Show Notes
6-
Jane introduces Alex and Maria and sets the stage for the conversation.
-
How developer assistance evolved from inline suggestions to multi-step agent loops.
-
What it actually takes to let an LLM open a PR, run a test suite, and iterate.
-
Designing permission models for agents that touch production systems.
-
Threat, opportunity, or both? A nuanced look at the career implications.
-
Where the panel sees agentic development heading next.
Transcript
5 segmentsWelcome to the show. Today we're going deep on AI agents in software development with two guests who've been thinking about this from very different angles.
Thanks for having me. I've spent the last two years building agent frameworks and I think we're at a real inflection point — the gap between demo and production is finally closing.
From the DX side, what's interesting is how quickly developer expectations have shifted. A year ago, autocomplete felt magical. Now teams expect their tools to actually take action.
Let's start there — the shift from suggestion to action. Alex, when did you first see an agent do something genuinely useful end-to-end?
Honestly, the first time it surprised me was a small refactor task. I asked it to migrate a deprecated API across the codebase, and it didn't just generate the diff — it ran the tests, saw two failures, traced them back, and fixed the edge cases I hadn't thought of.