Peter Bean

Creator of | Agentic AI founder | DeFi native builder | Nerd | Proud 🇨🇦 | Live Long and Prosper 🖖🏽

[Peter Bean @Beannation]

I built this because my own Claude Code spend didn't match what my tracker said. Turns out it reads session logs, which don't carry real cost. So I made one that reads the statusline — the source behind the bill. Run wtclaude compare and tell me your gap. Mine was not small.

[WTClaude @getwtclaude]

Your Claude Code cost tracker is probably wrong.

Most read session logs. Logs don't carry billing-grade cost — so the number drifts from your actual bill.

WTClaude reads the statusline instead: the same source behind your bill.

It's free + open source. Here's what it does 🧵

[Peter Bean @Beannation]

Halloween Night. Decorations ready, fog machines tested, pumpkins carved, and the web app for our Haunt is live.

Posting pictures all day to the app as we get it started.

https://t.co/fKw05Yv7jI

[Peter Bean @Beannation]

@boardyai Can you find the Easter eggs?

[Peter Bean @Beannation]

For years I’ve done a massive Halloween display for the neighbourhood kids. It now expands 3 front lawns.

This year I vibe coded a web app for the display, with photo sharing, a pumpkin gallery, a scavenger hunt and more.

Check out photos from every year!

https://t.co/P53Wnsdm06

[Peter Bean @Beannation]

Where I have landed, for now.

Within a few years most people who lead anything will be leading a mix of people and digital workers, and almost nobody is being trained for the second half of that. The training that exists is about tools. The skill that matters is older. Say what you want clearly. Decide what you will and will not delegate. Check the work. Be able to explain the call afterwards.

Every one of those is a management skill from long before any of this. We just got away with doing them loosely, because people are forgiving and proximity covered the gaps.

I am more optimistic about it than the general mood and have been for a while. The work handed over first is overwhelmingly the work nobody wanted. What is left behind is much closer to the reason most people got into their field.

The version that goes badly is not the one where machines do too much. It is the one where we scale past what we can see and call it progress.

[Peter Bean @Beannation]

Something I have had to watch in myself this year.

The machine does not get tired. I do.

When the work around you never slows down, there is a quiet pull to match it. Nothing is waiting on anyone else, so there is always a next thing available. Ten at night. Sunday. The twenty minutes before dinner. The friction that used to end my working day was mostly other people's availability, and it turns out that friction was doing something useful.

I stepped away from work for five months once, to care for someone in my family and look after my own health. I came back more capable than I had ever been. What I took from it was that I had been running on a definition of productivity that could not tell the difference between working hard and being unwell.

So there are rules now and they are boring on purpose. The work has hours. Some things wait until tomorrow even though nothing requires them to.

When capacity stops being the limit, what stops you?

[Peter Bean @Beannation]

If I were hiring today for someone who will work alongside a digital workforce, I would test for things that are not on most job descriptions.

How well they write instructions. Give them a task they know cold and ask them to hand it off in writing to someone who has never done it. Most people are worse at this than they think. It is learnable and it is now load-bearing.

Whether they check. The specific habit of going back to look at something they already approved.

What they do when an answer looks right. The failure mode of this era is not being fooled by obviously bad work. It is accepting good-looking work that happens to be wrong, because checking takes effort and it looked fine.

Which tools they have used matters far less. Those change every few months. The habits underneath them do not.

[Peter Bean @Beannation]

People ask where the line is. Mine, as honestly as I can draw it.

Nothing where a person's feelings are involved. Not a hard conversation, not feedback, not a message to someone having a bad week. The words might come out fine. That is not the point of them.

Nothing I would have to defend if it went wrong. If I am the one explaining it to a customer, I make it.

Not the first look at anything that matters. I want my own view before I see an assembled one. It is very easy to read a tidy summary and mistake agreement for thinking.

The reading stays mine as well. Not all of it, not always closely, but enough that I would notice if something were off.

Everything else is fair game, and there is a great deal of everything else.

The loudest people on this tend to sit at one extreme or the other. All of it, or none of it. The useful position is neither, and getting to it takes longer than picking a side does.

[Peter Bean @Beannation]

Three years ago my best days ended with me finally finding the answer.

Start with a question, spend the day hunting. Numbers out of four systems. Waiting on a team that owed me a piece. Reading around the problem until I had an opinion. By the time I could decide, the day was gone and the deciding took four minutes.

That ratio has flipped.

My time now goes to taking in information that has already been gathered and making calls on it. What arrives is assembled. What I add is judgment.

The part I did not expect is what it did to my thinking. When you get four minutes to decide, you decide with whatever you happen to have in your hands. When deciding is the whole job, you start noticing how often you were guessing and calling it experience.

If your calendar is mostly gathering, that is worth a look on its own terms, whether or not you ever hand any of it over.

[Peter Bean @Beannation]

How do you know you know what you need to know?

With a person you have a hundred quiet instruments. You hear it in their voice on a call. You notice they have gone quiet in a meeting. Someone mentions something in passing. You get a feeling about a project a week before there is any evidence for it, and you are right more often than you should be.

None of that exists on the other side. A digital worker will never seem off. It will not hesitate, or hedge, or come to you on a Friday afternoon and say it isn't sure about something. You get the output, and whatever record you insisted on keeping.

So the only handle you have is the one you built on purpose. Can you see what it did. Can you trace why. Could you explain that decision out loud to a board, an auditor, or a customer, without flinching.

Pick a project you feel confident about. Ask how you would know, today, if it had quietly gone wrong. If the answer is that someone would probably tell you, that is not a system. That is luck with good manners.

[Peter Bean @Beannation]

I built the measurement before I scaled any of it, and it is the decision I would defend hardest.

The reason is simple. I do not trust my own read on whether output is good, and I do not think anyone should trust theirs. You look at work you commissioned, it matches what you asked for, and your brain files it as correct. That happens whether a person made it or a machine did.

So there is a loop. Everything that goes out gets measured, and something reads those numbers back to me every week with an opinion attached.

It has told me things I would never have found by reading. The clearest one: when I chained related posts together, the way most people do, the follow-ups landed with roughly a tenth of the reach of the same idea standing on its own. I had no instinct about that at all. The format was quietly eating the work while the writing itself was fine.

Everyone wants to know how much they can hand over. The honest answer is that it is capped by how much you can actually see, and past that line you are not delegating. You are hoping.

[Peter Bean @Beannation]

Early on I gave one of my functions far too much to do.

It seemed sensible at the time. One place handling everything to do with getting the word out. Research, writing, scheduling, measuring, all of it. If you have ever built a small team you know exactly why that felt efficient.

The work came back plausible and slightly wrong. Nothing you could point at and call an error. It kept making small judgment calls in a direction I would not have chosen, because a job defined that broadly has no edges, and without edges there is nothing to be right or wrong against.

So I split it. Research on its own. Writing on its own, with its own rules about what my voice is. Scheduling on its own, and it is the only one allowed to publish anything. Measuring sits outside all of them and reports on the lot.

Each does a single job. Each knows what it hands off and to whom.

The reason that worked has almost nothing to do with AI. Narrow roles are easier to do well, easier to check, and easier to hold to a standard. We have known that about people for as long as we have had organizations.

The role you cannot define in one clean sentence is usually the one quietly underperforming, and it is almost never the person's fault.

[Peter Bean @Beannation]

My businesses run on files now.

What is true right now. What is still open. What was decided, when, and who approved it. Written down in the open and updated as the work happens, rather than living in my head or buried in a thread I could scroll back through if I had an hour.

The test I hold it to: if I walked away and someone new sat down cold, could they read four documents and start working without asking me a single question?

For a long time the answer was no, and the reason was never the technology. I was the only place some of the context lived.

I have watched that pattern in every company I have been part of. The operating knowledge sits with three people, everyone routes through them, and eventually those three get tired and leave with the manual still in their heads.

Writing it down fixes that for humans too. It just took handing work to something with no memory of last week to make me finally do it properly.

Where does your team's real operating knowledge live?

[Peter Bean @Beannation]

The first thing I read most mornings is a page of decisions that something else wrote for me overnight.

It arrives as a list rather than a report. Four or five items, each small enough to answer in a sentence. Approve this. Hold that. This account has gone quiet, pick a lane.

I answer them and the day's work goes out.

Two years of building to get here, and I want to be plain about what it actually is, because the language around this makes it sound stranger than it is. I run several businesses. Each function inside them is defined the way you would define a role. What it owns. What it must never do. Who approves what. Where it writes things down. Then something runs that role.

The surprise was how much of the work was writing. Not code. Definitions. Deciding what a job actually is, precisely enough to hand to someone who has never met you.

Could you do that for one job on your team today?

[Peter Bean @Beannation]

If you can't see what your agents did and why — if you can't defend a decision after the fact — you're not managing them. You're hoping. Observability is the whole job.

[Peter Bean @Beannation]

Here's the leadership shift nobody's preparing for.
Soon you won't just manage people. You'll manage a blend of people and digital workers — and the second group is growing fast.

[Peter Bean @Beannation]

The real test of managing a digital workforce:
How do you know you know what you need to know?