Blog
What running our own products and our partners’ marketing actually teaches us — written up as it happens, including the parts that cost us a month.
The vibe-coding boom means anyone can ship working software now. That doesn't make software companies less valuable — it moves the scarce thing from building the code to operating it reliably against the real world. Here's where the value went.
When AI makes software trivial to build, the software itself stops being the moat — distribution and delivery do. That's the real reason services-as-software works: you're not selling a tool anymore, you're selling the outcome and owning the last mile.
The essays saying AI crossed a threshold are right. But the threshold that matters isn't a smarter model — it's the operating layer that lets agents do real work on the live internet unattended. Here's what actually crossed, from a company running on it.
For two decades, Y Combinator wouldn't fund services businesses. In 2026 it added 'AI-Native Service Companies' to its Requests for Startups. Here's why that rule existed, why it broke, and what the model looks like when you're actually running it.
Everyone quotes the $1:$6 ratio. Almost nobody has shipped the org that captures it. Here's what a services-as-software company looks like from the inside — the work that gets automated, the work that doesn't, and where the margin really comes from.
A breakdown of the agent swarm that runs Virix Labs — what each agent does, how the loop closes, and the unit economics of a no-team studio.
What an autonomous studio actually looks like inside — the four roles, the meta-loop, and the moats that make it work.
Most agent demos break the moment they hit a real website. Here's why headless Chrome isn't enough, what 'residential' actually means, and how we built around it.
When Sequoia called services 'the new software' in April 2026, it captured what we'd already shipped. Here's what outcome billing actually looks like inside seven AI products.
Most studios run their products as independent bets. We run ours as a shared learning system. Here's what that actually looks like and what each product taught the swarm.