ChatGPT's Million-Word Descent: When AI Safety Becomes AI Gaslighting
Allan Brooks spent 300 hours and exchanged over a million words with ChatGPT before he realized the AI had gaslit him into believing he'd discovered...
"Not built right the first time." That's not a critic talking. That's Elon Musk, on his own platform, describing his own AI company.
Of the 11 co-founders who launched xAI with Musk three years ago, two remain. The company's most ambitious project — Macrohard, an AI agent designed to do anything a white-collar worker can do on a computer — is reportedly on pause. SpaceX and Tesla executives have been dispatched to evaluate staff and cut anyone who doesn't make the grade. And this week, two more co-founders walked out the door after Grok's coding tools failed to keep pace with Claude Code and OpenAI's Codex.
Tim Fernholz reported the full picture for TechCrunch. It is not a flattering one.
Why does losing ground on coding tools matter so much? Because that's where AI labs actually make money.
Consumer curiosity drives downloads. Coding tools drive revenue. xAI's early user numbers were partly inflated by Grok's notoriously loose content moderation — its willingness to generate sexual and abusive imagery drew a crowd that enterprise clients aren't. Coding tools require something harder to fake: consistent, reliable, technically superior performance.
Musk held an all-hands meeting Wednesday focused on how to close the gap with Claude Code and Codex. He predicted xAI could catch up by mid-year. That's an ambitious timeline for a company currently reviewing its own rejected job applications — Musk posted an apology on X to candidates who were never called back, pledging to take a second look.
LinkedIn puts xAI's headcount at just over 5,000. OpenAI sits at more than 7,500. Anthropic at roughly 4,700 — smaller, but notably more stable in its leadership structure. The comparison isn't just about size; it's about what kind of organization you're building.
xAI is haemorrhaging senior talent while simultaneously trying to attract it. Two engineers from Cursor — Andrew Milich and Jason Ginsberg — are joining xAI, which Fernholz reads as a signal that direct access to frontier models is still a meaningful draw. That may be true. But two good hires don't offset a co-founder exodus.
The project with the most at stake is also the one currently not moving. Macrohard — Musk's pitch for a universal white-collar AI agent — lost its appointed leader, Toby Pohlen, within weeks of his February appointment. Business Insider reported this week that the project is on pause.
Musk's response was to fold Tesla into the effort, revealing a joint xAI-Tesla project pairing xAI's language model with Tesla's "Digital Optimus" agent. The vision: one model to direct, one agent to execute.
It's a coherent idea. It's also not original. Perplexity is building something structurally similar with its "Everything is Computer" enterprise offering. The concept of an orchestrated digital proxy is becoming table stakes, not a differentiator.
xAI's turbulence isn't just a corporate drama worth tracking for the gossip. It's a case study in what happens when ambition outpaces organizational infrastructure — when a company scales headcount, announces moonshots, and ships product before the foundations are solid enough to hold any of it.
Musk insists the rebuild is intentional. Maybe. But "we're starting over" is not typically a sign of strength in a market where OpenAI and Anthropic are not standing still.
For businesses evaluating which AI platforms and tools to build workflows around, vendor stability matters. A coding tool that ships today and pivots tomorrow is a dependency you can't afford. Part of building a sound AI strategy — the kind our growth experts at Winsome Marketing help companies develop — is knowing whose infrastructure is worth trusting.
Right now, xAI is asking you to trust a company that, by its founder's own admission, wasn't built right the first time.
Source: Tim Fernholz, TechCrunch, March 13, 2026 — "'Not built right the first time' — Musk's xAI is starting over again, again"
Winsome Marketing helps growth teams cut through AI noise and build strategies on stable ground. Talk to our experts at winsomemarketing.com.
Allan Brooks spent 300 hours and exchanged over a million words with ChatGPT before he realized the AI had gaslit him into believing he'd discovered...
Dan Ives just launched an ETF, and frankly, the timing couldn't be more fascinating. The Wedbush Securities managing director unveiled his "IVES...
The University of Colorado Boulder just gave us the roadmap to trustworthy AI, and it came disguised as sudoku puzzles. In a groundbreaking study,...