I spent a good chunk of today actually using Claude Code and agents to do real work. Not playing with it. Not experimenting. Just working. Full audits, cross-referencing docs against live pages, writing everything up. The kind of stuff that used to take half a day or more is just part of a day now that has other things in it. The division of labor that works: agents browse, read, cross-reference, write. I direct, review, decide. That’s it. That’s the thing. I’m not sitting there prompting over and over hoping it gives me what I want. I’m setting up the task, letting it run, checking the output, shipping it or fixing it. Same process I used managing human teams. Same process I use managing AI teams. The tools changed. The fundamentals didn’t.
The division of labor that works: Agents browse, read, cross-reference, and write. You direct, review, and decide. That’s the entire workflow.
The Details That Actually Matter
Been testing the audit workflows on various properties I run. What impressed me wasn’t one big flashy thing. It was all the little details. The stuff I always think about but never get to. The things that matter but get skipped because we run out of time, or clients want something else, or they’re not earth-shattering so they fall to the bottom of the list. They’re not headline stuff. But they’re important. And now they actually get done. That’s the shift. Not replacing the big work. Not automating strategy or decision-making. Just making sure the important-but-not-urgent stuff actually happens instead of living on a list somewhere. The broken link that nobody notices until a customer complains. The meta description that’s technically fine but could be better. The documentation that exists but is two versions out of date. The edge case in a workflow that works 95% of the time but breaks on the 5%. None of that is dramatic. None of it would make a good case study. But it’s the difference between a site that works and a site that works well.Focus on the important-but-not-urgent: The details that matter but always get skipped are exactly where AI agents add the most value. Not the headline work. The unsexy stuff that compounds over time.
From iTerm to tmux (And Maybe Vim Next Week)
Got Ghostty set up with tmux today. Was using iTerm2 before. I like it as a fallback, it’s familiar, I used it for decades. Got it dialed in and customized the way I like. But as a former front-end developer, tmux always seemed like a thing I wanted to stay away from. And here we are. Probably next week I’ll be back in Vim too. Pretty wild. As a former engineer, the terminal environment matters. It’s not just aesthetics. It’s having the right tool for the right workflow. The setup itself isn’t going to be interesting to everybody. To some people it’s obvious. To others it would make their head spin. Both reactions are fine. But for me, former engineer, been in and out of technical work through health stuff and a bunch of life pivots, being able to work this way again, and do it faster and more powerfully than before, is genuinely mind-blowing. I know that’s not a universal feeling. But it’s mine. Just watching Claude agents manage all of that at once in one interface, multiple panes, pretty powerful. There’s something about getting the environment right. When your tools fit your brain, when the terminal does what you expect, when the workflow matches how you think, the work gets easier. Not because the work changed. Because the friction disappeared.Environment matters: When your tools fit your brain instead of forcing your brain to fit the tools, friction disappears. That’s not about being faster. It’s about being able to think clearly while you work.
Step by Step, Even at a Million Miles an Hour
The way I think about it: continually iterating, finding more solutions. I want to go a million miles an hour but I know it has to go step by step. If you skip a step, you’re doing yourself a disservice and you’re going to miss something. There’s always something to be learned, found out, absorbed along the way. Seven days ago was a different week, but it feels like a month ago. That’s the compounding. Keep learning something new each day. Keep iterating. Keep playing. A little something each day goes a long fucking way. Last week’s setup makes this week’s work possible. This week’s work becomes next week’s foundation. That’s the part most people miss. They see the output and think it’s about speed. It’s not. It’s about building systems that get better the longer you use them. Systems that learn your patterns. Systems that anticipate what you need before you ask for it. You can’t buy that. You can’t shortcut it. You have to build it. And building it takes time. But once it’s built, it changes everything. The compounding isn’t just about productivity. It’s about understanding. Every iteration teaches you something. Every problem you solve makes the next problem clearer. Every system you build becomes a lens for seeing the next system you need.The compounding effect: Last week’s work makes this week possible. This week’s work becomes next week’s foundation. Seven days feels like a month because the systems stack.
Building in Public Pays Off
A former colleague reached out. Really interested in the things I’m doing and building. More to come on that later in the week, potentially. You never know. That came from sharing on the blog. Building in public. Being honest about what’s happening. The right people find you when you put real stuff out there. These brain dumps sometimes feel like a waste of time. Sometimes they lead to a conversation like this. Worth it either way, but especially when it’s the latter. I used to think building in public meant having everything figured out before you share. It doesn’t. It means sharing the mess. The process. The stuff that’s working and the stuff that isn’t. The people who need that honesty will find it. The people who want polished marketing copy won’t. That’s the filter. Whoever you’re thinking about reaching out to, just reach out. You never know. It might make sense. The conversations that matter don’t start with a perfect pitch. They start with “here’s what I’m working on, here’s what I’m figuring out, does this resonate?” If it does, great. If it doesn’t, you saved both of you time.Nobody Can Keep Track of Everything
That Peter Levels interview with Lex Friedman stuck with me. And OpenClaw from a couple weeks ago, still something that resonates. A lot of people are saying a lot of stuff. Can’t say enough about Anthropic’s podcast either. It’s a wild time. Nobody can keep track of everything. That’s the thing about this moment: there’s too much happening to track all of it. The pace is accelerating faster than anyone can document. By the time you finish reading about something, three more things launched. So you have to choose. You have to decide what matters to you and let the rest go. You can’t be on the cutting edge of everything. You can barely be on the cutting edge of one thing. I’m not trying to know everything. I’m trying to know what matters for the work I’m doing. The rest can wait. Or not. Mostly not. And that’s fine. As always, communication and collaboration: those that are good at it will continue building and doing cool shit in the next couple years. The lone genius myth is dead. It’s about who you can work with, who you can learn from, who challenges your thinking without making it a competition.You can’t track everything: The pace is too fast. Choose what matters for your work and let the rest go. The value is in focused depth, not broad awareness.
Everything Goes in the Vault
My Obsidian vault is massive now. I have a very firm rule: everything goes in there. But I sometimes let my chief of staff agent help me organize things when they get sloppy, which tends to happen throughout the day. The agents have read access, but I let the chief of staff help reorganize when the mess builds up. Everything goes in the vault. That’s the rule. No exceptions. Notes from meetings. Links I want to read later. Ideas that hit at 10 PM. Code snippets. Client feedback. Personal reflections. Health tracking. Plumbing school notes. All of it. One place. Why? Because if it’s not in the vault, the AI can’t use it. And if the AI can’t use it, I have to remember it. And I’m done trying to remember everything. The vault is my external brain. The AI agents are the assistants who help me sort through it. I’m the one who decides what matters and what to do about it. I think that’s a thing about all of this: you have to keep trying, keep playing, keep tinkering. Be willing to let go of some control, but have guardrails in place too. Get yourself set pretty quickly. Don’t go too far before it gets too messy. Stay aware of those things. The balance isn’t about control versus automation. It’s about knowing when to let the system handle something and when to step in. That’s judgment. That’s the thing you can’t automate.The vault rule: Everything goes in there. No exceptions. If it’s not in the vault, the AI can’t use it. If the AI can’t use it, you have to remember it. Stop trying to remember everything.
