OpenClaw founder on why going viral can break an AI startup

This article summarizes an episode of Y Combinator’s video series featuring Peter Steinberger, founder of OpenClaw.

Peter Steinberger, founder of OpenClaw / Photo credit: Y Combinator

AI startups are learning that sudden popularity can be dangerous, forcing small teams to handle complex security issues before their software is even finished.

Peter Steinberger, founder of OpenClaw, warns that growing too fast traps developers in a never-ending cycle of maintenance and hardware problems.

Feature velocity creates permanent maintenance debts

Steinberger advises founders to ignore business theories and build to solve their own daily frustrations. Because AI generates code instantly, accepting outside contributions seems productive but forces the core team to support complex configurations indefinitely.

Founders must follow guidelines to avoid early feature bloat:

• Target specific annoyances: Focus on recurring work issues rather than theoretical enterprise needs.
• Test internally first: Share the simplest tool with peers to gauge emotional reactions rather than polite feedback.
• Reject unnecessary updates: Limit volunteer developer contributions to prevent the software from becoming complicated to manage.

Prioritizing personal utility over collaborative scale keeps developers grounded in reality and avoids bloated roadmaps. The benchmark for success is simple, as Steinberger insists that “user number one should be you.”

Media scrutiny amplifies early security flaws

Scaling these personal projects into collaborative tools exposes startups to corporate scrutiny:

• Relying on single AI model providers creates failure points when access rules change.
• Simple warning labels fail to satisfy early users who demand stable commercial systems.
• Addressing continuous vulnerability reports drains engineering time away from progress.

The backlash can damage a brand before it matures. Steinberger experienced this firsthand when the press claimed 20% of OpenClaw skills were malicious. “In the months after the release, we got absolutely crushed by security reports,” he says.

His team scanned all 67,000 skills and found the real threat level was 0.3%. As he puts it, “a correction never travels as far as a scare.”