The real barrier to AI adoption is not fear. It is the first task.

Most AI-adoption commentary is written by and for software engineers. It talks about IDE integration, agent harnesses, prompt engineering. For most people in a business, none of that is the barrier.

The barrier is smaller and more mechanical than fear. Their environment is not set up. They have not done one real task of their own with it yet. They get handed something that already has AI built into it and are asked to approve steps they do not understand. That is not scepticism. It is a threshold nobody has walked them across.

The fix is not a training deck. It is sitting with someone through one real task, their task, start to finish. I have watched the light go on more than once: from no AI to what I would call an ally, in the time it takes to do one thing that actually mattered to them.

Part of what lowers that threshold is which tool sits in front of them. Claude's auto mode reduces the number of decisions being pushed onto someone who has no way to evaluate them, and it removes a layer of risk at the same time. Fewer approval prompts a non-technical person cannot judge is not a convenience feature. It is most of the barrier, removed.

One person walking one colleague through one task does not scale to a company. That is the next problem the fix creates, and it is the one I am solving for now: a tool-agnostic AI gateway and marketplace, where allies across the business surface the skills that help people in their own department do their jobs to a standard, using AI and what the company already knows, gated behind their existing access. People are already sending skills in.

Engineering sits apart from this. They already have repository access, so contributing there was always a soft option, and what they build rarely applies company-wide. What I get from their involvement is not distribution. It is added standards, different perspectives, and collaboration on the schema itself before it goes out to everyone else.

Here is the honest part. Right now, I personally collect, verify, and tailor every skill before it goes into company-wide distribution. That will not scale, and I know it will not scale. It is a deliberate bottleneck, not an oversight: the only way to build a shared standard is to work closely with every early contributor while there are still few enough of them to do that properly. Centralise control just long enough to learn what good actually looks like. Then push it outward.

I am the bottleneck today, on purpose.

If you are the one responsible for getting AI in front of people who are not engineers, the fix is not a rollout plan. It is sitting with one person through one task that is actually theirs. Everything after that, the marketplace, the gateway, the shared standard, is just what happens once enough people have crossed that one threshold.