2026-05-29

When Claude Code Became a Movement

From a Claude / Claude Code builder’s perspective, this story is interesting because it’s not really about a single product launch — it’s about the moment coding agents stopped feeling like demos and started feeling like infrastructure. The article captures that shift through the people actually using these tools at ridiculous intensity, which is often where the real signal lives.

Key Points

The article argues that Claude Code, and then OpenClaw, helped kick off the practical age of AI agents for technical users.
Anthropic’s Claude Code evolved from an early coding assistant into something that could handle longer, more complex tasks, retain more context, and coordinate multiple subagents.
The “Claudeholic” culture grew fast: people were spending huge amounts of time coding with it, talking about it like a superpower rather than just a tool.
Boris Cherny, who helped build Claude Code at Anthropic, describes a progression from primitive automated coding to a system that could understand architecture and solve real programming problems.
The big turning point was the release described as Opus 4.5, which Anthropic said could run for many hours and manage teams of AI subagents.
Users like Y Combinator’s Garry Tan and Flexport’s Ryan Petersen became extremely deep users, treating Claude Code as a major multiplier on their output and attention.
Peter Steinberger extended the idea beyond terminal-based coding into a more general personal agent, first through Clawd and then OpenClaw, built on top of Claude Code or other coding tools.
OpenClaw is designed to access your data, apps, and even a credit card, then act autonomously in the background to complete tasks.
The project exploded in popularity on GitHub, eventually becoming the most popular open source project in GitHub’s history, according to the article.
The article emphasizes both the power and the risk: these tools are messy, imperfect, and potentially unsafe if used without care.
OpenClaw’s appeal is that it can automate not just coding, but a broader range of personal and professional work like ordering, tracking shipments, and coordinating logistics.
The story frames this as a larger computing transition: AI agents may become the default way people use computers, whether or not they explicitly choose that future.

My Take

What strikes me is how quickly the center of gravity moved from “AI helps me write code” to “AI is the interface for getting things done.” That’s a much bigger claim, and honestly a much more interesting one. Claude Code sounds like the kind of product that quietly changes habits first and ideology second: once you get used to delegating a hard coding task to a model that can keep going for hours, your tolerance for manual workflows drops fast.

I think the most compelling part of the piece is not the hype around AGI-adjacent language, but the concrete behavior it describes: people spending nights and holidays inside these tools, bending their work around them, and then building new tooling on top. That’s the real developer story here. When a tool starts inspiring wrappers, agents, and open-source extensions almost immediately, you’re no longer looking at a feature — you’re looking at an ecosystem.

At the same time, the article is very clear that this is not magic. Claude Code and OpenClaw are powerful, but they’re also risky, messy, and easy to misuse. I’d be very cautious about giving an agent broad access to apps, data, and especially money. That’s where the enthusiasm can turn into overconfidence really fast. I think the “Terminator persistence” framing is emotionally accurate, but operationally it’s exactly what should make developers pause.

If I were using Claude Code, I’d probably lean into the parts that are easy to verify: refactors, scaffolding, test generation, codebase exploration, and long-running background work with human review at the end. I’d be curious whether the broader agent dream actually survives contact with real-world permissions, edge cases, and accountability. That might be where the next wave of product work lives: not in making agents smarter in the abstract, but in making them safer, more observable, and less likely to do expensive nonsense while you’re away.

The takeaway: this is a genuine inflection point for Claude-style developer tooling, but it’s also a reminder that the leap from “impressive” to “reliable” is still where the hard work begins.