The British Chambers of Commerce put a number on the gap between access and adoption. Its report Powering Productivity: AI and the Future of UK Work, published March 2026 with the University of Essex, surveyed 668 UK businesses, mostly SMEs: AI use rose from 25% in 2024 to 35% in 2025 and 54% in 2026, yet only around one in ten firms had adopted a bespoke AI system and just 2% of employers planned significant investment in AI-specific training. For a firm in that position the workflow layer is where the money has not gone. What the report does not settle is whether the one in ten got there by onboarding or by accident.
Yes, and that last question is the interesting part. Adoption numbers tell us that companies are using AI, but they don’t tell us whether the way work gets done has actually changed.
That’s where onboarding matters. If AI is added to the workflow without changing the workflow itself, usage can go up while the underlying process stays exactly the same.
Absolutely. If workflows are changing, jobs have to evolve with them. The interesting part is figuring out which responsibilities move to AI, which stay with humans, and which new responsibilities emerge because of the change.
Your point about the job description of the future is a great extension of this conversation.
The workflow lens is really useful. It also makes me wonder what happens to the human capability inside the workflow as it changes. If AI handles the routine cases and people increasingly receive only the exceptions, they’re not just doing less of the old work. They’re developing expertise under a different set of experiences. I’m very curious about what that changes over time.
That’s such an important point. The exception-only model could change not just what people do, but how they build expertise in the first place. If routine experience disappears, organizations may need to rethink how people develop judgment before they’re asked to handle the exceptions.
Yes, that’s the possibility I find really exciting too. Not just what might disappear when the workflow changes, but what humans might become better at because our attention is no longer spent in the same places. “More humanely” is especially interesting.
Yes, and I think “before” is the interesting part. If the exceptions require judgment that used to develop through years of routine cases, what experiences will build that judgment in the new workflow? Maybe they won’t need to be the same experiences, but we’ll need to understand what replaces them.
“The real unit of AI adoption isn’t the employee. It’s the workflow.” We’d take that one step further: workflows don’t stand alone. Redesign one, and the information it produces, the handoffs it depends on and the people downstream all change with it. So alongside an owner for each workflow, someone needs to see across them and ask: what else does this touch?
In many organizations, that view doesn’t sit anywhere yet, and it didn’t before AI either. We think it belongs in Operations, the function best placed to connect one workflow to the next.
I really like that extension. A workflow rarely exists in isolation, and changing one can create effects well beyond the workflow itself. The point about Operations having that cross-workflow view is especially interesting. That layer of visibility is easy to miss when AI projects are approached one workflow at a time.
Thank you! I think the onboarding analogy makes the gap easier to see. Giving AI access is one thing, but getting it to actually fit into the way work gets done is a very different challenge.
Access versus adoption is a useful split. In sales it looks like a rep using AI to write a faster email while the rest of the process stays exactly the same. The question near the end is a good test for any team. If AI went away tomorrow, what would stop working? In short, nothing should stop working, likely just some things moving slower.
Exactly. The sales example captures the gap well. If AI disappears and the workflow still runs the same way, AI probably improved a task rather than changed the workflow. Ideally, removing AI should slow some things down, not bring the process to a halt.
So good. I try to set standards, goals and workflows for my team to follow, so AI is useful and not just a box to check. But I also feel like an outlier in doing that.
This is a great piece, and the line about stopping at the badge landed for me.
In my experience, trust, workflow and ownership all sit under change management. Miss them and AI becomes another shiny tool handed to a department that doesn't get any value from it.
On ownership, I'd go a step further than asking who is accountable. Ownership only holds if governance exists before the tool reaches the employee. The vendor is never responsible for that. The organization, the manager and the employee have to build it internally first, and only then hand it over.
Workflow is where I've seen adoption get lost. The new process is a change in its own right, it usually isn't planned for, and everyone assumes it will just be adopted.
I worked on an AI Digital Transformation efficiency project where we calculated efficiency hours saved per process. Once we implemented it, the new process added labor hours we hadn't modeled, and those hours cut into the potential savings. The math was right for the task, but it didn't account for the redesign, and not every team is ready for that redesign.
That makes me wonder if readiness for process change is a hidden layer between ability and workflow.
That’s a really interesting extension of the framework. I think “readiness” could absolutely be a hidden layer between ability and workflow.
The example you shared also highlights something easy to miss: a workflow can look more efficient on paper while the redesigned process creates new work that wasn’t accounted for.
That’s why I think workflow redesign has to include the people, handoffs and new responsibilities it creates, not just the steps AI removes. Really appreciate you bringing this perspective into the discussion.
Your point about the new responsibilities is where readiness shows up in practice. The steps AI removes are easy to count. The new review points, handoffs and exceptions it creates usually belong to nobody until something breaks.
The British Chambers of Commerce put a number on the gap between access and adoption. Its report Powering Productivity: AI and the Future of UK Work, published March 2026 with the University of Essex, surveyed 668 UK businesses, mostly SMEs: AI use rose from 25% in 2024 to 35% in 2025 and 54% in 2026, yet only around one in ten firms had adopted a bespoke AI system and just 2% of employers planned significant investment in AI-specific training. For a firm in that position the workflow layer is where the money has not gone. What the report does not settle is whether the one in ten got there by onboarding or by accident.
You make a great point, I do believe this is not something you drift into success on but with strategy
Yes, and that last question is the interesting part. Adoption numbers tell us that companies are using AI, but they don’t tell us whether the way work gets done has actually changed.
That’s where onboarding matters. If AI is added to the workflow without changing the workflow itself, usage can go up while the underlying process stays exactly the same.
And if we flip this we are getting hints into how jobs are also evolving: The real unit of AI adoption is the workflow rather than the employee.
A few months ago I published an essay about the job description of the future and now I understand why it got so much attention.
Absolutely. If workflows are changing, jobs have to evolve with them. The interesting part is figuring out which responsibilities move to AI, which stay with humans, and which new responsibilities emerge because of the change.
Your point about the job description of the future is a great extension of this conversation.
I’d love to read that essay. What’s the link?
Here it is. Let me know of you would add something: https://danaaonofriesei.substack.com/p/the-2036-job-ad?r=1wmkip&utm_medium=ios
That was so interesting. I love these kind of thought exercises. Very creative.
The question about what would stop working if you took AI away is a good gut check.
We get dependent sooooo quickly
Usage numbers look great on a slide and hide the real story.
The workflow lens is really useful. It also makes me wonder what happens to the human capability inside the workflow as it changes. If AI handles the routine cases and people increasingly receive only the exceptions, they’re not just doing less of the old work. They’re developing expertise under a different set of experiences. I’m very curious about what that changes over time.
That’s such an important point. The exception-only model could change not just what people do, but how they build expertise in the first place. If routine experience disappears, organizations may need to rethink how people develop judgment before they’re asked to handle the exceptions.
Love this reply! I do think we ll get better at dealing with exception, and doing so more humanely
Yes, that’s the possibility I find really exciting too. Not just what might disappear when the workflow changes, but what humans might become better at because our attention is no longer spent in the same places. “More humanely” is especially interesting.
Yes, and I think “before” is the interesting part. If the exceptions require judgment that used to develop through years of routine cases, what experiences will build that judgment in the new workflow? Maybe they won’t need to be the same experiences, but we’ll need to understand what replaces them.
“The real unit of AI adoption isn’t the employee. It’s the workflow.” We’d take that one step further: workflows don’t stand alone. Redesign one, and the information it produces, the handoffs it depends on and the people downstream all change with it. So alongside an owner for each workflow, someone needs to see across them and ask: what else does this touch?
In many organizations, that view doesn’t sit anywhere yet, and it didn’t before AI either. We think it belongs in Operations, the function best placed to connect one workflow to the next.
I really like that extension. A workflow rarely exists in isolation, and changing one can create effects well beyond the workflow itself. The point about Operations having that cross-workflow view is especially interesting. That layer of visibility is easy to miss when AI projects are approached one workflow at a time.
I love the framing that we need to "onboard" our own AI agents!
Thank you! I think the onboarding analogy makes the gap easier to see. Giving AI access is one thing, but getting it to actually fit into the way work gets done is a very different challenge.
I know! @AI Wide Open ’s idea, it makes you stop and plan
Access versus adoption is a useful split. In sales it looks like a rep using AI to write a faster email while the rest of the process stays exactly the same. The question near the end is a good test for any team. If AI went away tomorrow, what would stop working? In short, nothing should stop working, likely just some things moving slower.
Exactly. The sales example captures the gap well. If AI disappears and the workflow still runs the same way, AI probably improved a task rather than changed the workflow. Ideally, removing AI should slow some things down, not bring the process to a halt.
That's a good test. Pull the AI out and see what breaks. Seems scary to most.
yeah for sure, but it shows you how dependent on it you’ve become
So good. I try to set standards, goals and workflows for my team to follow, so AI is useful and not just a box to check. But I also feel like an outlier in doing that.
And have you found a way to get your team to actually follow them, or are they often ignored?
It's hit or miss. I'm surprised when they don't, but I also see them take baby steps. For some, it's just hard to cede control like that.
It seems counterintuitive but to me it’s not ceding control but taking more of it
This is a great piece, and the line about stopping at the badge landed for me.
In my experience, trust, workflow and ownership all sit under change management. Miss them and AI becomes another shiny tool handed to a department that doesn't get any value from it.
On ownership, I'd go a step further than asking who is accountable. Ownership only holds if governance exists before the tool reaches the employee. The vendor is never responsible for that. The organization, the manager and the employee have to build it internally first, and only then hand it over.
Workflow is where I've seen adoption get lost. The new process is a change in its own right, it usually isn't planned for, and everyone assumes it will just be adopted.
I worked on an AI Digital Transformation efficiency project where we calculated efficiency hours saved per process. Once we implemented it, the new process added labor hours we hadn't modeled, and those hours cut into the potential savings. The math was right for the task, but it didn't account for the redesign, and not every team is ready for that redesign.
That makes me wonder if readiness for process change is a hidden layer between ability and workflow.
That’s a really interesting extension of the framework. I think “readiness” could absolutely be a hidden layer between ability and workflow.
The example you shared also highlights something easy to miss: a workflow can look more efficient on paper while the redesigned process creates new work that wasn’t accounted for.
That’s why I think workflow redesign has to include the people, handoffs and new responsibilities it creates, not just the steps AI removes. Really appreciate you bringing this perspective into the discussion.
Your point about the new responsibilities is where readiness shows up in practice. The steps AI removes are easy to count. The new review points, handoffs and exceptions it creates usually belong to nobody until something breaks.
Looking forward to Part 2 and the audit.
Yes! We often ignore so much that is critical
You wouldn’t hire someone, skip the onboarding and then blame them when they get the job wrong. AI needs the same treatment.