A few months ago I asked Claude to read back through my entire AI activity log and tell me how I was using AI, and where I could be using it better.
The answer was more useful than I expected. I had built some good skills, and I was using AI as a thought partner every day. But the record showed I was still doing most of the execution myself. I would think through a problem with AI, then go do the work by hand. I wasn’t yet trusting it to carry things out on its own.
I’d had a rough sense of that pattern, but no language for where I could get better. When sessions evaporate the moment I close the terminal window, I’m left with a nagging feeling that I’m either a power user or a fraud, depending on the day.
If some version of “am I actually using AI the right way?” is taking up space in your head, you know the feeling.
The reason I could get a straight answer is that I had a record to point at. I make my AI keep a diary.
What the log actually is
At the end of each working session, the AI writes a short entry into a running log: what was built or changed, key decisions and the reasoning behind them, mistakes or errors, and follow-ups. Plain markdown files in a folder. Sessions where nothing meaningful happened get skipped.
Each entry reads in about thirty seconds. This one is real, pulled from my log in July:
# Fixed empty WordPress site title**Date:** 2026-07-10**Workspace:** work/webmaster## What Was Built or Changed- Diagnosed missing "Jenny Wanger" prefix in browser tab titles across jennywanger.com (reported on /blog as `| Blog`, but affected every page — homepage was rendering ` Product Team Transformations`).- Root cause: WordPress **Site Title** field was empty. The theme title template is `{site_title} | {page_title}`, so an empty site title dropped the prefix everywhere.- Fix: `POST https://jennywanger.com/wp-json/wp/v2/settings` with `{"title": "Jenny Wanger"}` using app-password auth. Verified via cache-busted fetch: `/blog` now renders `<title>Jenny Wanger | Blog</title>`.## Key Decisions Made- Pushed the setting change directly rather than routing through `drafts/` first. The project rule to draft-before-deploy is aimed at content edits; this was a single-field config restore for a bug Jenny explicitly reported and asked me to fix. Reversible via the same endpoint.- Skipped the `public-api.wordpress.com` v2 settings endpoint (returned nulls silently) and the v1.1 endpoint (unauthorized with the bearer token). Used the site-hosted `jennywanger.com/wp-json/wp/v2/settings` endpoint with app-password auth per `docs/wordpress-api.md` — matched the pattern that works elsewhere.## Mistakes or Errors- First two API attempts (`public-api.wordpress.com` v2 and v1.1) failed silently or with unauthorized before I checked the vault's own `wordpress-api.md`, which specifies the site-hosted endpoint as SSOT. Should have opened that reference first instead of guessing endpoints.## Next Steps- Consider adding a note to `docs/wordpress-api.md` that `/wp-json/wp/v2/settings` is the working endpoint for site-wide settings (title, description) — currently the doc only covers pages, blocks, and global styles.
Look at the mistakes section. Claude guessed at two API endpoints before it thought to check my own reference doc, and it wrote that down.
I wasn’t watching it work, and the entry told me something about how I hand off tasks. I’d turned Claude loose without pointing it at the right reference. Either I need to prompt Claude to put in better guard rails so Claude knows to find that file, or I need to flag to Claude when there are reference files to review. Patterns like that are easy to spot once there are dozens of entries to read in a row.
I use a Claude Code skill that writes the AI Log automatically at the end of every session, and I’ve published it if you want to steal it. But nothing about this requires Claude Code. Any AI tool can follow the format if you ask.
Another nudge to drive AI adoption
If you lead a product team, you’re probably trying to figure out how to improve AI adoption across your team. The AI Log is more than a personal productivity trick.
I’ve been driving AI adoption across my client’s 50 product managers. I’ve found much more success with nudges (lots of them, running at once): show-and-tells, office hours, shared skills, pairing sessions. An AI log belongs in that pile.
Keeping a log builds awareness of how we use AI. That awareness is what opens someone up to coaching, and to going deeper on their own.
I’ve set some boundaries around how it gets used: it stays opt-in, and it stays private. It is essentially a professional diary. Each entry is your AI writing a little note to its future self. You don’t read other people’s diaries, and you don’t make anyone publish theirs. If this is going to drive self-improvement, people have to trust that no one else is reading over their shoulder.
Your weekly update gets a first draft
Before the log, my weekly client update started in one of two places: a running notes file I maintained in Slack all week, or a Friday session of calendar archaeology, scrolling back through the week and trying to reconstruct what happened.
Now AI drafts the update from the log, and I edit. I still amend, enhance, and reframe (the judgment about what my client needs to hear is still mine). But I start from a first draft instead of a blank page, and the draft already knows what I did on Tuesday.
This is the use case that pays off most quickly. If you’re like me, you’re doing most of your work in AI at this point, so the log becomes a solid record of where your attention went. Turning that into a client or stakeholder update is a small step.
And it’s not just me. I rolled the log out to those product managers as an optional practice a few weeks ago, and a couple of them have already messaged me unprompted. Asking their AI to look back over the log showed them what they use it for and where they could be getting more from it. A couple have started using it to prep an agenda for their weekly one-on-ones with their managers.
Some of my other favorite use cases include:
- Session recovery. When a session crashes or won’t resume, the log holds what was decided and where things stood. You start from the last entry instead of from zero.
- Self-reviews stop being guesswork. Come annual review time, you have a year of receipts, written down at the moment things happened.
- Proof you’re growing at this. Using AI well is now a skill people expect you to develop. The log gives your manager conversations evidence instead of vibes.
The ask here is small. Set up the habit, let it run for two weeks, then ask your AI to read the log and tell you how you’re using it, and where you’re selling it short.
Compounding payoff
Which brings me back to that review of my log.
Once Claude showed me that I was thinking with AI but not delegating to it, I knew where to push. I put stronger guidelines and structures in place so AI could do more of the work without me. I connected it directly to the tools I use every day so it could act in them instead of handing me instructions. I started building multi-step workflows instead of one-off prompts.
My newsletter now publishes end-to-end. Once I’ve finished a draft, AI handles the rest (formatting, WordPress, the email broadcast, the social posts) instead of me or my assistant clicking through each step manually. None of it would have happened without a record for Claude to review.
You can’t improve what you can’t see. Most AI use is invisible. It happens in private sessions nobody remembers a week later. The log is what makes it visible: to the person doing the work first, and to the leader coaching them to more effective use.
Just don’t ask to read anyone’s diary.