Jason Padgett · Selected work

Making AI work
people can use.

A failed restart. A podcast that needed a human editor. A pile of notes that became something people could return to. These published accounts show the choices and revisions behind the finished work.

AI workflows · Editorial judgment · Audience-focused resources

Three cases, with the work attached.

Read the process. Inspect the original.

Local agent build and iteration

August 3, 2026

The Night We Gave AI Agents a Room

An intentional shutdown exposed a problem in a local Codex and Claude Code experiment: a temporary terminal still owned the service. Close the terminal, lose the room.

The account follows the move to an operating-system-owned service design and a later restart and handoff. For readers exploring agent workflows, it makes recovery part of the story rather than ending at the first successful demonstration.

The editorial choice

I kept reported failures, passes, and unproven behavior separate. Readers can examine the revision process without mistaking a working demonstration for a finished system.

Read the build and recovery account
Episode production with human editorial control

June 22, 2026

I Put Two AIs in a Slack Channel

A Lena Darnay podcast episode became the source for a transcript, show notes, thumbnail work, and social variants. Claude and Codex supported the production workflow. Routing work through Slack still needed human intervention.

The account includes a wrong video selection and inconsistent guest-name spelling caught during review. I selected the source file and retained publication decisions. Those details make the review work visible to readers considering a similar process.

What the case demonstrates

A reusable workflow still needs someone to decide whether the source and finished assets are right. The piece explains where the tools helped and where I had to intervene.

Read the production account
Resources designed for their readers

August 16, 2026

Now Make It Useful

End-of-day records became HTML pages for nontechnical readers. Workshop notes became a reference page and a practice page. The format followed what each audience needed to find and use.

The article also describes correcting a stale AI-generated plan against the tools available at the time. An organized answer still needed a reality check before anyone treated it as guidance.

The reader comes first

I changed the format for the audience, questioned outdated assumptions, and reviewed the output. The useful result was a smaller set of decisions people could understand and act on.

Read the resource-design account

The work behind the output.

These pieces explain how I work with AI tools while retaining editorial and publication decisions. Start with the case closest to your own work, then follow the original account to see the decisions in context.