A friend brought up The Phoenix Project to me recently. I recommended it to him years ago, then mostly stopped thinking about it. His timing turned out to be excellent.

The book follows a software organization drowning in work. Requests arrive faster than the teams can finish them, dependencies pile up, and the one person who understands the system is stretched across every initiative. They keep starting new things and wonder why nothing finishes.

The way out is not heroics or working harder. It is boring, counterintuitive stuff: make work visible, protect your constraint, and above all, limit your work in progress.

This spring, somewhere between my seventh terminal tab and my laptop fan doing its best jet engine impression, it hit me:

The Phoenix Project has been resurrected. Again. And this time, we are all Bill Palmer.

Managers now, all of us

Once you start using AI agents for real work, part of your job turns into management.

You're still engineering, but you're also delegating, checking in, reviewing, unblocking, and every now and then discovering that your direct report has confidently done something completely different from what you asked for.

Like every first-time manager in history, my instinct was to take on more. Way more.

This spring and summer I've been thoroughly consumed by agents. Five at a time. Seven. On a good, or possibly bad, day, ten. I ground my CPU and memory into paste and, if I'm honest, did roughly the same to my own mental scheduler.

Some of the time it feels amazing. Like running a tiny software company whose entire staff lives in terminal tabs. You fire off a refactor in one window and a research task in another. For a glorious twenty minutes everything hums and you feel like you've been handed superpowers.

Plenty got done. The problem was the kind of work I had started choosing. Small, bounded tasks fit the system I had built because I could hand them to an agent and keep them loosely in my head alongside six others. Mostly refactors and bits of plumbing. Useful, but less ambitious than what these tools should make possible.

AI should let one person take on work that used to feel too large. But ambitious work needs a coherent direction and enough attention to notice when the implementation drifts. The faster the agents get, the more focused I need to become.

Meanwhile, I was still jumping between half-finished conversations, reviewing code I barely remembered asking for, and trying to recall which branch was stacked on which.

Anyone who has read the book will recognize this immediately. It is Parts Unlimited all over again, except the unplanned work arrives through a terminal, and Brent is me.

At one point I found myself coding with AI on a boat in the open ocean between Flores and Bali, around Komodo National Park. Laptop open, sea on every side, absolutely delighted with myself. It was genuinely great. It was also, in retrospect, a symptom.

Komodo with agents. Santorini before them. Apparently the symptom predates the diagnosis.

The case for a personal WIP limit

In the book, the turning point comes when they stop starting and start finishing. Work in progress is the silent killer. Every extra item in flight makes everything slower. The tasks are only part of the cost. Switching between them is what gets you.

Agents do not repeal that rule.

So I'm trying an intentional WIP limit of two agents at a time. Three, max.

I can run more. My electricity bill can confirm. But two or three is roughly the point where I can still hold the full context in my head, review the work properly, and choose what matters instead of inventing tasks to keep the machines busy.

More finished work would be nice. The bigger benefit is leaving room for the larger project I might otherwise avoid because ten smaller tasks are easier to juggle.

I'm trying to make this less dependent on willpower. A little system I've been building, Loopkeeper, reads roughly the last 800 meaningful characters from every agent thread from the previous two weeks. That cheap first pass tells it what might still be unfinished and which threads are worth reading in full.

It also counts the mess I leave behind. Three linked worktrees in one repository trigger a warning. At five, it blocks another from being created. Dirty state older than 48 hours, unpushed commits and a Git operation left hanging all get dragged back into view.

That is what making work visible looks like in practice. Sometimes the most useful agent simply asks why I have five versions of the same job open.

When two or three long-running tasks are churning away, the hard part is resisting the hunt for a fourth. I'm going to try to sit there. Twiddle my thumbs. Do some pushups. Maybe even go outside without bringing the laptop.

The token-burning Olympics

The age of AI was supposed to give us abundance and freedom. Instead, I think I'm more locked to the screen than ever.

In my circles I've started noticing something like sportsmanship around it. How many tokens did you burn this week? How many agents do you have running? And the ultimate flex: how much did AI do for you while you slept?

Some overnight runs are genuinely useful. I'm not trying to be a Luddite here. But I want to make the case for the opposite move: don't go hunting for tasks to feed the machine before bed just because you can. If there's no natural task, there's no natural task.

Have the agent write a handoff summarizing where things stand, then close the lid.

Humans need sleep. Sometimes computers need to be restarted so memory can be released. Agent sessions need the same reset. A new thread is a new day.

In my experience, the handoff is worth more than the overnight run anyway. Future you gets a clean briefing instead of waking up to 40,000 lines of code to untangle before coffee.

Pro tip, and this one is free: START NEW THREADS. More often than you think. Way more often. And for the love of God, don't rely on auto-compacting to carry you through a marathon session.

A compacted thread is like working with a colleague with amnesia. They remember fragments and sound confident, then casually reintroduce the bug you fixed together two hours ago.

A fresh thread with a tight brief beats a compacted marathon every time.

Stop starting, start finishing

I want to get better at using these tools, not use them less. I think the people who get the most out of them over the next few years will be the ones who manage them well. Management is an old discipline with old lessons. It's just funny that a thirteen-year-old novel about a fictional auto parts company turned out to contain the operating manual.

I'll report back on how it goes. Knowing myself, there's a decent chance you'll find me in a month with nine terminals open, insisting this time is different. But hey, Bill Palmer didn't fix Parts Unlimited on the first try either.

Enjoy the rest of the summer, everyone. I certainly will. Mexico next week. The agents are coming with me. So is the WIP limit.