AI becomes an answer machine or a thinking partner depending on one thing: the norms your leadership team sets. Left to default, it almost always becomes the former.

TL;DR
  • AI defaults to an answer machine. Leaders have to deliberately design it into a thinking partner.
  • The norms you set in the first 90 days will define how your team uses AI for years.
  • Workflow redesign predicts AI success more than the technology itself, according to 37 independent studies.
  • Do this with your people, not to them. Adoption beats resistance every time.

The Lazy Default

Here is what happens when a company buys an AI tool and does nothing else.

Someone opens a chat window. They type a question. They get a response. They copy it, maybe lightly edit it, and ship it. Nobody asks whether the output was right. Nobody pushes back. Nobody notices when the reasoning is shallow because the question was shallow.

Two months in, that person is producing more output. They look more productive. But ask them what they actually learned, and they shrug. Ask them to solve the next version of that problem without AI, and they struggle more than they did before.

This is what we mean by an answer machine that makes you stupider. The tool is not the problem. The integration is.

When AI hands you answers without requiring you to think, your ability to evaluate those answers quietly atrophies. You stop forming hypotheses before you query. You stop questioning outputs because you have come to expect them to be mostly right. You outsource the judgment alongside the task.

This is not a theoretical risk. It is a pattern we see repeatedly in organisations that adopt AI without a deliberate change in how work is structured.

What a Thinking Partner Actually Looks Like

An AI thinking partner operates differently. It does not just hand you an answer. It helps you arrive at a better one than you would have reached alone.

The distinction is not in the technology. It is in how you engage with it.

A thinking partner interaction looks like this: you bring a half-formed problem, you share your current thinking, and you ask AI to challenge your assumptions, surface what you might be missing, or model out scenarios you have not considered. You stay in the driver's seat. The AI rides shotgun.

That sounds simple. But it requires a different posture from both the user and the organisation. It means slowing down before you type. It means developing the habit of bringing your own thinking first, not asking AI to generate thinking for you. And it means leaders who model that behaviour publicly, because people watch what the top of the house actually does, not what it says.

The Leader Sets the Norm

This is why AI transformation has to be owned at the CEO level. Not delegated to IT. Not handed to a department head with a budget and good intentions.

IT can configure the tools. A department head can run a pilot. But the cultural norms around how AI is used? Those flow from the top. Every time a senior leader shares an AI output without interrogating it, they signal that interrogation is optional. Every time they publicly work through a problem with AI and show their reasoning, they signal something very different.

The IMF estimates that 60% of jobs will be materially changed by AI. The WEF projects 92 million roles displaced and 170 million new ones created. This is not an IT project. It is a shift in the nature of work itself. Treating it as anything less is how you end up on the wrong side of what economists are calling the K-curve: organisations that transform well will accelerate, and the ones that do not will fall behind. Right now, only 4 to 6 percent of organisations are on the right side of that curve.

The technology is not the constraint. The thinking is.

Find the Constraint First

One of the most common mistakes we see: companies buy AI and then go looking for a use case. They should do it the other way around.

Before any tool selection or rollout, the question is: where is work actually breaking down? Where are decisions slow, inconsistent, or too dependent on one person? Where do people spend time on tasks that produce nothing a customer would pay for?

That is where AI earns its keep. And more importantly, that is where redesigning a workflow around AI will produce something that actually sticks.

Thirty-seven independent studies have now looked at what predicts AI success in organisations. The number one factor, consistently, is not the model, not the vendor, not the budget. It is whether the organisation redesigned its workflows to take advantage of AI capabilities. Companies that bolt AI onto existing broken processes get broken processes with faster output. Companies that redesign how work flows first get compounding improvement.

So before you ask what AI can do, ask: what is the constraint? Where is the actual bottleneck? Fix that. Build AI into the fix. That is the sequence.

A concrete example

A mid-sized professional services firm came to us with a proposal process that was taking 11 days on average. They wanted to use AI to write proposals faster.

We pushed back. Writing was not the constraint. The constraint was that the information needed to write a good proposal lived in three different people's heads, and getting them aligned took a week. AI could not fix that. A better intake process could.

Once they rebuilt the intake process, they added AI to the drafting step. Average proposal time dropped to four days. The AI was a 20% contributor to that result. The workflow redesign was an 80% contributor. That ratio is typical.

Do It With Your People, Not To Them

The difference between adoption and resistance usually comes down to one thing: whether people feel like AI was done to them or with them.

"To them" looks like a rollout. Tools appear. Training decks circulate. People are told this will improve their work. Efficiency targets go up. Nobody asked whether anyone was worried about their job, or had ideas about what actually needed to improve, or knew something about the work that the strategy deck missed.

"With them" looks like a conversation before a decision. It looks like involving the people closest to the work in identifying the constraint and designing the new workflow. It looks like being honest about what is changing, what you do not know yet, and what is not changing.

This matters for performance. Organisations that involve employees in AI redesign consistently see higher tool adoption, fewer workarounds, and better outcomes. But it also matters for trust. And you cannot build an AI thinking partner culture on a foundation of distrust.

Develop Change Leaders inside the organisation

Top-down change management has a ceiling. At some point, the message has to travel peer to peer.

The most effective approach is to identify Change Leaders inside the organisation. Not the most senior people. The most credible ones. The people others actually go to when they have a question or a problem. Give those people early access, deeper training, and a real role in shaping how AI gets used in their part of the business.

This serves two purposes. First, it builds a distributed capability to keep improving how AI is used, rather than depending on periodic pushes from the top. Second, it signals to the broader organisation that this is not something being imposed. It is something being built together.

The 10-80-10 Method

When people ask how to think about where AI fits in a given workflow, we use a simple frame: 10-80-10.

The first 10 is setup. Framing the problem, defining the constraints, bringing context. This is almost always human work. AI is only as good as what you give it.

The 80 is execution. This is where AI can carry real weight. Drafting, analysing, comparing options, synthesising information, identifying patterns. A well-designed workflow hands this part off to AI and lets it run.

The last 10 is judgment. Reviewing the output, checking against what you know, making the call. This is always human work. Not because AI cannot generate a recommendation, but because the accountability stays with you.

The failure mode most organisations fall into is letting the last 10 slip. They accept the output of the 80 without the final review step. That is how you get confident-sounding nonsense in a client presentation. That is how the answer machine makes you stupider.

When you build AI into a workflow deliberately, you make the last 10 explicit. You build in a checkpoint. You treat AI outputs as drafts, not decisions.

Anchor the Change So It Sticks

Most AI improvements do not fail in the first 30 days. They fail in month four or five, when the initial energy fades and people quietly revert to what they knew.

The reason is almost always the same: the improvement was not anchored in daily management. There was no consistent measure that surfaced whether the new way of working was actually happening. There was no rhythm of review. There was no moment each week where a leader asked: how are we doing, and what do we need to adjust?

Anchoring looks like building the new workflow into existing operational routines. A weekly team check-in that includes a question about AI use. A dashboard that shows a metric connected to the workflow that changed. A regular conversation about what is working and what needs to change, held by someone who is curious rather than someone who is policing.

Control and daily management are not about surveillance. They are about signal. They tell you whether the change is taking hold or quietly unwinding. And they give you the data to improve rather than just repeat.

What leaders actually need to do

If you are running a business and you want AI to be a thinking partner rather than an answer machine, the practical actions are straightforward.

  • Name this as a leadership priority, not an IT initiative. Say it publicly, more than once.
  • Start with the constraint, not the tool. Find where work is actually breaking down.
  • Involve the people closest to the work in designing the new workflow.
  • Identify two or three Change Leaders in the organisation and give them real responsibility.
  • Build the new way of working into daily management routines, so it has somewhere to live.
  • Model the thinking-partner behaviour yourself. Show your team what it looks like to push back on AI output and use it to arrive at a better answer.

None of this is complicated. Most of it is just leadership, applied to a new context.

Frequently asked questions

An answer machine generates output in response to a query, and the user accepts it with minimal scrutiny. An AI thinking partner is used to challenge assumptions, explore scenarios, and arrive at a better answer than you would reach alone. The difference is not the technology. It is how you engage with it and what norms your organisation sets around that engagement.
Because the norms around how AI is used are set by behaviour at the top, not by policy documents. If senior leaders use AI as a shortcut and accept outputs uncritically, that signal travels fast. If they model rigorous, thinking-partner engagement, that travels too. IT can configure the tools. Only leadership can shape the culture of how those tools are actually used.
The most reliable method is anchoring changes in daily management routines. Build the new workflow into existing operational rhythms, track a metric that tells you whether the change is holding, and create a regular review conversation to surface what is working and what needs adjustment. Changes that are not embedded in day-to-day management routines almost always erode within three to six months.
Recap

AI defaults to an answer machine. It hands people outputs, and without deliberate design, people stop questioning those outputs. Capability quietly erodes. The organisations that avoid this trap do so by treating AI adoption as a leadership initiative, not a technology deployment. They find the constraint first, redesign the workflow around AI, involve their people in the redesign, develop Change Leaders inside the organisation, and anchor the new way of working in daily management routines.

The next action: before your next AI conversation, write down what you already think the answer is. Use AI to challenge that thinking, not replace it. Do that once. Then make it the norm.