Most weeks, Ryan and I sit on a call with a leadership team that has been thinking about AI for the better part of a year and is still no closer to a settled answer. Different industry each time, different size, different vintage of business. The script barely changes.
Someone senior says, in some version, "We can see the opportunity, we just don't know what the safe path looks like." Someone closer to delivery adds, "Our team is excited, but a few of them are using these tools in ways that worry me." A founder, sometimes more quietly, says, "I think I will have to make some hard calls about roles in the next eighteen months. I don't know how to prepare for that conversation, or whether I should."
I have stopped being surprised by the pattern. Underneath the differences, every business we sit with is wrestling with roughly the same four things at once. None of them are the things LinkedIn fixates on.
The first wrestle: the plumbing, not the tools
The popular story is that AI adoption is a tooling problem. Pick the right copilot, run a few prompt training sessions, sit back and watch productivity climb. My hunch is that this story is wrong almost everywhere it gets told.
The actual bottleneck is the layer beneath the tools. It is the contracts with clients that carry penalty clauses for data leakage. It is the InfoSec review that a Fortune 500 customer will run before they let you anywhere near their information. It is the ten-year-old folder structure that looks tidy until you ask an agent to find something in it. It is the file storage system that has slowly become the most expensive line on the SaaS bill, holding data nobody can usefully query.
I sat with a panel beater last year who still ran three different tools to do work that one good tool could now handle. I sat with an aviation servicing business this year that paused its AI roadmap entirely to run a ninety-day RFP for a new ERP, because the underlying tech stack simply could not carry what they wanted to layer on top. AI is the easy bit. The plumbing is where the real work is.
If you cannot easily find your data, classify your data, or trust the rails it runs on, you do not have an AI problem. You have a plumbing problem dressed up as one.
The second wrestle: outsourced thinking
A delivery lead said something to me recently that I keep returning to. "The risk I'm seeing is that people are using AI to help them think, and they are outsourcing the very learning that makes them valuable to clients." She had noticed it in her own team. People were generating frameworks, generating ideas, generating the building blocks of advice they would later be expected to defend in front of senior client stakeholders. Then they would arrive in those rooms thin, unable to answer the next question.
This is not a tooling problem either. It is a craft problem. It is the difference between a junior who has wrestled with a regulation for two days and a junior who has asked a chat box to summarise it in two minutes. They produce surfaces that look similar; the depth underneath is radically different. A client cannot always tell on first contact. They can always tell by the third meeting.
The teams I trust most are the ones whose leaders have started teaching, explicitly, where AI ends and the human craft begins. Use it to draft, not to think. Use it to research wide, not to conclude. Use it to compress the boring parts of the work, not to skip the parts that build judgement. Leaders who do not name this line will watch their bench quietly thin, then wonder in eighteen months why their best people stopped sounding like their best people.
This is not a problem of lazy employees. Laziness, where it shows up, is almost always a downstream symptom of leadership that has not given the team language, tools, and licence. The people closest to the work are usually the ones most aware of the risk. They just need to hear, from above them, that the craft is still the point.
The third wrestle: the pace of the thing
You can spend three months building out a stack and find that ten better tools have shipped in the meantime. This is disorienting for leaders who came up in a slower software cycle, where you chose an ERP and ran it for a decade. Several of the people I meet have stopped trying to keep up and have started, instead, hoping the dust will settle. It will not, at least not on any timeline that helps them.
The way through is not omniscience. It is a small, disciplined operating posture. Pick a horizon you can actually act on, three to six months. Choose tools that integrate cleanly into the layer below, so the cost of swapping them later is low. Train your people on the shape of the technology rather than the brand on the login screen, because the brand on the login screen will change.
This is closer to running a business in wartime than in peacetime. You are making decisions with incomplete information, on a moving front, with the knowledge that some of those decisions will not age well. That is fine. The cost of moving slowly is now higher than the cost of getting some things wrong.
The fourth wrestle: the role question
Eventually, every leader I speak to gets to the question they did not want to ask. Will I need fewer people next year? Will the people I have now be the right people? Should I be hiring at all, given that I might have to let some of them go?
I want to be careful about easy answers here. My suspicion is that most knowledge work businesses will not shrink in headcount; they will shift in shape. The administrative burden inside professional services, the gap analyses that take forty hours, the reconciliations that take days, the formatting and reformatting and rechecking, all of that compresses dramatically. What sits on top, the relationship, the judgement, the storytelling under pressure, becomes more valuable, not less. The team that produced one report a quarter can produce three. The team that managed fifteen clients can manage forty-five. The service offering does not shrink. It deepens.
The companies that handle this well are the ones whose leaders are already saying out loud, to their teams, what is going to change and what is not. Leaders who duck the conversation produce anxious teams who use AI in secret, badly, and the worst of both worlds arrives a year later: less learning, more leakage, no productivity gain to show for it. Leaders who name the change, even imperfectly, give their people room to adapt with them.
This is not a binary between protecting jobs and embracing technology. It is a third path, harder than either, where the leader takes responsibility for both. Should leadership protect the team from disruption? No. Should leadership protect the team's dignity inside the disruption? Absolutely.
How we come alongside
When Ryan and I work with a business on this, we tend to start in the same place. We interview every person on the team. We map the workflows that actually exist, not the ones the org chart implies. We pull out the painful bits, the time sinks, the duplicated tools, the places where confidentiality is being held together with the equivalent of duct tape and a polite warning. We give that back to the leadership team as a single document, with the easy wins and the hard ones flagged honestly.
What we do not do is assume the team is the problem. The teams we meet are almost always sharper than the systems they are forced to work inside. The people closest to the work usually know exactly where AI could help. They rarely have the language, tools, or licence to say so. Our job is to give them all three.
After that, the sequence is straightforward in shape, even when it is hard in execution. Set up a private environment so people can use these tools without breaking a contract. Train them on what is actually relevant to their work, not on a generic prompt engineering deck from 2023. Pick the two or three workflows where AI will compound, and build there. Leave the report behind so the leadership team can keep going without us.
We are not trying to be the centre of your AI story. We are trying to make sure your team becomes the centre of it.
A closing word for leaders
If you are running a business and you sense the slow tide of this technology rising under your feet, you are not alone in the feeling. Almost every leader we speak to is somewhere on the same map. Most of them are further along than they think, and further behind than they fear.
The work is patient and unglamorous. Map your plumbing. Name your craft. Pick your horizon. Bring your people with you, slowly enough that they trust you, quickly enough that the tide does not catch you. That is the season we are in.
We would rather walk it with you than watch it from the outside.