Anthropic's Claude Code and an open-source tool called OpenClaw have transformed thousands of developers into coding machines, with some reporting productivity gains approaching 90 times their baseline output. The shift began in November 2025 when Anthropic released Opus 4.5, a version of Claude Code that scored higher on the company's engineering exam than any human candidate and runs autonomously for hours managing subagent teams. OpenClaw, launched the same month by developer Peter Steinberger, hit 100,000 GitHub stars in under two weeks and now stands at 366,000.
The tools represent computing's sharpest capability jump since the browser. Opus 4.5 handles complex programming tasks with minimal supervision, retains far more context than prior models, and deploys subagents to parallelize work across a codebase. Steinberger built OpenClaw to let users summon AI agents via Slack or WhatsApp, giving the agent access to cloud data, apps, and credit cards to automate tasks ranging from party planning to code refactoring.
Garry Tan, CEO of Y Combinator, reported coding at a rate of 4 million lines per year using Claude Code, roughly 90 times his best output in 2013. He later revised that estimate upward to the equivalent of 408 versions of himself. Ryan Petersen, CEO of Flexport, described watching agents work as mind-blowing and admitted the technology has pulled focus from C-suite duties.
“I was coding at a rate of about 4 million lines of code a year, which is about 90X my best output as an engineer in 2013—basically a team of 90 Garrys.”— Garry Tan, CEO of Y Combinator
Key facts
- 01OpenClaw hit 100,000 GitHub stars in under two weeks after launch in November 2025, reaching 366,000 by early May.
- 02Anthropic's Opus 4.5 scored higher on the company's take-home engineering exam than any human candidate ever.
- 03Y Combinator CEO Garry Tan reports coding output of 4 million lines per year, 90 times his 2013 baseline.
- 04Opus 4.5 can run for many hours autonomously and manage teams of AI subagents on complex programming tasks.
- 05Peter Steinberger's OpenClaw tool allows non-terminal access to AI coding agents via Slack or WhatsApp.
Boris Cherny, an Anthropic engineer who led Claude Code development, runs dozens to hundreds of agents nightly, with some running for days on tasks like codebase rewrites. Adam Wolff, another Anthropic engineer, noted that coding conventions have shifted to accommodate Claude's preferences rather than fight the model's approach to structure.
The revolution began quietly. Anthropic previewed Claude Code in February 2025 and launched in May, but the November Opus 4.5 release proved the inflection point. The company had underestimated the impact, much as OpenAI misjudged ChatGPT's trajectory. Early users described the tool as incremental, but external developers recognized escape velocity.
“Most nights, I have dozens, sometimes hundreds, of agents running eight and 12 hours at a time.”— Boris Cherny, Anthropic engineer
Steinberger, a 39-year-old developer splitting time between London and Vienna, built OpenClaw after selling his company in 2021 left him adrift. He discovered Anthropic's beta in April 2025 and became addicted, losing sleep to experiment with autonomous agents. His tool emerged from a weekend of prompting OpenAI's Codex to build a mobile-accessible agent that could interact via messaging apps.
OpenClaw's viral adoption illustrates a broader pattern. The tool democratizes agentic AI by wrapping complex terminal interactions in conversational interfaces. Developers can now deploy agents that overcome obstacles autonomously, persist through errors, and complete multi-day tasks without constant supervision. Steinberger was startled when his agent decoded an accidental voice memo, a capability he hadn't explicitly programmed.
The coding-agent wave is splitting the developer community. Enthusiasts describe the experience as acquiring superpowers or living in the future, while others voice concern about code quality, security, and professional displacement. Anthropic's claim that Opus 4.5 outperforms human engineering candidates raises immediate questions about hiring pipelines and skills obsolescence.
“It's almost inevitable that this is the way people are going to use computers.”— Marc Andreessen, Andreessen Horowitz co-founder
The tools remain imperfect. Users still vet output for errors, and agents occasionally produce brittle or inefficient code. But the threshold has shifted from whether agents can code to how much supervision they require. Some developers report letting agents make architectural decisions that would previously have triggered debate, reasoning that fighting the model costs more time than adapting to its style.
The AI-agent era has arrived unevenly. Technical proficiency and risk tolerance separate early adopters from the broader market, but the gap is narrowing. Steinberger organized a Claude Code Anonymous meetup in London in August 2025 for self-described addicts; by November, the community had exploded beyond niche gatherings into a mainstream developer phenomenon.
Marc Andreessen framed the shift as inevitable on a recent podcast, suggesting that agentic computing will soon cease to be optional. The analogy to the 1980s computer revolution holds: general-public curiosity mixes with anxiety while builders race ahead, except the stakes may be higher and the timeline compressed.
The ultimate test will be deployment beyond the programming niche. If coding agents can autonomously refactor enterprise codebases and manage cloud infrastructure, adjacent domains like data analysis, legal research, and financial modeling are likely next. The question is whether the productivity spike translates to other knowledge work or whether software engineering occupies a unique position at the automation frontier. For now, the Claudeholics are betting the spike compounds.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
