everything compounds into a product that stands for something, or fragments into something that merely looks productive.

In a cheap-execution world, one job matters above all others: holding the organization together while everything accelerates. That is the leader’s job. And it must never be delegated.

Two Operating Systems, One org chart

Most organizations don’t fragment all at once. They fragment unevenly.

Some teams, typically the ones recently rebuilt around AI-accelerated ways of working, are moving fast. They iterate quickly, ship constantly, and make decisions at a pace the rest of the organization can’t match. Their scope is clear, their tools are sharp, and their output is visible.

They look like exactly what the future of work is supposed to look like.

Other teams, call them the load-bearing teams, are carrying a different kind of weight. Not because they are less sophisticated, but because they operate under fundamentally different constraints. Compliance structures. Established SOPs. Cross-functional commitments forged over years of negotiation. Customer agreements that predate the current roadmap. Governance frameworks that exist because someone, at some point, learned a hard lesson. These teams are not slow because they haven’t modernized. They are slow because they are holding the institutional load that keeps the company functioning.

Both are doing exactly what they should be doing.

The problem is what happens at the boundary between them.

The fast-moving team looking to ship a feature that touches customer data hits a compliance review that wasn’t built for 48-hour releases. The load-bearing team flags a security sign-off that the fast-moving team didn’t know was required. Neither side is wrong. They are operating under different physics, inside the same org chart.

This is not a culture clash. It is a structural incompatibility. And it is where the coherence gap becomes most visible; not in the metrics, not in the dashboards, but in the friction between teams that are supposed to be building towards the same thing.

What makes this particularly difficult is that the incompatibility compounds without announcing itself. The fast-moving teams keep moving. The load-bearing teams keep holding. The distance between them grows. And at some point the organization is not running one operating system with some variation. It is running two, simultaneously, with no mechanism to reconcile them.

This is where the instinctive response reveals its own limitation. Some organizations embed a senior leader close to the fast-moving teams; a sponsor who can hold coherence by proximity, translate between the two worlds, and make the calls that the org chart can’t. At the scale of a handful of teams, it works. But it is a person standing in for a mechanism. It doesn’t scale, was never designed to, and offers no answer for the ten other teams moving just as fast in different directions.

This is not a new problem. Companies like Amazon, Google, and Meta had to confront it directly, not as a future risk but as an operational reality. At their scale, coherence could not be left to chance. I saw this from the inside of two of those companies. They had to make it explicit, deliberately, and at real cost.

Of course, most organizations are not at that scale. But AI is compressing the timeline. What took hyperscalers a decade to encounter at ten thousand people, a mid-sized company can encounter in eighteen months at two hundred people.

The fragmentation builds in the background, across rational decisions, until it’s too distributed to trace. By then, the question is no longer how to prevent it. It’s how much it will cost to reverse it.

It shows up first in places easy to dismiss. A support ticket that doesn’t fit a clean category. A CS rep spending an extra thirty minutes untangling something that should have been straightforward. A sales engineer on an enterprise call who can’t quite explain how two parts of the product work together, not because they don’t know the product, but because the product no longer has a clean answer.

And then, eventually, it shows up in a QBR. A customer who has been using the product for three years, who relies on it every day, who cannot articulate what it is for anymore. Not because they’ve lost faith. Because the product stopped telling a coherent story and they absorbed that confusion without knowing where it came from.

By the time that conversation happens, the fragmentation is already baked in. It lives across dozens of reasonable decisions, hundreds of justified features, thousands of individual moments where the incentive system worked exactly as designed. There is no single decision to reverse. There is no team to blame. There is only the accumulated distance between what the organization intended and what it actually built.

Here is what made that distance possible.

The friction that used to force coherence accidentally is gone, removed by the same acceleration that made the teams feel so productive. The incentive structures kept rewarding shipping. The surface area kept expanding. Leadership attention kept drifting toward execution. And two operating tempos kept pulling apart inside the same org chart. Nobody designed any of it. Nobody noticed until the distance became impossible to close.

None of it was planned. None of it was anyone’s fault. The mechanisms that used to prevent it were never designed, they were inherited. And that inheritance is gone.

What replaces it, how organizations build and maintain coherence intentionally at the speed the AI era demands, is the question that matters most. And it is one I am digging into next.

The ones who recognize it early enough will have a choice. The ones who don’t will find it in a QBR.