Try Supabase for Free: https://supabase.com/

Cloudroom: https://cloudroom.dev/

Well done David, the calibre of your guests is skyrocketing! Guillermo is a cool cat, lots of insights in this interview 👌

i am loving that this channel's thumbnails are just obituaries

Guillermo, you and Miguel de Icaza have been true role models and aspirational mentors for me, and part of my entire generation. Seeing achieve what you  have is incredibly inspiring and completely shaped our tech careers. Te queremos mucho, Guillermo! Greetings from Barranquilla, Colombia! 🇨🇴🙏

00:00:00 Post-AGI Era: Creative Control, Taste & Human Ambition
00:05:27 Finding Your Edge: Anticipation, Timing & AI Evals
00:11:37 Speed vs Velocity: Building Autonomous Software Factories
00:21:39 Guillermo’s Agent Setup: Fast CLIs, Ghostty & Unix Philosophy
00:32:43 Instant App Generation & The Evolution from v0 to FX
00:40:20 Verification Engineering, Latency Numbers & Systems Ontology
00:54:35 Embracing Technical Debt for Extreme Iteration Speed
01:01:24 AI Cybersecurity: Deepsec, Finding Zero-Days & Code Auditing
01:12:04 Cloud Agents vs Local Execution & Sandbox Security
01:21:35 Cutting Through AI Hype & Calibrating High Quality Standards
01:32:53 The Future of Software Architecture & 100x Startup Ambition

How the fuckkkk did this guy get this dude in an interview?! Holy crap. Aight guess I cant keep talking shit lol

Really fantastic conversation! Thank you!

Haha “mini farm in their balcony” 😂 that’s a wild way to say garden

just wanna say thanks. really value right here!

Man! The best podcast you made! Perfektní !

Man I'd love to eat some LSD with this guy. Dude is immensely thoughtful.

Great interview, thanks.

When Guillermo speaks, I listen.

What an incredible guest, instant click!

Wow you really got this GOAT on your channel, good job

Holy crap David u got to interview Guillermo

Great podcast!

Super interesting!

The main message of the video is that the role of software engineers is shifting from writing code manually to designing and supervising systems in which AI agents produce code.
Guillermo Rauch argues that engineers still need strong technical understanding — architecture, APIs, data flow, performance, security — because someone has to define the goal, constraints, tests, and decide whether the agent’s output is actually correct.
A key idea is the “software factory”: instead of using one coding assistant in a terminal, companies will build persistent environments where agents have access to context, tools, tests, observability, and clear boundaries, and can execute engineering tasks semi-autonomously.
The downside is that faster code generation also means faster creation of technical debt and security debt, so agentic development needs strong verification, sandboxing, logging, and governance.
In one sentence:
The future software engineer will spend less time typing code and more time designing, directing, and validating the systems that generate it.

"unbottle the agents" :-)

They say there are specialized models—or ones that are more robust for certain tasks than others—so why not implement a pre-classification layer or a "mini-model"? This component would take a user's prompt, categorize the topic, and then—acting as an orchestrator—send refined, simplified prompts (like "caveman style") to the appropriate model. It would then manage the responses and charge a single subscription fee. The company building the classifier would handle negotiations with providers like Claude and ChatGPT, manage costs and margins, and offer a unified price to the customer, effectively delivering the best of all models. Just a thought.