My latest book is live and available now. This one is all about navigating the harsh realities of enterprise software projects, which claim to be "agile," but not really.
Sprints, standups, fixed fall scope, sprint-end QA batches, velocity on a leaderboard. If most match, you have a map of pressure, not a verdict on the team.
This is Gemma 4:12B running on a 16 GB MacBook Air M1. It sat for two hours before outputting anything but Working. Now itโs generating about 1 word every 30-40 seconds. Memory use on the MacBook is over 21GB.
@twwilliams I was running it inside of Codex against Ollama + Gemma 4:12B. I asked it to refactor a web page by editing a button. Two hours later it was still thinking, so I killed it. I'm now trying something a little simpler in VS Code against the same Ollama + Gemma.
.@GoogleDeepMind's Gemma 4 - 12B is available on Ollama!
Chat:
ollama run gemma4:12b-mlx
Hermes Agent:
ollama launch hermes --model gemma4:12b-mlx
Claude Code:
ollama launch claude --model gemma4:12b-mlx
and more ๐๐๐
(Note, this currently works via MLX)
@KyleHebertMusic@intheworldofai Running it now on a 16GB M1 MacBook Air. "Working" is a bit optimistic. I asked it to replace a button in an HTML file. An hour in, it's still "thinking" and the Mac is warmer than it's ever been. Maybe I should have started with something simpler. ๐
@krishravela In some ways, I miss working for startups. I was with a company called eTelemetry back in 2008. I could draw a very solid line between my code and our sales.
I added a new feature based on sales feedback. They sent me to trade shows to demo it. I did it all. ๐
@chrisbbehrens Had another project once where the manager proposed comparing developers by the number of their commits per week. That one blew up fast, as I'm sure you can imagine. ๐คฃ
A stakeholder asked us to raise velocity twenty percent. I said we could add twenty percent to all estimates. Joke. Not entirely a joke. The number is not the work. Not sure many people get that.
If your AI setup has twelve config files and the team cannot explain it in one standup you have a hobby not a workflow. What is the smallest setup that stuck?
Story estimation often goes like this: team says eight, management writes five, delivery is thirteen, everyone acts surprised. The number was always a guess, not a promise.