For $1.3 million a month, OpenClaw founder Peter Steinberger runs 100 AI agents that code, review PRs, and find bugs

The more pointed detail here is organizational, not technical: three people managing 100 concurrent agents implies that human oversight is already the bottleneck, not model capability or compute availability. The $1.3M figure is less a cost benchmark than a signal about who gets to run these experiments and therefore who shapes the norms.

This is largely disconnected from recent activity in our archive, as we have no prior coverage to anchor it to. It belongs to a cluster of stories about agentic coding infrastructure that includes Cognition's Devin rollout and various Codex-adjacent announcements, where the recurring tension is between headline capability claims and actual deployment economics. Steinberger's setup is notable precisely because it makes the economics explicit rather than burying them, which most vendor-adjacent coverage avoids doing. That transparency is useful for practitioners trying to model what scaled agentic development actually costs before token pricing compresses.

Watch whether OpenAI adjusts Codex pricing tiers within the next two quarters in response to customers like OpenClaw normalizing this spend level. If volume discounts appear targeting multi-agent workloads specifically, that confirms the inference commodity thesis the summary gestures at.