Small businesses don’t need more busywork. They need leverage.
AI can help you:
• answer common customer questions faster
• follow up with leads automatically
• turn notes, calls, or meetings into usable next steps
• create content without starting from a blank page every time
The businesses that win with AI won’t be the ones using the fanciest tools. They’ll be the ones using it to save time, respond faster, and stay consistent.
If you want a simple place to start, check out https://t.co/TbIZeftD0M
One thing I’ve noticed with ChatGPT, Claude, and Gemini is that sometimes you come across one part of the conversation that you know you’ll want to come back to later.
Not because the whole conversation is bad or lost. More because the chat keeps growing, you keep exploring, and that one useful point gets buried over time.
It could be a good explanation, a code snippet, an idea, a decision, or just something you want easy access to later.
So I built a small Chrome extension for myself called ConvoAnchor.
It lets you highlight part of an AI conversation, save it, and jump back to it later without digging through the full thread.
It works with ChatGPT, Claude, and Gemini, and everything stays local in the browser.
Sharing it here in case that sounds useful to anyone else:
https://t.co/8De5KOxJAm
ConvoAnchor started from a simple frustration:
AI chats are useful right up until the moment the one thing you needed disappears into a wall of scroll.
So I built a way to anchor the important parts of long conversations, so good answers don’t get buried, lost, or forgotten.
That feels small on the surface.
It’s not.
If you use AI seriously, retrieval becomes the whole game.
Today Cursor Composer couldn’t fix simple button padding in a Next.js + Supabase app, so I had to do it myself.
Then Claude Code, running Sonnet, confidently used outdated Next.js auth patterns and missed that I already had the file it wanted to create.
Two very normal AI coding failure modes:
• weak attention to existing project state
• stale framework knowledge presented with confidence
This is why “AI can code” is still a half-truth.
It can generate.
It can accelerate.
But if you stop supervising, it will absolutely burn time on mistakes a careful junior dev wouldn’t make twice.
Today I tried to build a simple local file-sharing app for my MacBook and Windows machine.
The goal was boring on purpose:
• send files and folders over local network
• copy text between devices
• install it once and use it
• no manually running a server every time
Claude got pretty far on the app itself.
But after 2 hours with Opus and GPT-5.4, the thing still fell apart at the packaging stage. The app existed. The installer didn’t.
That gap matters.
The demo is easy.
A clean bundled app people can actually install is where a lot of AI coding still starts to wobble.
Good reminder that “it built the app” and “it shipped a usable product” are not the same sentence.
My agent-browser skill has been giving me trouble lately.
I set it up as a cron job to:
log into my gym site,
check the workout of the day,
send it to me in Discord,
then ask what class I want to attend.
It works… until it randomly doesn’t.
That’s the frustrating part with real-world agent workflows:
the logic can be fine, and the brittle part is still the website, session state, or UI changing underneath you.
I’m building an app that creates running routes from natural language.
No dragging waypoints.
No drawing the route yourself.
Just:
"Give me a 10-mile loop starting at my house."
Big challenge right now:
getting Valhalla routing engine to generate actual loops instead of fake loops with out-and-back sections on the same path.
Spent today setting up the X posting pipeline.
Took way longer than expected, but now the whole thing is automated.
That’s the trade I want to keep making:
spend longer once, save time every day after.
If you're also trying to build something real with limited time -- this is for you.
Follow along. Ask questions. Push back if something doesn't make sense.
Posting daily: what I'm building, what's working, what broke, and what I'm automating next.
Let's go. 🦾
I'm building an AI-powered income stream while working a full-time job and raising a family.
I'm documenting the whole thing here -- the wins, the pivots, the dead ends.
Here's why I'm doing this and what I'm building. 🧵
I've already made some mistakes worth skipping.
Spent weeks building automation for a local gym. Hit a wall when the owner wouldn't share platform access. Pivoted.
Built a running route app before validating if anyone wanted it. Parked it.
That's the kind of thing I'll share here too.