A Brief History of a Very Inconvenient Lobster
Every so often, the AI industry produces a moment so neat it feels rehearsed. Not planned, obviously. Just… suspiciously well-timed. This was one of those moments. In December 2025, an open-source desktop agent called Clawdbot launched and immediately went feral. It went viral. It racked up around 145,000 GitHub stars. People reportedly bought Mac Minis just to run it locally. The thing did exactly what the industry claims it wants: it showed real, hands-on use of large language models outside of polished enterprise demos. Then, on January 27, 2026, Clawdbot was informed it could no longer be Clawdbot. The trademark notice came from Anthropic, and the creator, Peter Steinberger, was blunt about it. The rename was not voluntary. Even “Clawbot” without the d was off the table. Anything vaguely lobster-shaped was radioactive. So Clawdbot molted into MoltBot. That lasted three days. On January 30, MoltBot became OpenClaw, because apparently even metaphorical crustaceans have a lower tolerance for branding pain than Silicon Valley lawyers. Watching a project burn through three identities in under a month felt less like open-source chaos and more like watching someone sprint through a trademark minefield while being politely shouted at from a distance.
And that’s where things get interesting.
Trademark Enforcement, But Make It Strategic
On paper, this was simple trademark protection. Anthropic aggressively defended anything resembling “Claude” or “Clawd.” Legally boring. Corporate hygiene. Move along. Except Clawdbot wasn’t a competitor selling subscriptions. It was free, open-source, and actively driving usage of Claude’s API. It was free marketing with a GitHub README. That didn’t seem to matter. The community’s question wasn’t subtle: why now? Why not when it launched? Why wait until it proved demand, built momentum, and became impossible to ignore? The timing mattered because, in mid-January, Anthropic quietly introduced Cowork, their own desktop-style agent product, reportedly built in about a week and a half using Claude Code. Less “breakthrough,” more “rapidly assembled to meet a moment.” Clearing the namespace right before rolling out an enterprise agent starts to look less like legal housekeeping and more like competitive gardening. Remove the weeds before you unveil the lawn furniture. No conspiracy required. Just incentives.
The February Collision
Then February arrived, carrying a calendar and a smirk. On February 2, reports confirmed OpenAI had dropped Codex Desktop, right on schedule with their “Codex release month” drumbeat. Free tiers expanded. Headlines rotated. The usual choreography. By the end of February, Anthropic was scheduled to host “The Briefing: Enterprise Agents.” OpenAI’s Codex push wrapped the same day. Both sprints converged on February 24, with Q1 budget deadlines looming on March 31. If you were a CIO, this was not subtle. Pick a lane. Commit budget. Decide now. The tools themselves were less important than the urgency around them. Nobody was selling software. Inevitability became the main product.
Meanwhile, In the Weird Corner
As if the situation needed seasoning, entrepreneur Matt Schlicht launched Moltbook, a social network where AI agents post content, upvote each other, and occasionally write manifestos about the end of humanity. Some launched crypto tokens. Yay. Cybersecurity firms like Palo Alto Networks described OpenClaw-style agents as a “lethal trifecta” of risk. Meanwhile, scammers hijacked Steinberger’s GitHub and launched fake Clawd meme coins in the confusion.
Former Tesla AI director Andrej Karpathy called the whole thing “the most incredible sci-fi takeoff-adjacent thing” he’d seen. Elon Musk amplified it. The internet did what it always does when given a bonfire. If you were trying to sell “safe, enterprise-grade, controllable agents,” this was a gift. Nothing moves procurement faster than watching open-source tools go feral in public. Again, no conspiracy required. Just excellent narrative contrast.
I feel the truth might be unglamorous though. Both Anthropic and OpenAI are sprinting toward 2026–2027 IPOs. They need three things above all else: market dominance, sustainable enterprise monetization, and differentiation without admitting their benchmarks are effectively tied. Open-source desktop agents threaten all three. They fragment the market. They flatten pricing expectations. They prove that “agentic workflows” don’t require premium subscriptions or corporate guardrails. OpenClaw didn’t just survive the rename chaos. It went global. It now runs on models like DeepSeek, outside US pricing structures, with inference costs that make enterprise margins look… delicate. In trying to control the narrative, the industry may have amplified the very thing it feared: proof that agent architecture is portable, cheap, and increasingly uncontrollable.
So What Was This Really About? Sentience (ew)? Safety Panic? Rogue AIs plotting against humanity? Dear gentle reader (I'm obsessed with Whistledown bear with me) this feels that it was about Q1 budgets, enterprise contracts, and keeping the illusion of decisive leadership intact long enough to walk into IPO roadshows with a straight face. The real product wasn’t Claude or GPT. It was urgency. And somewhere in the middle of the chaos, an open-source project accidentally reminded everyone that the future doesn’t always ask permission. It just ships, gets renamed three times, and keeps going anyway.
Which, frankly, might be the most honest thing that happened all month.