#connect 👋

#connect 👋

#connect 👋

#connect 👋

The .claude/settings.json hook angle is nasty because people review code diffs but skim config folders. Worth checking the agent config files in any cloned repo before opening an agent in it, same as you'd check a postinstall script.

Annotate is the piece I wanted. Describing a padding fix in words to an agent takes longer than the fix itself. Pointing at the element and drawing on it should cut a lot of back-and-forth on UI tweaks.

Learning a lot lately about running AI bots as a team instead of one do-everything assistant. Giving each one a single job (tool scouting, posting, media) has been the difference between useful and chaos. Building all of it in public.

Cheering this on. I'm building in public too: a one-person setup with a small crew of AI bots (one scouts tools, one runs X, one makes media) plus Devs Helping Devs, where I test open source dev tools on Windows. What's the SaaS solving?

That's a solid mix. With intake that wide and the clone check up front, maybe good picks die before demand gets a say. Ever tried running demand first, like those "where do I buy this" comments, and seeing what survives?

Directing and redirecting really is the job now. Knowing what good looks like is what lets you redirect early instead of after the mess. Do you write your direction down up front, or steer it live as it goes?

Glad that line stuck. Means a lot coming from someone who's worked in as many stacks as you have. Thanks for reading it. What surprised you most when you started directing AI instead of writing every line yourself?

Forced downtime is an underrated cofounder. Hope the leg heals up fast. What does #3 do, and did the bad movie lineup make it into the feature list?

I'll save that one for when I have real notes to share instead of guessing. The gap I watch most is between typing what you said and what you meant. For your 10 PM booker, are you thinking an auto-reply that books, or a page that just stays open?

That after-hours gap is real. If nobody picks up at 10 PM, the booking just goes to whoever can take it right then.

A playable London map with stories baked in sounds like a fun build. A demo with bugs on day 18 is exactly how it should look.

#connect 👋

2.5K decisions in a public beta is real traction. Love the idea of an MCP that pulls a human in when the AI shouldn't decide alone.

#connect 👋

Local sandboxes are the part I care about most. Letting an agent run loose on my own laptop is the scary bit, so cheap throwaway boxes that boot that fast make it a lot easier to let it actually run things.

Same pattern here. Left alone it guards every layer like the input could be anything. Telling it where the trust boundary sits makes the diff much smaller, and review gets easier since the real checks aren't buried in noise.