Who gets to build private AI tooling

A designer at a bank opens an internal chat tool that already knows the bank's components, the bank's tone rules and last quarter's research. A designer at a six-person startup opens the same model everyone else on the internet is opening.

The gradient is unusually clean

Over 900 designers told Designer Fund and Foundation Capital whether they use internal AI tools, in a report dated September 16, 2026. The answers sort almost perfectly by headcount.

Company size	Using internal AI tools
2,000 and above	74%
501 to 2,000	48%
51 to 500	34%
1 to 50	26%

Nearly three quarters at the top. Roughly a quarter at the bottom. Internal tooling is the one part of the AI stack that money still buys outright.

The companies without it lean on it harder

Here is the part that makes this worth more than a resourcing complaint. In the State of Prototyping survey of 1,478 designers, 38.8% at startups said AI is now central to their workflow, against 34.7% at enterprises.

So the group least likely to have tooling built around their own context is slightly more likely to describe AI as central to how they work.

That is not ironic. It is causal in the obvious direction. A small team has no platform group, so AI is doing more of the work, with less of the company loaded into it.

What an internal tool actually holds

The advantage is not a better model. Everybody has the same models. The advantage is what sits in front of the model before anyone types.

An internal tool carries the component names your team actually uses, the decisions already made and closed, the phrases legal has rejected, the shape of last quarter's research. A designer at a large company gets that for free on every prompt. A designer at a small one supplies it by hand, or does without and edits the result.

The cost of doing without is not visible as a failure. It shows up as slightly generic output, slightly more editing, every single time, forever.

Most of it needs no platform team

The useful news is that the expensive part of an internal tool is the integration work, and the valuable part is the context. You can have the second without the first.

59.1% of designers have already built themselves some tool in the past six months, so the capability is not the blocker. The blocker is that nobody has written the company down.

Start with one file, about forty minutes, no tooling decision required.

CONTEXT.md  (keep it under two pages)

Who we are:      product, audience, one sentence each
Components:      the names we use, and what each is for
Decided already: choices that are closed, with the reason
                 (so nothing re-litigates them)
Never do:        phrasing, patterns and colours that are out,
                 and why
Words we use:    our term -> the term everyone else uses

Paste it at the top of a session, or point your tool at it. That is the whole mechanism an expensive internal tool provides, minus the integration.

Of those five headings, "decided already" is what pays for the file. Most editing of AI output is not error correction. It is re-rejecting options your team settled two years ago, one generation at a time.

Write it once, or pay for it daily

The gap between 74% and 26% is real and it is not closing on its own, because it tracks something small companies genuinely do not have, which is a platform team.

It does not track something they cannot have. Two pages of written-down context is the same asset, delivered badly. It gets stale, it needs re-pasting, nobody maintains it properly.

It also takes one afternoon, and it is the difference between a model that knows where it is and one that is guessing from the average of the internet.