3 min read

The Flat Five Months: What AI Adoption Looks Like Before Results

Picture of Chad Walters Chad Walters : Updated on October 1, 2026

Engineering AI

October 1, 2026

The Flat Five Months What AI Adoption Looks Like Before Results

From June through October 2025, counting the way the board counted then, our monthly merged PR totals were 28, 45, 19, 40, and 24. Five months under 50. We were building the machines, but features were not getting faster.

Chad WaltersVP of Engineering, CampusIQ

What we mean by the flat five months

June through October 2025: five months under 50 merged PRs a month, while we built the skills, guardrails, review automation, and telemetry that would eventually turn feature delivery into a production line. Then November hit 72, and December closed at 91.

The chart

Then the Curve Broke

Then November hit 72, and December closed at 91. The curve broke and never came back down. The board deck’s own caption summed it up well: the hard part, five months under 50, until the machines.

Merged PRs. The board’s monthly totals.

June through October 2025, under fifty. November never came back down. December closed at 91.

The part that looks like failure

Nobody Publishes This Part

If you’re five weeks into AI at your company and the numbers haven’t moved, you’re looking at the part of our chart that looked like failure. For us, that part lasted five months. It’s easy to look at the chart now and focus on the inflection, but the flat period matters just as much.

Your team will eventually see both, and they need to understand that months of flat results don’t necessarily mean nothing is happening. In our case, a lot was happening behind the scenes. It just wasn’t showing up in feature delivery yet.

The machines

What We Were Actually Building

We weren’t focused on features or product during those five months. We were building the infrastructure that would eventually turn feature delivery into a production line: skills, guardrails, review automation, and telemetry. A second engineer started on August 12, right in the middle of that period, and the chart still didn’t move. That’s important because it showed us this wasn’t simply a headcount story.

The process mattered more. I didn’t care if something was a two-line change. I cared that everyone followed the same process because we needed as much determinism as possible around a fundamentally non-deterministic technology. For those five months, the process was the product.

October 16, 2025

The Date That Matters

October 16, 2025, was the turning point. Anthropic released Agent Skills, and five days later, the first commit landed in our own repo. We were working with the capability almost as soon as it existed.

My numbers moved first, jumping from 21 merged PRs to 62, then 81. At the company level, November climbed to 72 and December closed at 91. But the rest of the team didn’t see the same change yet. Excluding me, their weekly averages went from 11.5 to 7.5 to 8.5, showing that the capability was working before adoption had caught up.

Agent Skills released

October 16, 2025

Five days later, the first commit landed in our own repo.

My merged PRs

21 → 62 → 81

My numbers moved first.

The rest of the team, weekly

11.5 → 7.5 → 8.5

The capability was working before adoption had caught up.

The rest of the team, January to June

25 → 62 PRs per week

The rest of the team inherited them over time.

About two months

The Lag Is the Useful Part

The rest of the team started moving in January, climbing from 25 PRs per week to 62 by June. That roughly two-month lag taught us that capability can arrive before adoption. One person may get faster while the rest of the team stays flat, and that doesn’t necessarily mean the tools aren’t working.

For us, each improvement also made the next one easier to build. The person closest to the system saw the benefits first, and the rest of the team inherited them over time.

If you are trying to copy this

What We Learned From the Flat Five Months

If you’re trying to build something similar, expect a lag between the moment a new capability works and the moment it shows up across the team. Ours was roughly two months, although that’s our experience rather than a universal timeline.

The five flat months matter because they show what the later numbers leave out. We spent months building the systems behind the work before feature delivery reflected any of it. Then one person’s numbers changed while everyone else’s stayed flat, and only later did the rest of the team follow.

The flat part of the chart looked like failure at the time. In hindsight, it was the part immediately before the way we worked began to change.

We’re hiring a security leader, an engineering leader, and a VP of Sales. We’re looking for judgment and taste, proven by shipping. Have your AI send us what you built in the last 30 days.

About the author

Chad Walters

VP of Engineering, CampusIQ

Chad leads CampusIQ’s move to agentic engineering. Four engineers now merge about 1,150 production PRs a month, roughly the output of a 50-person team, with agents writing most of the code and engineers owning the outcomes, the quality bar, and the calls. He built the machine behind it: a standing merge queue, a fleet of estimation and review bots, executor-validator quality gates, and live telemetry on velocity and AI cost. Last month, more than 20 people across sales, ops, product, and onboarding shipped production code on the same system.