Your teams and agents can generate change faster than anyone can steer it. A denial of service from your own people

Linear CEO Karri Saarinen just named a problem most engineering leaders feel but rarely say out loud. Agents and teams can now produce change faster than any leader can understand it, and leadership is quietly losing the ability to steer. Here is what he describes, and what it takes to steer an engineering organization in the AI SDLC.

A denial of service, from your own people

In his essay, Saarinen describes how hard it has become to live up to a principle he has written about before: you can outsource work, but not your understanding.

Nobody is attacking the company. Everyone is doing work they believe is useful. But all of it draws on the same limited capacity to understand, review and decide, and that capacity has not grown at the speed of generation. Saarinen compares it to a car speeding up until it becomes impossible to steer. Leaders never saw every piece of work, but the work can now change the direction of the company faster than leadership can respond.

He is careful to say this is not about people making poor decisions. Linear invests heavily in hiring and in an environment built for good judgment. Even so, he describes unclear or shifting direction and overlapping or conflicting features. Something can go from idea to working implementation before anyone outside the immediate team knows it exists.

“It already exists” is not a reason to ship it

When something takes twenty minutes to build, asking to discuss it at tomorrow’s meeting starts to feel unreasonable. Saying “looks good, let’s go” is easier. Saarinen’s answer is the core of the essay: the fact that work exists does not tell you whether it is the right thing to do. We have to stop treating output as proof of value. A working implementation proves something can be built, not that it belongs in the product, and some changes have negative value.

But the opposite extreme fails too. You cannot hold an alignment meeting for every one of the 1,000 pull requests a team might open in a day. Leaders are stuck between rubber-stamping everything and becoming the bottleneck.

Two moves, and one gap

That gap is the heart of the problem. Intent and gates are inputs. Steering also needs feedback: a continuous, honest picture of where the work is actually going. And it is not just a CEO problem. Saarinen points out that every leader whose judgment you rely on is also trying to keep up with everyone else.

You cannot steer what you cannot see

Every steering system has the same shape: a direction, an engine, and instruments that show where you are actually heading. Most engineering organizations upgraded the engine with agents and kept the old instruments.

The aggregate view is what turns a pile of individually reasonable changes into something a leader can steer. In practice, it answers questions like these every week, without reading 1,000 pull requests:

The answer is not more surveillance of people

Saarinen makes a point worth repeating: he trusts his people, and the problem is not bad decisions. That matters for how leaders respond. The surge of change is driven by agent throughput, not by people suddenly working differently. Watching engineers more closely will not restore steering. Understanding what the agents are producing, where it goes and whether it holds up, will.

That is the direction we have taken at Waydev: measure AI agents, not people. Which agents generate the most change, how much of it lands, how much creates review load or rework, and whether it moves the bets the company actually made. People stay accountable for direction and judgment. The instruments watch the engine.

Make agents check their work against your intent

Saarinen’s first move, writing intent down so people and agents can check their own work against it, only works if agents can actually read that intent alongside how your organization builds software: ownership, standards, what is in flight, and what broke last time. For most enterprises, that context is too sensitive to send to a public model. So it either never reaches the agents, or it leaks.

A weekly steering check for engineering leaders

Responsibility does not scale down with speed

Saarinen ends on a point every engineering leader should hold onto: stepping back does not remove responsibility for where it all leads. Each team can make reasonable decisions, and leaders still own what those decisions add up to. In the AI SDLC, the way to keep that responsibility without becoming the bottleneck is not more meetings. It is better instruments, clearer intent, and agents that work inside both.