AI Gives You the Ability to Run Ten Things at Once. That's Not Always a Good Idea.

AI gives you the ability to run ten things at once. The temptation is to do exactly that.

More tools, more threads, more agents firing in the background. More progress, theoretically, all at once.

But cognitive load doesn't scale with capability. And that's the part most people are not accounting for.

The WIP problem has not gone away

I've spent years getting product and engineering teams to limit work in progress. Not because ambition is wrong, but because context switching kills completion.

When too much is in flight, things pile up. Quality drops. The board fills with activity but nothing actually ships. You're busy in a way that produces very little.

The principle behind WIP limits is simple: finishing matters more than starting. A smaller number of focused threads, completed well, beats a large number of half-finished ones every time.

That principle does not stop applying just because AI is doing more of the work.

Directing AI is still cognitive work

When you run multiple AI threads in parallel, you are still the person responsible for qualifying the outputs. Reading them, assessing them, deciding what to do with them, feeding the next prompt.

That work has a cost. Run too many threads and you end up with half-finished outputs, lost context, and no clear picture of what you were trying to achieve in the first place. The AI moved fast. You didn't keep up.

The failure mode looks exactly like an overloaded sprint. Lots of motion. Not much progress.

Agents make it harder to see the problem

If anything, agents amplify this. With a standard AI thread, you're in the loop at each step. You can see what's happening and course-correct.

Agents can move faster than your ability to qualify what they've done. By the time you're reviewing outputs, several decisions have already been made on your behalf. Dependencies have been missed. Outputs have drifted. And you've lost the thread of what you were actually trying to achieve.

The WIP problem doesn't disappear with agents. It just becomes harder to spot until you're already in it.

The right frame

The goal is not to maximise how much you have in flight. It is to shorten cycle time on the things that matter.

Use AI to move faster through individual pieces of work. Finish them. Then move on. That is how you end up doing more, without the overhead of managing a sprawling set of half-finished threads that only you can untangle.

More done. Just not all at once.