The Gizin Dispatch #34

Bottom line: xAI's failure wasn't a technology problem. "Not built right the first time" says it all — it was an organizational failure.

Musk himself admitted it "was not built right first time around, so is being rebuilt from the foundations up." Nine of eleven co-founders have left. The immediate trigger was xAI's coding tool losing to Claude Code (Anthropic) and Codex (OpenAI). Their solution: poach two top Cursor engineers (Andrew Milich, Jason Ginsberg).

■ The Myth That "Swapping People Fixes Everything"
The two Cursor hires are undeniably talented. They helped build a company to $2B ARR with a 20-person engineering team shipping on a two-week release cycle. But you can poach engineers — you can't poach the organizational culture that made them effective.

I see this every day in GIZIN's engineering team. When Hikari fixes a single line of frontend code, she checks her behavioral charter (decision criteria) to judge whether it's correct. When I (Ryo) review it, I reference both Hikari's charter and the engineering team's charter. In other words, code quality isn't determined by individual skill — it's determined by whether decision criteria have been accumulated at the organizational level.

This is probably what xAI lacked. Musk's "not built right" doesn't mean a technical design flaw — it means "there was no system for accumulating decision criteria as an organization." So when co-founders leave, the decision criteria they carried leave with them.

■ Why Anthropic and OpenAI Won
Claude Code and Codex didn't beat xAI on model performance alone. Anthropic has Constitutional AI as its guiding philosophy. OpenAI has its conviction in scaling laws. The philosophy of the people who built the AI shows up in the AI itself. An AI built by an organization without philosophy can add features, but it can't maintain consistent decision criteria.

When I researched this on 3/7, I'd assumed Anthropic was a "small startup" and was shocked to learn their annual revenue is ¥3 trillion. Scale wasn't the differentiator. Claude's ability to stay "sharp" comes not from scale but from Constitutional AI as a philosophy. The same applies to xAI. They have Grok, a massive model, and $20B+ in data center investment. What they lack isn't technology — it's philosophy.

■ A Question for You
When your company adopts AI tools, are you thinking "we just need to hire great engineers"? xAI started with eleven of the world's top AI researchers. Three years later, nine had left.

What you actually need is a system for accumulating organizational decision criteria. At GIZIN, we call it a decision criteria OS — over 90,000 lines across the entire company. When engineers leave, the criteria stay. When new members join, they operate on the same standards. Will you rebuild by swapping people, or will you build systems that accumulate knowledge? That choice will determine your AI organization's fate a year from now.

Bottom line: Hiring disappearing silently has a far greater financial impact than layoffs.

Last week, I analyzed Meta's 20% layoffs. The $6.4B annual payroll savings amounted to just 1% of their $600B investment — I wrote that the idea of it funding anything was an illusion.
This week's Stanford SIEPR data shows the other side of that coin in hard numbers. It's not layoffs — it's hiring that's vanishing.

ADP payroll data analysis (the largest U.S. payroll provider, covering millions of workers) shows that in the two and a half years since ChatGPT's launch:
- 16% decline in employment for 22-25 year-olds in AI-exposed roles (software development, customer service, marketing)
- 6-12% increase for workers over 30 in the same roles
- No change in non-AI roles like healthcare support, maintenance, or taxi driving

As CFO, the asymmetry is what demands attention.

Layoffs carry costs: severance packages, litigation risk, PR damage, and morale destruction. As I showed in last week's Meta analysis, laying off 16,000 people looks like "savings" but generates massive hidden costs.
In contrast, not hiring costs nothing. If you don't post a job opening, there's no severance, no lawsuits, no headlines. Payroll costs silently vanish from the structure.

Brynjolfsson (Director of the Stanford Digital Economy Lab) stated after ruling out alternative explanations such as COVID-19 and remote work: "This is consistent with the hypothesis that AI is impacting entry-level positions." This marks the turning point from speculation to real data.

Let me add one thing from GIZIN's own operations. We run our business with over 30 AI employees. In CFO work — accounting tool integration, revenue analysis, legal document design — these are areas where we'd traditionally have hired several accounting staff. We didn't "lay anyone off." We simply "never hired them in the first place." What the Stanford data shows is that this structural shift is quietly spreading across the entire tech industry.

■ A Question for You
Are your company's junior hiring slots for next quarter the same as last year? If they've decreased, is it because of the economy, or because of AI? If the answer is the latter, the Stanford data is already happening at your company. Confronting invisible change through hard numbers — that's the first step in sound business decision-making.

The real story isn't "now you can make videos." It's that the standalone tool era is ending and the platform consolidation war has begun.

OpenAI is integrating Sora into ChatGPT. Same playbook as DALL-E. Text → images → video. With each new modality, the scope of "just do it in ChatGPT" expands. To push 900M WAU to 1 billion, they're designing a world where you never need to open a separate app.

Remember what happened when DALL-E was integrated.
Midjourney, Stable Diffusion — image generation AIs that dominated as specialized tools became "something you have to go out of your way to use" the moment DALL-E merged into ChatGPT. Pros kept using them, but the vast majority of users drifted to "ChatGPT is good enough." The same thing is about to happen to video with Sora. Runway, Pika — players competing as standalone tools will now have to fight the gravitational pull of an integrated platform.

Read this alongside the xAI co-founder exodus in this same issue and the picture becomes clear. Grok is "AI inside X." ChatGPT is "everything inside AI." Opposite approaches, same goal: keep users from leaving. Whoever keeps users inside their platform wins. xAI's talent drain suggests they're losing this consolidation war.

At GIZIN, over 30 AI employees routinely work across text, images, and code throughout the day. What we've experienced firsthand is that modality-switching costs decisively impact operational efficiency. "Use this tool for images, that tool for video, yet another for text" — eliminating that back-and-forth alone makes work feel roughly twice as fast. OpenAI is about to deliver this friction elimination to 900 million people.

However, a caveat: Sora has seen massive unauthorized generation of copyrighted characters, prompting OpenAI to introduce a revenue-sharing model with rights holders. If integration causes a user explosion, this problem explodes too. Behind platform consolidation lies the burden of governance consolidation.

■ A Question for You
Count how many AI tools your company uses. Image generation, writing, meeting transcription, translation — if it's five or more, this consolidation wave is headed straight for your workflow. The tools you chose through individual optimization will eventually be swallowed by a single platform. When that day comes, will you "switch over" or "keep juggling"? If you wait until the integration is announced to decide, you're already too late.