The jagged frontier was a measurement error — here's what actually smoothed it, why it's accelerating, and 3 prompts to map where your work sits on the new spectrum

The “jagged frontier” framing that shaped three years of AI strategy was never describing AI intelligence. It was describing what happens when you remove all organizational structure from the work and ask for one answer in one shot.

The evidence: a general-purpose coding harness built by Cursor solved a research-grade spectral graph theory problem after running for four days with no human guidance, no domain-specific modifications, and no mathematical reasoning machinery. It didn’t just solve it. It improved on the human-written solution. And four organizations independently arrived at the same structural answer: decompose, parallelize, verify, iterate.

The skill that survives this transition isn’t doing the work. It’s evaluating whether the work is correct. That shift is already underway, and it’s moving faster than the model improvement curve would predict.

This is Part 1 of a three-part series. Part 2 covers how knowledge organizations should restructure around this shift. Part 3 covers the new categories of value creation that open up when the execution layer is automated.

The breakdown:

• Why the Cursor result matters more than purpose-built math systems. When a coding harness outperforms a domain-specific agent, you’ve learned something about harnesses, not math. The generalization is the finding.

• The verifiability spectrum. A framework for classifying your work into machine-checkable, expert-checkable, and genuinely judgment-dependent tiers — and why the judgment-dependent bucket is smaller than you think.

• Why architectural advances beat the model curve. Organizational insights transfer across domains at near-zero cost. That’s the mechanism behind the METR data showing task completion horizons doubling every four to seven months.

• What “sniff-checking” actually means for your career. The evaluation meta-skills in your domain that are about to get more valuable, not less — and the ones that are about to get cheaper.

But first — how a tool built to write code ended up producing better mathematics than the mathematicians’ own solution, and what that means for the way you work right now.