xAI Admits It Was 'Built Wrong': Will Musk's Big Gamble Save Grok or Sink It?

On March 13, 2026, Elon Musk posted on X — his own platform — that xAI "was not built right first time around, so is being rebuilt from the foundations up." He added: "Same thing happened with Tesla."

The comparison is meant to reassure. It should do the opposite.

What Actually Happened

The day Musk made that post, Guodong Zhang — the co-founder who led Grok's coding agent and Grok Imagine, xAI's image and video generation tool — confirmed his departure. Zhang described his time at xAI as a "wild journey" on X, hours after Reuters reported that Musk had personally blamed him for shortfalls in xAI's coding products.

Earlier that week, Zihang Dai, a former Google researcher with a PhD from Carnegie Mellon, quietly removed his xAI affiliation from his X profile. Another exit confirmed.

That brings the tally to 9 of the original 11 co-founders gone since the company launched in 2023. The two who remain: Manuel Kroiss and Ross Nordeen. Everyone else — including Toby Pohlen, Jimmy Ba, Tony Wu, and Greg Yang — has left since January 2026.

Pohlen's case is the most telling: he was appointed to lead the ambitious Macrohard coding project, then left 16 days later.

The Product Impact on Grok Users

Zhang's departure isn't just a personnel story. He owned two of Grok's most-used capabilities:

Grok's coding agent — the feature meant to compete with Claude Code and OpenAI's Codex. Musk publicly admitted this product was falling behind rivals. Zhang was the person responsible for it. He is now gone.

Grok Imagine — xAI's image and video generation tool. Zhang oversaw it. He received an update approximately three days ago (enhanced video and image generation), and then its architect walked out the door.

For Grok users, this means the two areas where xAI was most actively competing — coding and multimodal generation — have just lost their lead architect, under circumstances that include a public blame-and-exit pattern from Musk himself.

The Money Makes This Worse

This is not a startup quietly reorganizing. In January 2026, Tesla disclosed a $2 billion investment in xAI in its Q4 earnings report. Days after that, SpaceX acquired xAI in a transaction valuing the combined entity at $1.25 trillion.

Musk is now telling the world that the company he just extracted $2 billion from Tesla to fund — the same company he merged with SpaceX at a $1.25 trillion valuation — was "not built right."

Tesla shareholders are already suing Musk for breach of fiduciary duty over xAI's founding. His public admission adds a fresh layer to those legal challenges. Meanwhile, xAI is also under government investigation in multiple jurisdictions over Grok's image generator, which previously enabled generation of non-consensual sexual images.

The Rebuild Plan

Musk's response is to import "fixers" from Tesla and SpaceX to review performance and implement cuts. He also reached out publicly to candidates xAI previously rejected: "Many talented people over the past few years were declined an offer or even an interview at xAI. My apologies."

Two senior engineers from Cursor — Andrew Milich and Jason Ginsberg — have joined, reporting directly to Musk. Cursor is itself rumoured to be in talks at a $50 billion valuation, making the hires a signal of intent, if not a solution.

Musk's stated target: catch up to OpenAI and Anthropic in AI coding by mid-2026.

The Watchdog Take: Big Gamble or Slow Sink?

Musk's Tesla analogy is the rhetorical move here, and it's worth examining. Tesla did go through a painful early rebuild. It also nearly went bankrupt, required multiple emergency funding rounds, and took years before producing a reliable product at scale. Users who stuck with Tesla during that period absorbed the risk.

The question for Grok users is the same: are you willing to absorb that risk now?

The case for staying: xAI has infrastructure ($20B third data centre announced March 10), aggressive hiring ambitions, and Musk's track record of eventually delivering at scale — even from chaos.

The case for switching: Grok's two most directly consumer-facing product areas (coding and image generation) just lost their architect. The company's internal culture, by its own employees' accounts, involves 19-36 hour shifts and a pattern of high-profile exits under blame. The rebuild timeline is "mid-2026" — months away. Claude Code and GPT-5.4 Codex are shipping now.

What This Means If You Use Grok

If you use Grok for coding assistance: This is the capability Musk himself flagged as behind — and the person responsible for it just left under pressure. Treat current Grok coding performance as the ceiling for the near term, not a floor.

If you use Grok Imagine: Zhang's departure removes the direct owner. Expect slower iteration on image and video generation while xAI stabilises leadership.

If you chose Grok for its "truth-seeking, unfiltered" positioning: The rebuild Musk describes — importing Tesla/SpaceX auditors, cutting underperformers, reaching out to rejected candidates — is a corporate turnaround playbook, not an AI research reset. The model may improve. The internal chaos is real and current.

If you're evaluating Grok for enterprise use: The combination of shareholder litigation, regulatory investigations, and a self-declared foundational rebuild is a procurement risk that procurement teams will flag.

Not sure if Grok is still the right fit? Run your AI tool match again — xAI's rebuild changes the stability calculus for users who need reliability over the next 6 months.

Sources: