Putting AI agents into a small company's software process without losing control of it
What changes when code is written by agents: identity, review, testing and the boring safeguards that make it safe to go faster.
· 12 min readWhat we use, and why
Tool reviews, white papers and market notes from our own builds. If we recommend something to a client, we have run it ourselves first, and the scores come from that use.
6 entries · newest first
All entries
A new entry most weeks, written from what we built or tested that week. We say what we measured and what we did not.
What AI models cost, per job, and what we can price so far
The prices and usage we recorded, the traps in reading them, and how we are measuring cost per job.
· 6 min readSelf-hosted LLM gateway: one key, many models
Routing to the cheapest model that passes, with fallbacks, and why a fallback can hide a dead model for weeks.
· 7 min readA shared memory service for coding agents
Why sessions forget, and whether one network service fixes it. Search got good; getting agents to read it did not.
· 8 min readVoicebox: a free local voice studio, and the clone is the weakest part
Stock voices that are fast and natural on a CPU, a voice designed from one sentence, and a clone that disappoints.
· 5 min readClaude Code edits a video: what 37 seconds became 19, and where it stops
A free skill, local Whisper and FFmpeg turned a rambling take into a titled, captioned clip. It overlays; it does not cut.
· 5 min readWant this on your project?
Everything in the Journal comes from software we build and run ourselves. If you want the same discipline applied to your project, tell us what you are trying to do.