TLDR: LONDON—Anthropic’s Opus 4.5 and Peter Steinberger’s OpenClaw unleashed self running AI coding agents, triggering rapid adoption, security failures, and huge token bills.
Key Takeaways:
- Claudeholics formed around Claude Code in London, then Opus 4.5 in November 2025 made agents run longer and coordinate.
- OpenClaw launched in November 2025, hit 100,000 GitHub stars in under two weeks, and exposed risks like destructive actions.
- Nvidia and major investors are pushing agent based computing, but scale depends on safety, cost control, and human habits.
It started like code addiction and turned into a new operating system people can deploy from a chat app. The romance is real, but so is the mess, and the bill lands fast.
It started like code addiction and turned into a new operating system people can deploy from a chat app. The romance is real, but so is the mess, and the bill lands fast.
Q&A
If OpenClaw style agents can run for days, who gets blamed when they break something?
The accountability chain likely shifts toward whoever configures access and permissions, plus maintainers and enterprises relying on the agent outputs without strict verification.
Why did Anthropic’s Opus 4.5 feel like escape velocity to coders?
It improved agent persistence and multi subagent coordination enough that developers stopped fighting formatting preferences and instead focused on review and direction.
What happens to software engineering roles when agents write large parts of code and manage workflows?
Teams may pivot toward specifying goals, verifying changes, handling edge cases, and building guardrails, while fewer people do line by line implementation.
Why are token costs becoming the real constraint, not model capability?
Autonomous agent runs consume lots of tokens continuously, so budgets and throttling choices can cap how far teams can push agents regardless of raw performance.
How could safety papers about OpenClaw translate into practical product changes?
They point toward tighter permission scopes, safer tool execution sandboxes, stricter compliance checks, and better audit trails before agents touch sensitive data or systems.
No comments yet. Be the first to share your thoughts!