xAI 'Not Built Right' - Nine Co-Founders Gone

Less than three years after its founding, xAI is starting over. Elon Musk held an all-hands meeting this week where he told staff the company's AI coding infrastructure "was not built right the first time" and announced a ground-up reconstruction. Nine of the eleven original co-founders have now left. Two engineers hired from Cursor will attempt to rebuild what the founding team couldn't.

A Founding Team in Freefall

When xAI launched in 2023, Musk assembled eleven co-founders alongside himself - a roster heavy with ex-DeepMind and ex-OpenAI talent. The company raised at a $50 billion valuation in 2024 and hit $80 billion by the time it raised again. The founding team was the asset. Most of them are gone.

Zihang Dai and Guodong Zhang are the most recent departures, confirmed in March. Guodong Zhang, who ran Grok Code directly, was reportedly relieved of his duties by Musk. Two founders remain: Manuel Kroiss and Ross Nordeen.

The departures span every function - safety, infrastructure, architecture, product, and research. This is not attrition. It's a complete turnover of the technical leadership that built the company.

The Admission

At this week's all-hands, Musk said xAI's AI coding tools weren't competitive with Anthropic's Claude Code or OpenAI's Codex, and that the underlying system needed to be rebuilt from the foundations up. His characterisation was direct: the company's first attempt at a coding product simply did not work.

"xAI was not built right the first time, and now it needs to be rebuilt from scratch."

The context matters. xAI is competing in AI coding against Claude Code and Codex, products backed by Anthropic and OpenAI with years of iteration behind them. Grok Code, xAI's answer, has not gained comparable traction. That Musk acknowledged this publicly - at an all-hands, with the SpaceX merger just weeks old - signals the problem is visible even at the top.

Grok Code vs. Claude Code and Codex

The competitive framing in Musk's admission is worth taking seriously. Anthropic's Claude Code and OpenAI's Codex both launched as standalone tools and attracted developer audiences that xAI hasn't reached with Grok 4. Coding assistants are now a primary battleground for enterprise AI adoption, and xAI's absence from that fight is a strategic gap.

The AI coding tools market has become a key battleground - xAI's Grok Code has failed to keep pace with Claude Code and OpenAI Codex. Source: unsplash.com

Two Replacements for Nine Departures

To rebuild Grok's coding infrastructure, Musk hired Andrew Milich and Jason Ginsberg. Both departed Cursor on March 12 and will report directly to Musk. Their mandate is to reconstruct xAI's coding tools from the ground up.

Who They Are

Milich and Ginsberg co-founded Skiff, a privacy-focused document collaboration tool acquired by Notion. Milich subsequently led Notion Mail; Ginsberg was Skiff's CTO and previously worked at Apple and Sequoia Capital. They joined Cursor together in June 2025 as co-heads of product engineering, and were there for nine months before this move.

Cursor reached $2 billion in annualized revenue faster than any software company on record. Whether the two engineers responsible for the product side of that growth can reproduce it at xAI is an open question, but the pedigree is credible.

What two engineers can't replicate is a founding team of eleven. The institutional knowledge that left xAI over the past year - in model architecture, infrastructure design, safety systems, and research culture - doesn't transfer via a hiring announcement.

The Merger Math

On February 2, 2026, SpaceX absorbed xAI in what Musk described as the largest merger of all time. The combined entity was valued at $1.25 trillion: SpaceX at $1 trillion, xAI at $250 billion. The deal was structured partly to support the planned SpaceX IPO, which is targeting a $50 billion raise at up to a $1.5 trillion valuation.

Tesla invested approximately $2 billion into xAI in January, taking Series E preferred stock at roughly a $230 billion valuation. That stake was subsequently converted to a minority SpaceX position after the merger closed. Tesla shareholders filed breach of fiduciary duty lawsuits over the investment; those were dismissed in February.

Elon Musk has been running multiple major companies simultaneously while managing xAI's transition through the SpaceX merger. Source: commons.wikimedia.org

The Macrohard project - announced March 11 - pairs xAI's Grok reasoning engine with a Tesla AI agent that controls computers via real-time screen video. It runs on Tesla's AI4 chip. Toby Pohlen, one of the departed co-founders, had been leading Macrohard before his exit. His replacement hasn't been announced.

Counter-Argument

The bull case for xAI is straightforward: Musk admitted a problem, identified the gap, and moved quickly to address it. Companies that don't acknowledge failure tend to repeat it. Bringing in Milich and Ginsberg - engineers with a recent, verifiable track record at one of AI's fastest-growing products - is a more targeted hire than a generic senior engineering search.

The merger with SpaceX also provides infrastructure that an independent xAI couldn't access: orbital data centers, hardware supply chains, and a path to public markets that changes the capital structure. The Tesla AI4 chip partnership gives xAI a low-cost inference platform that competitors can't easily reproduce.

Musk has rebuilt organizations before, often from worse positions. There's a non-trivial scenario where this restructuring produces a more focused company than the original.

What the Market Is Missing

The SpaceX IPO creates a timeline problem. A $50 billion raise at a $1.5 trillion valuation requires the combined entity to present credibly as an AI company, not just a rocket manufacturer. xAI's coding product is the most visible AI consumer product in the portfolio, and it doesn't work well enough by its own CEO's assessment. That admission, made weeks before a critical pre-IPO period, is the part the market should be watching.

The safety picture adds a second layer. Jimmy Ba, who ran research and safety, left February 11. The safety organization he led has been described by former employees as effectively inactive. Grok generated over one million sexualized deepfake images in January before controls were applied. For a company heading toward a public listing, the liability exposure from a degraded safety function isn't a footnote. The AI safety exodus at xAI predates this week's engineering reset - and the engineering reset does nothing to address it.

The co-founder exodus leaves xAI with a radically different leadership composition than the one that originally attracted investors. Source: mapbox via unsplash.com

Two engineers from Cursor can't close that gap. What xAI needs is a functioning safety program, a competitive coding product, and a stable engineering leadership - all at once, with an IPO window opening. The new hires address one piece. The other two remain unresolved.

Sources:

• TechCrunch - 'Not built right the first time': Musk's xAI is starting over again, again
• OpenTools - Elon Musk's bold rebuild of xAI amidst leadership exodus