Peter Steinberger, creator of OpenClaw, and his team published a billing example showing roughly $1.3 million spent over a 30-day period for about 100 Codex agents. That works out to roughly $13,000 per agent per month.
What the spending covered
According to the breakdown, much of the agents’ activity was not pure software development. The work list included community handling, bug cleanup, issue triage, managing regressions, code reviews and dealing with support noise. These tasks belong to the operational layer around software rather than to the creation of new features.
Why that distinction matters
If agents primarily perform maintenance and operational work, the implications differ from the simple narrative that “AI replaces one programmer.” The AI can absorb the messy operational layer: duplicate issues, Discord reports, failing tests, benchmark drift, security checks, stale pull requests and user complaints. Those are repetitive, high-volume tasks that currently consume significant human time.
Cost perspective
Seen from one individual’s bill, $1.3 million in a month looks extreme. From the perspective of a company exploring labor restructuring—especially where large volumes of operational work exist—that sum can be interpreted as comparatively inexpensive for the scale of automation being tested.
What this could mean going forward
If this model is commercialized and unit costs fall, the major consequence won’t simply be “100 agents writing code.” More consequential could be 100 agents taking over the invisible, operational tasks that have sustained many software jobs. That shift may reshape software team composition, workforce needs and operational cost structures.
Summary
The OpenClaw billing example highlights that AI agents are capable not only of generating code but also of handling the maintenance and support workload around software. If such agent deployments become practical and cheaper, the resulting changes to jobs and operations could be substantial.
