you can just build "things"

I'm so glad prompt engineering died. It was always stupid.

Welcome Edinburgh and welcome to this year’s @kcduk_io! Excited to be back as my first international talk happened here at @KubeHuddle (OGs will know this conference) in 2022!

🤣 I couldn't help myself. "Generate an SVG of a walrus playing the bagpipes"

Neo in a @Docker Docker Sandbox: Using Pulumi’s Coding Agent for All Things Infra Safely and Securely @PulumiCorp pulumi.com/events/neo-in-a-d…

My best explanation of the “software factory” is to literally think about engineering output as a factory. Before AI, software was mostly craftsmanship. The craftsmen were software engineers, and we spent decades optimizing everything around them: PMs writing specs so engineers worked on the right thing, better tooling, CI/CD, testing, infra, etc. But ultimately the throughput of the factory was still constrained by how much software humans could physically write. Then the first wave of AI made the craftsmen much faster. Autocomplete, Cursor-style coding, pair programming with AI. Now agents can increasingly take a feature end to end: understand the spec, inspect the codebase, write the code, test it, fix it, ship it. So for roughly the same human input, the output of the software factory starts going up dramatically. And I think the end state people are chasing is a mostly autonomous software factory. Engineers write a spec / markdown file, define the tests and constraints, and become much more like verification engineers, while swarms of agents run 24/7 looking at telemetry, customer bugs, regressions, feature requests, then designing, coding, testing and shipping software. But for that to actually work, the rest of the factory has to be insanely good too: CI/CD, tests, fast deployments, observability, security, rollback, evals, all of it. It’s basically going from Ferrari building cars with incredible craftsmen doing a huge amount by hand, to the most advanced Tesla-style factory where more and more of the production line is automated. And the big assumption behind all of this is that your product will eventually be as good as your software factory. If your factory is better, your agents can iterate faster, learn faster, ship more, and ultimately just build a better product.

Coffee prep for my upcoming trip. Most of the coffee places I go to fail on creating a good drip/batch coffee. It’s not always due to cheap beans but mostly the baristas lack of knowledge. Sometimes you are lucky with a good one but consistency is the key!