Ex-Google Senior Engineering Manager. Before that, a decade at startups. Left in late 2025 to build on my own.
Now I mostly build things with AI and write about what I learn, under the GammaVibe name (YouTube, blog, the occasional course). I also make indie desktop apps under Sunset Mesa Software: buy-once, no subscriptions, BYOK or local models, your data stays on your machine.
OP here. I originally launched my autononomous startup idea pipeline close to 3 months ago. It's been running smoothly since then, I've learned a lot, and continued to tune it. So I updated the post to reflect my learnings and improvements (such as podcast generation and better candidate synthesis).
I think it's all up to you what you make of this situation. For context: I've been programming for well over 40 years, and I've worked as a software engineer (or eng manager) for close to 30 years. I've gone through cycles of coding a lot and cycles of focusing more on architecture, product, or the business side. And I loved coding when I first started. Making pixels appear on the monitor felt magical back then. But coding got old over time, and I've found myself drawn to higher level concerns, including business, product, engineering architecture trade-offs, etc.
Lately, I've found this new AI assisted coding world immensely empowering. In the last few weeks, I've built web apps (e.g. Astro), an agentic pipeline (Python, Pydantic AI), a proof of concept for a roguelike game (C++ with libtcod), and I'm now working on a cross-platform desktop app in Tauri 2 (Rust and TypeScript). And I'm learning a ton. Not necessarily low level language details - I'm increasingly trusting those to the LLM - but application architecture, product design, marketing, agentic engineering, etc. I've found this genuinely empowering and haven't felt this motivated about application development in decades.
I'd say zoom out a level and see if you can find the joy in learning some of the higher level concerns.
Op here. I posted a video where I walk through how I auto-generate audio overviews for my newsletter. I cover generating a script (in my case from a markdown post), selecting voices, defining the hosts and their audio profiles, setting the scene, and finally generating the actual audio.
Happy to answer any questions that folks might have.
I recently shared the text write-up of my autonomous market research agent. Based on the questions I got about the specific implementation, I recorded a video walkthrough of the architecture and codebase.
In the video, I show:
The Pydantic AI orchestration logic.
The state-based pipeline architecture using Postgres/pgvector.
Two bugs I ran into with Gemini Flash, e.g. how I prevented it from truncating UUIDs.
A demo of the new Audio Overview feature (generating a podcast from the research).
This project started on a whim a few months ago. I was curious if it was possible to automate ideation by feeding raw news into an LLM and getting coherent, viable business concepts out, rather than just generic summaries.
I ended up building a full pipeline (10 steps, db state driven) to test the theory. It runs daily on a DigitalOcean droplet, scans ~500 articles via EventRegistry, and uses Gemini 2.5 Pro/Flash to synthesize the trends and post a fleshed out idea to Ghost CMS each morning.
The stack is Python 3.13, Pydantic AI, and Postgres with pgvector. I found Pydantic AI particularly helpful for handling the structured outputs, and Logfire was a super helpful bonus. The vector search turned out to be critical for semantic deduplication so the agent doesn't get stuck in a loop suggesting the same ideas every day.
Happy to answer questions about the architecture or the prompt engineering logic!
Now I mostly build things with AI and write about what I learn, under the GammaVibe name (YouTube, blog, the occasional course). I also make indie desktop apps under Sunset Mesa Software: buy-once, no subscriptions, BYOK or local models, your data stays on your machine.
Interests: TTRPGs, hiking, camping, off-roading.
https://mirkof.com