Agents Need Platforms. Platforms Are Being Built Now.
Two case studies — Superset building an IDE for AI agents, General Intelligence building an agent platform with agents — show that the meta-layer of AI tooling is itself being assembled using AI. That loop is tightening.
Superset built an IDE for AI agents on Vercel. General Intelligence used agents to build an agent platform. These aren't just case studies about clever engineering — they're evidence that the tooling layer for autonomous software is being constructed by the very thing it's meant to support.
That recursive quality matters. When the people building agent infrastructure are already using agents to do it, the development cycle compresses in ways that won't show up in a press release. Features that would have taken a quarter get shipped in weeks. The rough edges get found faster because the system is also the tester.
For businesses that are still asking "should we explore AI agents?" — that question has a shelf life. The more useful question now is: what does your operation have that an agent could run repeatably, and what does it have that genuinely requires a human judgment call every time? The first category is larger than most founders admit.
The platform risk is real too. Building your agent workflow on a single vendor's sandbox and runtime means your operational continuity is coupled to their uptime and pricing decisions. That's a known risk worth pricing in — not a reason to wait.