More Than Just Code podcast - iOS, visionOS and Swift development, news and advice · Tim Mitra, Jaime Lopez Jr, Mark Rubin

Robots, AI, & Cephalopods with Alexis Gallagher

·1 hr 17 min·5 clips
Alexis argues vibe coding is not “shut off your brain,” but a different way of thinking about delegation.
1. More Than Just Code episode 397 centers on Tim Mitter and Alexis Gallagher discussing AI coding, WWDC memories, and San Francisco life. 2. Tim Mitter hosts from Toronto, Ontario, and Alexis Gallagher joins from San Francisco, California, after appearing on the show’s 2019 Live from WWDC broadcast. 3. The episode asks what AI coding tools actually change for iOS and Swift developers, and whether the shift is about speed, control, or delegation. 4. Tim brings up San Francisco’s “eye-wateringly high” cost of living, and Alexis says the cost is very high and especially difficult with “two kids” in the city. 5. Alexis says one reason to work in San Francisco is proximity to people “doing interesting work,” including meetups where “a thousand people” showed up for the open-claw conversation. 6. Tim corrects a previous fact about attending Macworld 2010, and he says he met Mark Rubin there, who became a co-host and collaborator. 7. Tim says he has used Claude heavily for vibe coding and code assistance, and he mentions Paul Hudson’s talk on setting up AI coding. 8. Alexis describes himself as a “machine learning AI guy” and says he started as CTO at Topology around 2016, where the role was “heavily iOS leverage” and “heavily ML leverage.” 9. Alexis says he later worked as a senior staff engineer at Google, managing applied ML research for Google Ads. 10. Alexis says he is “amazed at the pace of progress” and that AI is becoming “more and more relevant to software development.” 11. Alexis says he built AnswerAI’s Fast Migrate through what he calls vibe coding, using cloud code to produce the library without writing the code himself. 12. He distinguishes that workflow from the caricature of “you don’t look at the code,” saying the real method is incremental decomposition, testing, and frequent commits. 13. Alexis says current tools can now take much larger steps, like finding libraries for a wake-word detector and working autonomously for “30 minutes.” 14. He argues that skepticism about agentic coding often misses the point of delegation, comparing code review concerns to managing “100 engineers.” 15. Alexis says the key question is not whether he can inspect every line, but whether he can verify inputs, outputs, performance, error behavior, and future evolvability. 16. He tells a hackathon story where he used multiple Tmux windows and codex agents to answer questions about PDFs, spreadsheets, and text documents under time pressure. 17. Alexis says that worked because the hackathon grading was cheap to verify: wrong answers only triggered a cooldown, not a deeper penalty. 18. The conversation shifts into LLM mechanics, where Alexis explains in-context learning, saved Markdown files, and longer-term model training as separate layers. 19. The style is conversational and technical, with Tim asking practical follow-ups and Alexis answering in structured, example-heavy explanations. 20. Listeners who work in iOS, Swift, ML, or AI-assisted coding will get the most from it; listeners wanting light entertainment over technical detail may skip it.

As heard by us

A practical look at how AI tools become useful when communication stays explicit.

Alexis Gallagher's return visit centers on the practical machinery that makes AI tooling useful: skills as plain Markdown files, tool calling as a separate function, and the communication habits that keep both from drifting into vagueness.

Read the full review in PlayNext →

Why you'd press play

You want a plain-English look at how one developer uses skill files, CRC, and delegation, not hype.

Read the full recommendation in PlayNext →
Listen to the show on