From the Lab
Writeups on what we're building, what we're learning, and what we think about AI in 2026. Written for builders, not LinkedIn.
A productive day at Snowflake Summit '26 in San Francisco — the 'Making AI Real for Business' floor, a run-in with Mike Swift from MLH, and a look at their demo vibecoding inside the Snowflake IDE. Plus a 25-second walk-through.

Our biggest turnout yet, hosted in the Google space. We took people from a cold Claude CLI install to running it like a systems admin, got a non-technical attendee shipping in 30 seconds, and automated a co-founder's starscout platform live. Highlights reel included.

Bernt Wahl — mathematician, entrepreneur, and UC Berkeley Industry Fellow — has spent a career around the people who built the modern world. For the last stretch of ours, he's been opening doors, making introductions, and giving the kind of guidance that actually changes the trajectory of a team.

Our third time at the bi-weekly Gemini developer meetup. Live demos, a packed room, and a short clip from the floor.

Workshop #4 brought UX and UI professionals into the room to vibecode their own projects — from job interview portfolios to tennis court scrapers. Three hours in Downtown SF, hosted by Graybeam. Everyone shipped.

Workshop #3 brought a live podcast recording, a GRM build for one of our members, and a guest appearance from ENI6MA inventor Frank Dylan Rosario. Another solid session.

We set up shop at Giovanni's Cigar Shop in North Beach, went deep on image, and got a surprise visit from a Debian developer with 20+ years of experience. Lab #2 delivered.

We walked into PAI Palooza 2025 as a VeteranAI startup. We walked out as something else entirely. This is the story of what happened and why we don't look back.

starOS Labs sent four team members to the bi-weekly Gemini developer meetup hosted by Google DeepMind's Peter Danenberg. Here's what we heard, what we saw, and why it matters.

starOS Labs attended Stripe Sessions in San Francisco — two days of keynotes, Anthropic's revenue playbook, product launches, and real conversations on the expo floor.

Why Claude Code's slash-command interface — and the Python services built around it — is the most powerful, portable AI workflow available today.

What happens when you swap the stock OS on a Pixel 10, get root, install Termux, and realize you're carrying a full Linux development environment in your pocket.
The honest, non-hype explanation of how LLMs work — tokens, context, prediction, and why they hallucinate.

How to write prompts that actually work. Zero-shot vs few-shot, system prompts, and the single most important thing you can do differently.

The full story — picking a framework, writing strategies, paper trading, and deploying on Railway. What worked, what blew up.

Our honest, opinionated breakdown of the AI tools that actually save time vs the ones that just look impressive in demos.

How we set up a secure mesh network between our machines so Claude Code can orchestrate across all of them. Step-by-step.
