This isn’t fresh news. The trend of high-profile CTOs leaving C-suite roles to take IC seats at Anthropic has been building for over a year. Bailis left Workday. McCann left You.com. Parmar (one of the original “Attention Is All You Need” team) went from CTO at Adept to MTS. Shi from Super.com. Krieger stepped sideways from CPO into the Labs incubator. At least five names deep, probably wider.

What took me a while was understanding why it kept happening.

The two readings on offer were “SaaS is dead” and “the comp is good.” Both are partially true and neither explains the move. SaaS isn’t dead, most of these companies aren’t in trouble, and senior MTS comp at Anthropic doesn’t dwarf what a CTO at a public company already makes. If pay and panic were the drivers, the names on this list would look different.

The reading I’ve landed on is information arbitrage.

Being inside the lab gives you signal you can’t reconstruct from the outside, no matter how plugged in you are. You see the next two or three model generations before they ship. You see what quietly didn’t work. You see the failure modes that didn’t make the blog post.

That signal doesn’t pay back as your next title. It pays back as the quality of every call you make over the next decade, and not just at work. Where to put your savings. What to tell your kids to study. Where to live. Which skills are still worth deep investment and which are about to be cheap. It compounds quietly for ten years and then suddenly looks like genius in hindsight.

What’s most telling is that these CTOs were already as close as most operators ever get. They had internal data on AI in production, relationships at the labs, and large engineering orgs. They still concluded that wasn’t enough. They saw the gap and decided it was worth a career.

The follow-on that doesn’t get said often is that you probably don’t have to be inside Anthropic to capture most of this. The lab is the densest point of the signal, but the gradient extends outward. The companies deploying frontier AI at scale inside large enterprises see most of what matters: the integration failure modes, the behaviors that don’t reproduce in the demo, the places where the model is brilliant in a notebook and useless in a real workflow.

The CTOs aren’t downgrading. They’re buying a front-row seat during what may be the only window where the seats are still being assigned. And the rest of us are figuring out how close to that front row we need to sit.

Title compounds slowly right now. Proximity compounds fast.

That’s the trade.