The Accommodation We Already Built

Almost every large employer in this country bought artificial intelligence in the last three years. They bought it for productivity. They counted seats deployed, adoption rates, hours saved. What almost none of them noticed is that they also bought the most capable accommodation technology ever built, and then pointed it somewhere else.

I noticed because I work with one hand. Every tool I open is either an accommodation or an obstacle, and I know which one within about a minute.

Think about what accommodation has always cost. An interpreter booked days ahead. A reader working from someone else's calendar. Equipment through three approvals and a budget cycle. Extra time granted as a favour by a person who has to be told why. And underneath all of it, the part that never appears in the policy: to get any of it, you have to disclose. You have to explain your body to the person who decides your next promotion, and then wait.

That is why take up has always been low. And low take up has always been read as low need.

Now look at what an ordinary enterprise licence does today, without anyone calling it accommodation. It drafts from an instruction. It captions a meeting live. It transcribes and summarises what was decided. It reads a document aloud. It describes a chart. It answers a question about a two hundred page contract without anyone reading two hundred pages.

Here is the part the productivity story misses, and it is not the part people expect.

Dictation is not new. It has existed for twenty years, and for most of that time it was transcription. You still composed every sentence in your head, spoke it word by word, and then went back to repair what the machine misheard. For someone whose hands were the problem that was an improvement. It was never a solution, because composing was never the easy part. And it failed most often on the people who needed it most, because the conditions that make typing hard frequently affect speech as well.

What is new is that you can stop composing. You say what you need and the thing arrives. That is not a faster keyboard. It is the first interface in forty years that does not assume hands.

Let me be specific about my own day, because this argument usually dies in abstraction. By the afternoon my hand is tired. It is always tired by then. So I stop typing. I say what I need and it happens. The file opens. The document gets found. The message gets written and posted. I am not speaking into a box and watching words appear. I am directing the work, and something else does the operating.

That is the part I did not expect. I assumed the value would be that I could speak instead of type. The actual value is that on the days my hand is finished, the work is not.

Now the part I have to be honest about, because there is a version of this argument that is far too easy.

None of that happened on its own. I had to work out how to set it up. Which tasks to hand over and which to keep. How to ask for something so that I got it back the first time instead of the third. And the piece no software has a setting for: how long my hand lasts before I have to stop, and how to build a day around that answer.

That is not configuration. That is knowing your own body and then translating it into a way of working, and it took me months. I could do it because I understand how these systems behave and I had the time to experiment. Most people have neither. If the only people who benefit are the ones who can reverse engineer it alone, then this technology will track the advantage people already had, which is the exact opposite of what it is capable of.

So when I say turn it on for everyone, I am not saying turning it on is enough. Availability is not access. Somebody still has to sit with a person and work out what their day actually needs, and in most organizations that person does not exist. That is why the capability keeps arriving and nothing changes.

It would be comfortable to read this as a small population. It is not, and the reason is that most of it never gets recorded as disability at all.

When an experienced employee quietly stops volunteering for the work that involves long documents, nobody writes down a barrier. It goes down as changing priorities.

When someone leaves after twenty two years, nobody writes down an interface. It goes down as retirement.

When a strong candidate withdraws partway through an online assessment, nobody writes down an obstacle. It goes down as not the right fit.

In every one of those cases the organization loses a person and records a reason that is not the reason. Most disability is acquired during a working life rather than present at birth, most of it is invisible, and prevalence rises steeply with age at exactly the moment a person's judgment is worth the most it has ever been worth.

In most organizations, nobody does. Not because anyone objected, and not because of cost. Most of it is already paid for.

It has not happened because the decision has no owner. Artificial intelligence rollout belongs to technology and operations, and it is measured on adoption and efficiency. Accommodation belongs to human resources and legal, and it is measured on requests handled and risk avoided. Neither function is looking at the other's problem. So the question never lands on anybody's desk, and an unowned decision is not made slowly. It is not made at all.

The instruction is simpler than the analysis. Do not build a request process. Turn it on for everyone.

If dictation, captions, transcription and read aloud are available to the whole organization, then using them says nothing about you. Nobody discloses. Nobody asks permission. The people who need it use it, and so does everyone else, which is precisely what makes it usable.

Then say it once, plainly, in the rollout announcement. Some of you will find these tools remove an obstacle that has been in your way for years. That is what they are for, and you do not need to ask anyone or explain anything.

Check your measurements before launch rather than after. If your dashboards count response time, keystrokes, or handling speed, they will quietly punish anyone working through assistive technology. When culture and measurement disagree, measurement wins, and nobody has to be a villain for it to happen.

And put the paid tier within reach without routing the request through a line manager. The features that carry access sit disproportionately in the paid tier. An organization that puts everyone on the free tier has provided the appearance of access rather than the substance of it, and will read the resulting silence as proof that nobody needed it.

The arithmetic is not close. Licensing an organization at the tier that actually works costs less per year than replacing one experienced person who left because the job stopped fitting. Which means this was never an accessibility expense. It is a retention decision that produces accessibility, and retention has a budget owner in every organization I have worked in. Accessibility often does not.

I built my own firm on a version of this. I use these tools every day, not as a demonstration and not as a preference, but because they are the reason the work gets done. I did not set out to prove a point about accommodation. I set out to run a firm, and the tools turned out to be the accommodation nobody had offered me.

That is what I want leaders to sit with. You have already bought it. It is deployed, licensed, and sitting on the machine of every person you employ. Somebody in your organization is working around a barrier this week that the software on their own laptop would remove, and they do not know it, because nobody told them and nobody decided they were allowed.

The technology is not the missing part. We already built the accommodation. What is missing is somebody deciding that it counts.