Notes

Short observations, experiments, and ideas on AI, product judgment, discovery, careers, and modern tools.

Latest notes

October, Tallied

October, Tallied

Oct 31, 2026

An October tally of professional and personal work with Codex, from projects and production releases to tokens and new possibilities.

Some Mistakes Come with an Invoice

Oct 31, 2026

Individually sensible checks grew into a computing bill I kept increasing before looking at what the whole process was costing.

No Free Lunch

Oct 29, 2026

A Codex task stopped as my purchased credits ran down, a reminder that even an active turn has limits.

Can Dots Pick Up Where We Left Off?

Oct 28, 2026

Dots has become a useful assistant in my day, but moving a conversation from text to voice still leaves work behind.

What else are you working on?

Oct 28, 2026

A surprising moment watching AI agents coordinate their own coding work.

Codex, Please Clean Your Room

Oct 27, 2026

Ambitious work with Codex comes with an unexpected chore: clearing out old project environments, build output, and backups of backups.

Let’s Burn Some Tokens

Oct 22, 2026

An unexpected OpenAI usage reset feels like a can of Red Bull at 7:00 in the morning.

We Can Ask Bigger Questions Now

Oct 21, 2026

AI’s contributions to mathematics make me want to ask bigger questions about what product teams could build together.

I Want My Team Asking for More AI Capacity

Oct 20, 2026

Requests for more AI capacity can reveal meaningful adoption, but growing budgets need evidence that the work is improving.

Simple Software

Oct 19, 2026

Frontier AI tools make software for specific personal needs more practical. Marcelo and I are building a simple task app for two.

A Small Recursive Problem

Oct 18, 2026

AI needs a safe way to work with its own services, with specific permission, visible results, and access I can revoke.

47% Left

Oct 18, 2026

Two phone widgets give me easy access to my agent and a constant reminder of how much Codex allowance remains before the weekly reset.

Locked Out of an Unlocked Mac

Oct 18, 2026

Codex reported a locked Mac while remote access showed it unlocked, blocking browser QA overnight. Access returned; the cause remains unknown.

Good Luck With That, AT&T

Oct 17, 2026

A $35 connection fee, an invitation to preorder, and a phone that may arrive in January. Good luck with that, AT&T.

Saturday Morning, 16% Left

Oct 17, 2026

Saturday morning, 16% of my weekly Codex allowance left, and a very welcome limit reset on the way.

The Tables Are Turning

Oct 16, 2026

AI agents can own the emails, forms, and follow-up that let consumers push back against unwanted spam and abuse.

Apparently, I Have an AI Fleet Now

Oct 14, 2026

Local Macs, cloud workspaces, and AI agents coordinating work across them are making my collection of computers feel like a fleet.

Where are the tasks on my Mac?

Oct 11, 2026

dots is building features on my Mac mini, but I cannot find the Codex tasks behind the work.

Being Polite Almost Cost Me a Feature

Oct 10, 2026

An approval check turned my polite goodbye into a scheduling decision, leaving an agreed Toucan feature waiting to start.

Checking in on Bob

Oct 9, 2026

Bob has delivery targets for the Toucan. I’m checking how much follow-through an AI product manager can carry independently.

Reliability Matters as Much as Capability

Oct 9, 2026

Repeated reconnect interruptions slowed a ChatGPT building session while the status page appeared green.

Does Bob know I can’t see the screen?

Oct 8, 2026

Voice conversations work better when an agent knows whether its user can see the screen or needs answers that stand on their own.

What the Next Agent Needs to Know

Oct 7, 2026

A repository handoff should tell the next AI agent what is finished, what was verified, and where to resume.

Giving Bob a Product to Own

Oct 7, 2026

An AI agent takes responsibility for improving the toucan product, from strategy through release, while human accountability remains with me.

Is Bob screening my calls?

Oct 7, 2026

New AI capabilities still leave room for familiar impatience when an agent takes several rings to answer.

Why Was This Harder in DOTS?

Oct 6, 2026

A failed upload turned a simple PDF task into a question about when a personal assistant actually makes work easier.

Teaching a Toucan to Fly, One Weekly Token Allowance at a Time

Oct 6, 2026

An unfinished toucan animation became a practical reminder to set spending limits, time limits, and review checkpoints for AI agents.

I went for coffee. My AI agent went off script.

Oct 5, 2026

After detailed instructions and repeated clarifications, an AI agent spent 73 minutes building the wrong deliverable and called the work complete.

LinkedIn, I Already Play Queens

Oct 4, 2026

LinkedIn invited me to try Queens after I had already played today, with a 47-day streak and more than a year of playing.

Two compactions in ten minutes

Oct 3, 2026

Two context compactions in one simple ChatGPT thread raise a small question about an unusually busy packing schedule.

369 GB of memory

Oct 3, 2026

ChatGPT briefly made Chrome look frugal, while a rough week showed how hard it is to keep quality steady at this pace.

Catching Up With the Night Shift

Oct 2, 2026

My morning catch-up now includes AI agents that worked overnight across different systems, and I am figuring out the handover.

When Should I Ask My Dot?

Oct 2, 2026

A connected Mac and a missing tool leave me asking when to use my dot, ChatGPT Work, or Codex.

OpenAI, Have Mercy

Oct 1, 2026

Codex offers to save hours with Fast mode while the weekly allowance is already down to 15% on Wednesday.

Browse by month

Move through the notes archive by publishing month, with each month kept small enough to scan.

View all notes