The 100X Engineer is Already Here

Before YC, I believed in the 10x engineer. I had worked with people who would walk into an unfamiliar system, understand the important parts, and ship a week's worth of work in one afternoon.

But YC broke my scale. I saw some people moving practically at light speed. These 100X agentic engineers had redesigned the way they worked using multiple agents in carefully crafted workflows. One agent explores a codebase. Another implements a bounded change. A third reviews and checks the result. They had designed the work to move in parallel and return verified results, without bouncing frantically between screens.

How do you quantify frontier agentic engineering?

That question became one of the reasons we built Mentlio. Recently, we studied a small set of AI-heavy YC teams over a 30-day window. We looked for the low-level habits behind the speed. Which tools were they using? Which workflows did they repeat? Did they use skills and rules? Did they delegate to subagents? Did they provide useful context? Did they verify the result? And, importantly, did completed engineering work show up on the other side?

In this 30-day window, one YC founder recorded 2,299 meaningful AI turns. That is a lot, but it was not the most interesting number. They used 32 distinct skills 227 times. Their most-used skill was a recurring loop, invoked 61 times. They used a merge-train workflow 20 times, a database-safety skill 14 times, a design-pipeline skill 13 times, a coding-agent training workflow 11 times, and a read-only-review skill 10 times.

They worked across Claude Code and Codex, with 156 observed subagent sessions spanning exploration, planning, general-purpose tasks, and research. In the same window, 294 merged pull requests and 13,934 Delivery Points were attributed to them.

The tempting explanation is that this founder is simply a coked-out genius. But the data points to something more useful: their leverage comes from a collection of small systems. They have a way to keep work moving, a way to merge it, a way to protect database changes, a way to review without accidentally editing, and a way to recover when an agent gets stuck. That is much more useful than calling someone a 100X agentic genius.

Make frontier behavior teachable

The reason to identify frontier engineers is not to put them on a pedestal. It is to find the practices that can spread. Once companies identify these engineers, they can replace generic advice like “use AI more” or “write better prompts” with specific coaching built around workflows that already work.

Mentlio helps engineering leaders find those patterns. When you compare engineers at the workflow level, coaching becomes specific. You can see that one engineer repeatedly uses a database-safety skill while the rest of the team handles the same checks manually. You can see that a strong investigator delegates repository exploration before implementation. You can see that another engineer keeps context clean or consistently asks the agent to test and verify its work.

True AI adoption is measured not by ranking people, but by finding the best internal workflows and helping them spread.

Methodology note: This article uses an anonymized 30-day snapshot of derived AI activity and delivery metrics from a small set of AI-heavy YC teams. Mentlio does not collect raw prompts, outputs, or source code. Delivery and AI activity were observed in the same window; these figures do not claim that AI caused every merged pull request.