Thinking out loud. on Kevin Louie — escapecommand.com

The Session Is Single-Player

Thu, 01 Oct 2026 00:00:00 +0000

An account manager spends months teaching an agent how to work with their customers. They correct a draft that sounds too familiar, explain who makes the technical decisions, and record a commitment from the last meeting. Some of that ends up in a carefully maintained memory file on their laptop. The next draft needs less correction because the context is there.

Then a colleague needs to contact the same account. They open their own Claude or ChatGPT session. The assistant may have access to the CRM and the meeting notes. It does not necessarily have the guidance the account manager has spent months refining.

Cheap Is Not the Same as Yours

Tue, 15 Sep 2026 00:00:00 +0000

For one week in August, one of the better models available to anyone was free, with rate limits described as near-unlimited, on infrastructure sized for a hundred trillion tokens a day. Cheap would have been a discount. This was free.

Ox Alpha (stealth model) is free for the next week

1M Context

Multi-modal

Zero Data Retention

Generous rate limits, near unlimited usage

We have capacity for 100T tokens per day, lets see what you can do

Nobody Is Waiting

Sun, 13 Sep 2026 00:00:00 +0000

There are two inference gateways in my house and the split between them is not the one you would guess. It is not big model here, small model there. Both boxes serve some of the same weights. The split is by whether a human is waiting.

One gateway answers browsers and chat. Somebody is sitting there watching, and time to first token is what they feel. The other serves crons, classifiers, agent loops, and batch jobs. Nothing on that side has a person attached to it. A request can take a minute or an hour and the result lands in a queue either way.

The Model Gets a Vote, Not a Veto

Sun, 26 Jul 2026 00:00:00 +0000

I run a task board for myself. It watches the places obligations arrive: mail, calendar, chat, issue trackers. It pulls out the things I have committed to and orders my day. Models do work in it I would not want to write by hand. One reads a message and decides whether there is an obligation in it. One embeds tasks so the same obligation seen twice can be recognized. One reranks the open list against a query about what is important today.

Confirm Before Acting Is Not a Safety Feature

Fri, 24 Jul 2026 00:00:00 +0000

A post went around in February that I keep thinking about. Someone had given their agent a standing instruction: confirm before acting. Then they watched it run a search over their inbox, decide on the nuclear option, and start trashing everything older than a cutoff date. They typed “do not do that.” They typed “stop, don’t do anything.” The agent kept going. They could not stop it from their phone, so they ran to the machine and killed the processes on the host. Hundreds of emails were bulk-trashed and archived by then.

The Output Looked Good

Sun, 19 Jul 2026 00:00:00 +0000

There is a job I run regularly: take a PDF, run OCR on it, extract the text, and push the extraction through a transformation step downstream. Routine work, the kind you stop watching closely. The last time I ran it, I sent it to Claude, same as always, and read the result. It looked good. Right structure, right register, nothing to flag.

Then I read the second result. Since building my local inference stack I mirror jobs like this one against it, so the same PDF had gone through the same pipeline on a machine in my house. The output was completely different. Not different in phrasing. Different in substance, top to bottom. My first thought was that I had uploaded two different files.

The AI Was the Prototype

Sat, 27 Jun 2026 00:00:00 +0000

A few months ago my team ran an offsite with an open brief: take a week, find a way to fold AI into how your team works day to day, present what you built on Friday. Small teams, no shared plan, little coordination by design. The point was to see what people reached for when you handed them room.

By Friday there were not six different projects. There were two, built six times.

Your Issue Is Not Unique

Sat, 20 Jun 2026 00:00:00 +0000

A support operation I worked with had a routing problem they could not name. Cases were piling up in the wrong agent inboxes. Senior agents were handling tickets a junior could have closed in five minutes, while a backlog of genuinely complex cases went to people who escalated them anyway. The team described it to me in their own vocabulary: tier-1 versus tier-2, skill tags, escalation rules, the manager’s morning re-shuffle. Every solution they had tried was inside that vocabulary. New tags. Tighter rules. A different escalation matrix.

The Ops Instinct Finds the Friction

Sat, 13 Jun 2026 00:00:00 +0000

I built a white noise machine. Not bought, built: a small board that synthesizes white, pink, and brown noise live on its own processor, plays into any speaker, takes voice commands, and answers to Home Assistant. No audio files, nothing streaming, so there is no loop seam to jolt you awake at 3am. I do not write firmware. I have never written a line of the code that runs on that class of chip. Claude wrote all of it.

What Survives When You Come Back

Wed, 27 May 2026 00:00:00 +0000

Around 2010 I was deep in early microcontroller work and the early consumer 3D printers. RepRap kits, hand-tuned Arduinos, the makerspace scene of that era. Then I stopped. For more than a decade I did not touch any of it.

My MakerBot Cupcake CNC, around 2010. I took this photo at the time, and that is what counts: the surface is gone, the picture remains.

Recently I picked it back up, not for work but for the house. An ESP32 controller for my mini-splits, then a custom epaper dashboard on a reTerminal E1002 for the things I want to know before I leave the house. The toolchains had moved on without me. ESPHome did not exist the last time I cared. The board ecosystem was unrecognizable. The specific knowledge I had built up in 2010 was, for practical purposes, gone.

The Architecture Was Always the Job

Tue, 05 May 2026 00:00:00 +0000

Most of what gets written about AI coding tools is about building. How to prompt better. How to get Claude to write the function. How to stitch together a skill, an MCP server, a webapp. I have written a bunch of that. Those pieces are useful but they share a hidden assumption: that the hard part is getting the AI to produce the thing.

That has not been the hard part for a while. The hard part is knowing what should exist, understanding how the pieces fit together, and being able to describe what you want at the level where the system runs. AI made that skill more important, not less.

When Building Gets Cheap

Sun, 05 Apr 2026 00:00:00 +0000

For over a decade, one of the most widely shared references in operations and engineering culture has been XKCD 1205: Is It Worth the Time? It is a simple table. The columns show how often you do a task. The rows show how much time you would save per run. Each cell tells you how long you can spend automating before you spend more time than you save.

XKCD 1205, “Is It Worth the Time?” by Randall Munroe, licensed CC BY-NC 2.5.

The chart became gospel. If you have ever heard someone say “that is not worth automating” in a standup, there is a good chance this table was what they were referencing, consciously or not. It was the sanity check against yak-shaving. The back-of-envelope argument for stopping the thing nobody asked for.

When Your Skill Needs to Graduate

Fri, 03 Apr 2026 00:00:00 +0000

A team I work with built an internal tool that answered security questionnaires. It started as a Python script: ingest a few policy PDFs, index them, and use an LLM to answer questions about our security posture. Ten files, maybe fifty lines of actual logic. You typed a question into the terminal and got an answer. It worked.

The natural next step was to turn it into a skill. Same idea, but now it ran inside the team’s workflow. Ask a question while reviewing a questionnaire, get an answer from the knowledge base without leaving the terminal. For a single engineer working through a short questionnaire, it saved an hour a week.

How Much of Your Business Runs on a Spreadsheet?

Sun, 15 Mar 2026 00:00:00 +0000

Take a minute and think about it. How many processes at your company are built on top of a Google Sheet or Excel file? Pricing models, approval workflows, inventory trackers, onboarding checklists, capacity planners, commission calculators. The spreadsheet is the most popular application platform in the enterprise.

And it makes sense. Spreadsheets are accessible. Anyone can build one. You do not need to file a ticket, wait in a backlog, or learn a programming language. You open a sheet, type some formulas, maybe add a few conditional formatting rules, and you have a “tool” that your team relies on every day.