What does Adaptive Edge do?

Adaptive Edge is a UK strategy and innovation consultancy. We help organisations decide what to change, then make the change real: strategy built with the people who must carry it, AI adoption inside real work, and behaviour change that sticks.

How is Adaptive Edge different from a traditional strategy consultancy?

We do not hand over a deck and leave. Strategy is made in working sessions with your own teams, so the people who must deliver it own it from day one. We prove an approach in one team before asking the whole organisation to change.

Who does Adaptive Edge work with?

Global organisations across healthcare, industrials, media, retail technology and financial services, including ITAB Group, Freudenberg, Sky and Sandvik Coromant. Engagements run across Europe and beyond, remote and on site.

How does Adaptive Edge approach AI adoption?

We put AI to work inside real workflows rather than running abstract training. Teams learn on their own live work, we measure what actually changes, and we scale only what proves useful.

What does a typical engagement look like?

Most engagements combine focused research, co-created working sessions, and hands-on tools built for your context. They range from a single workshop to a multi-phase transformation programme.

How do we start a conversation?

Use the contact form on the site. We usually begin with a short call to understand the decision you are facing, and we will tell you plainly if we are not the right fit.

Managing Models, Managing People

By Nathan Waterhouse · 2026-08-28

I have spent a surprising amount of time this year teaching AI systems to write like me. Lists of banned phrases. Examples of my published work fed in as reference. Automated checks that scan a draft for the tells I hate before I ever see it. The models are extraordinary at the skill itself; they can produce competent prose on anything in seconds. Getting them to sound like me, to carry the judgement and taste that makes something mine, has taken months and is still not finished. None of that effort has changed the model itself. I could change it, in principle. An open model can be trained on everything I have written, and plenty of teams now do exactly that. I have not, and nor have most people, because it costs money and expertise I would rather spend elsewhere, and because a model I trained last spring would already be behind the one I can rent today. So what has built up over the months is not the model but a pile of documents: the rules, the examples, the checks. Some tools now hold that pile for me and hand it back at the start of each session, which looks like memory but is closer to filing. Everything that makes the writing mine sits outside the model, in those documents.

I was talking to Steve Baker about organisational culture when it struck me that I was dealing with a version of the cultural fit problem. The phrase has a bad reputation, deservedly, because it too often meant hiring people who look and think alike. Underneath the misuse sits a real management problem, and leaders have wrestled with it for as long as there have been organisations. Someone arrives perfectly able to do the job and still takes months to become effective, because what they are absorbing in those months is not the job. It is how things are done here: the standards, the shortcuts, the unwritten rules about what good looks like. Almost none of it has ever been written down, because until recently nobody needed it to be.

The comparison holds in an uncomfortable way. A model arrives with more raw skill than anyone I could hire, so none of the effort goes into capability and all of it goes into fit. The difference is that a person eventually finishes their induction and carries it with them afterwards. A model finishes nothing. It could, if I retrained it, but that is a bigger undertaking than most of us will take on, so the induction simply runs again tomorrow.

The question that interests me is what all this tuning is doing to the way leaders think about the people they manage.

The mirror

Most executives now manage two workforces. One is human. The other is a growing collection of models, agents and assistants, briefed, corrected, and increasingly given real work. Much has been written about borrowing from how good human teams are managed and applying it to teams of agents; I made that argument myself in What Managing Taught Me About AI, where the lessons of delegation, goals rather than tasks, turned out to map straight onto agents. Does the idea travel the other way too? Do the habits leak back? Does how we manage AI start to affect, positively or negatively, how we manage the real people on our teams? Management habits rarely stay politely on their side of a line, and I think these leak in two directions.

The dark direction treats people the way we treat models: as interchangeable capability, prompted harder when output disappoints, swapped for a newer version when a better one ships. The comparison turns personal quickly. The model did this for me in four minutes, so why does my team need a week? Once the model becomes the yardstick, a person is being measured against something with no bad days, no competing priorities and no stake in the outcome. The vocabulary is already drifting this way. When a leader spends the morning comparing model versions on benchmark scores, it costs something to spend the afternoon remembering that the sales team cannot be upgraded by switching providers, and that the analyst whose report disappointed does not come with a settings panel.

This is more than a worry about vocabulary. Hye-young Kim and Ann McGill ran five experiments on what they call AI-induced dehumanisation, published last year in the Journal of Consumer Psychology. People exposed to an AI that displayed emotional capability went on to rate other humans as less human-like, and the effect showed up in what they did rather than only in what they said: they were more supportive of workplace practices such as tracking devices and meal replacements, and less willing to give a bonus to a fund for employee mental health. What triggered it was not the machine's competence, because advanced cognitive ability on its own produced no effect at all. It was the appearance of emotional intelligence, which is precisely the quality this generation of assistants is being tuned for. The participants were members of the public rather than managers, so the finding does not transfer straight into a leadership team, but the mechanism is the one I would want to watch for in my own behaviour.

Organisations that already saw their people as containers of skills were always at risk of treating them as swappable. It is an easy view to hold without noticing you hold it. A person becomes the list on their CV, so replacing them turns into a procurement problem: find someone with the same list. What the list never carries is the part that took longest to build: who to ask when the official process fails, which rules can be bent and which cannot, and why the thing that looks broken was built that way on purpose. Working alongside models every day makes that view easier to hold, because with a model it happens to be true. The model is the list, and nothing else has accumulated. The organisations that choose to retrain rather than swap are refusing to carry that view across. They treat accumulated fit, everything a person has absorbed about how the place works, as an asset worth more than the cost of new skills.

The hopeful direction runs the other way, and I think it is within reach. Working seriously with models turns out to be a course in the fundamentals of good management. A model given a vague brief produces confident rubbish; so does a team. A model given worked examples outperforms one given abstract instructions; so does a new hire. A model performs best when specific feedback is close at hand and context is generous, and degrades when either is withheld; every person I have ever managed behaves the same way, with one difference that flatters the people. A person internalises the feedback. The model does not; short of retraining it, I am not teaching it, I am writing an ever-better briefing pack. The distinction is worth holding onto, because it is also the distinction between developing your people and merely instructing them. The best prompt writers I know sound like the best delegators I know. If the daily practice of briefing agents teaches a generation of managers to be clearer about intent, more generous with context and faster with feedback, the humans around them will inherit the benefit.

What the tuning teaches

The struggle for fit carries its own lesson. To get a model writing in my voice I had to articulate standards I had carried for twenty years but was not conscious of: which words I never use, how I open an argument, what I find irritating. The model forced the implicit to become explicit.

Many organisations are learning this at scale. The brand voice, the quality bar, the way we treat customers here: teams tuning models are discovering these were never actually defined, only absorbed slowly by humans through years of correction and imitation. The absorption route still works, and it is expensive and slow, which is exactly why cultural fit has always been so hard to hire for. A company that does the work of articulating its standards well enough for a model to follow has, almost by accident, made them teachable to people too. The tuning effort is not overhead on the AI project. It is the organisation writing down what it is.

Where to start

If any of this is worth acting on, the cheapest place to begin is a document you already have. Whatever you or your team wrote to get a model producing work you would put your name to, the style guide, the worked examples, the list of things it must never do, give it to your next new starter. It was written to make the implicit explicit, which is what an induction is supposed to do, and it will be better than whatever is sitting in the HR folder because it was written by someone who cared whether the output was right.

The harder one arrives disguised as a hiring decision. Someone leaves, or a role changes, and the choice is between finding a person with the right list of skills and growing the person who already understands how the place works. That decision usually gets made on cost and speed, and it is worth asking at least once what you are giving up when you choose the list. If the answer is nothing much, you have learned something useful about the role. If the answer is a great deal, you have learned something less comfortable about how you have been measuring people.

The habits leaders are forming with their agents right now are forming largely unexamined, and when nobody examines a habit it tends to drift the wrong way, towards treating people as replaceable rather than towards managing them better. The check is simple enough to run on yourself. Look back at how you treated your models this week, the patience you extended, the context you shared, the speed with which you discarded one that disappointed you, and then ask whether anyone on your team is receiving the same treatment. That check, run honestly across a whole leadership team, is where much of our AI-era change work at Adaptive Edge now begins.