---
title: ['Getting', 'Wider']
description: "Working inside other organizations as a forward deployed engineer, the AI adoption curve I see is getting wider: at the front each step makes the next one cheaper, at the back each step not taken makes the next one more expensive. Cloud and git closed that gap eventually; I don't know whether this one does."
---

I've been doing some AI work at various organizations this year, which is a good way to see how AI is landing in other organizations. What I keep seeing is one curve with its two ends moving apart.

The usual picture of AI adoption is the Rogers bell curve[^1], innovators at the left through to the laggards out on the right tail (Contrarian to Rogers, I drew my illustration is the reverse direction). The promise built into that drawing is that the back of the curve ends up where the front is. Later, but there.

I don't think that is what is happening[^2]. The curve is getting *wider*. Where you stand on it pushes you further in the direction you already face, and the two ends are getting further from each other.

<picture>
  <source srcset="/images/2026-11-06-getting-wider-curve-dark.png" media="(prefers-color-scheme: dark)">
  <img src="/images/2026-11-06-getting-wider-curve-light.png" alt="A broad solid bell curve surrounds a narrow dashed curve of the same height. A penguin surfs down the right slope, while another tries to climb the left slope but slips backward." width="1774" height="887" style="width: 100%; height: auto; display: block; margin: 2rem auto;">
</picture>

At the front, every step makes the next step cheaper. The tests are in order because the agents needed them to be. Deploys are cheap and reversible because that was the price of letting [one PR through](/start-with-one/) on its own. Review is sorted by risk rather than by who is free, and nobody on the team stops to decide whether to trust the output anymore. Each of those cost something once and is paid for now, so every new change rides on all of them.

When you spend [half your day sharpening the axe](/sharpening-the-axe/) (aka building the harness), you will slow down. But eventually those speed gains *compound*. We hit 100 PRs in a day recently[^3], and a few days later 200 in a day is normal.

At the back, every step not taken makes the next one more expensive. The distance to what is visibly possible becomes a reason in itself not to start; improving an outdated process is a sunk cost. The first attempt fails the way first attempts do, and the failure confirms what everyone already suspected. The tooling change that went for review "in Q1" comes back approved after the model it was written for has been retired.

Both of those are happening this month, with the same tools at the same prices. One organization has agents merging to production on their own and another is still going through multiple rounds of human review.

Rogers put time along the bottom of his curve, and on that axis the back does arrive eventually; everyone gets to next year. What I'm drawing along the bottom is the cost of the next step, and on that axis the two ends are moving apart.

Precedent may be against me. Cloud went like this for a while and then closed up; being late cost you a few years and after that nobody could tell. Git did the same. Whether this one closes the same way, I don't know.

### FOOTNOTES

[^1]: Everett Rogers published [Diffusion of Innovations](https://en.wikipedia.org/wiki/Diffusion_of_innovations) in 1962, and the five adopter categories, innovators through to laggards, come straight from it.

[^2]: I said in [Don't Believe the Hype](/dont-believe-the-hype/) that early adopters don't necessarily beat fast followers. I'd still say it about picking tools; this is about the organization around the tools.

[^3]: The day itself is in [One Hundred, and Fifty](/one-hundred-and-fifty/); the system that moves a hundred changes through one person without thrashing is laid out in [the original 100 PRs post](/building-towards-100-prs-a-day/).