Coding Agents Don't Scale Teams

󰃭 2026-09-11 | #ai #dim factory #guardrails #software factory #software-development #workflow

I’m watching Patrick Debois’ talk Coding Agents Don’t Scale Themselves. Neither Do Your Teams. for the third time. He describes a familiar transformation program: hackathons, lunch-and-learns, a shared Slack channel. We know these formats from Agile and DevOps.

And yet a company can run all of them and still depend on a handful of people maintaining private AI workflows.

That gap is what keeps me watching. Underneath it sits one question:

Continue reading 

The Factory Looked Good Until It Had to Work

󰃭 2026-09-07 | #agents #ai #ai engineering #dazzhub #deno #orca #software factory #testing #workflow

TL;DR
My workflow looked good until real DazzHub issues put pressure on it. Commands tested the wrong checkout, copied shell snippets drifted apart, and the journal executed text that should have been data. The fix was not a longer prompt. I moved repeated operations into a tested Deno CLI with cliffy and turned important rules into executable checks.

The first version of my agent workflow looked convincing in Markdown. It had a board, named states, skills for each job, worktrees, a CI command, and rules about when an agent had to stop.

Continue reading 

8 Stages to a Software Factory, and Where I Am Now

󰃭 2026-09-02 | #agents #ai #ai engineering #claude code #codex #orca #pi #software factory #workflow

TL;DR
I thought I already had an advanced agent setup. What I actually had was a group of capable workers with me acting as scheduler, state machine, and safety net. DazzHub and Orca moved parts of that setup towards Stage 7, but the interesting work was not better prompting. It was isolation, state, gates, and limited permissions.

A few weeks ago, I still started every coding agent myself. I opened a terminal, explained the task, watched it work, ran the tests, and decided what happened next. The agent wrote much of the code, but I remained the scheduler, the state machine, and the audit log.

Continue reading 

Orca Is My Agent Development Environment Now

󰃭 2026-08-23 | #agents #ai #ai engineering #claude code #orca #pi #self-hosting #software factory #workflow

I used to open an IDE, open a terminal, find the right file, make a change, and run the tests myself.

That is still a perfectly reasonable way to develop software. It is also no longer the way I want to work.

At the beginning of August, I moved my development into Herdr. Herdr gave me the first important step away from an IDE: one terminal interface for several agents and projects. I could send an agent to work, switch to another project instead of watching it, and see when the first agent had finished.

Continue reading 

Four Ways to Extract Knowledge from Unstructured Data (and Why RAG Isn't One of Them)

󰃭 2026-07-17 | #AI #knowledge extraction #LLM #ontology #RAG

The problem: knowledge trapped in the unstructured

Every video you watch, every long-form article you read, every podcast transcript sitting in a folder somewhere contains knowledge that is, structurally speaking, in the worst possible shape for reuse. It’s linear. It’s redundant. It’s mixed with filler, tangents, and repetition. The same concept might be explained three different ways across forty minutes, buried between an ad read and a rambling aside about the weather. If you want to use that knowledge later — cite it, search it, connect it to something you learned six videos ago, hand it to an agent as durable context — you cannot use the transcript directly. You have to extract.

Continue reading 