AI fluency is quickly becoming the most fashionable leadership skill in technology.
That should worry enterprise IT buyers.
Not because AI fluency is irrelevant. It clearly matters. Future IT leaders need to understand how AI tools work, where they create leverage, how they change delivery models and how they affect security, governance and productivity.
The problem is different.
AI fluency is being mistaken for IT judgement.
That distinction matters. A leader who can use AI confidently is not necessarily a leader who can challenge AI output, understand technical trade-offs, protect domain logic, manage people through change or identify when automation is being applied to the wrong problem.
Recent UK IT roundtable discussions showed this tension clearly. Senior technology leaders were not dismissing AI. They were actively exploring Copilot, Claude, agents, automation, governance models and AI-supported decision-making. Yet the deeper concern was not whether people could use AI tools. It was whether future IT leaders would still develop the judgement needed to use them well.
That creates an important shift for vendors.
Enterprise buyers are not only evaluating AI capability. They are evaluating whether technology partners can help their teams become more capable, not more dependent.
AI literacy is not the same as leadership readiness
AI literacy is often presented as a neat solution to the skills gap.
Train people on the tools.
Teach them prompt techniques.
Show them use cases.
Build adoption.
Measure productivity.
Scale from there.
That is a useful starting point, but it is not enough.
The roundtable discussion on future-ready IT leaders surfaced a more uncomfortable issue. Participants repeatedly came back to judgement, critical thinking, people leadership, technical depth, emotional intelligence and accountability. The strongest future leaders were not described as those who use AI most aggressively. They were described as those who know how to balance AI with human responsibility.
That is the part many enterprise AI narratives underplay.
A leader can be AI fluent and still be technically shallow.
A developer can use AI quickly and still miss domain logic.
A manager can encourage adoption and still fail to address employee anxiety.
A business team can produce outputs faster and still make weaker decisions.
A board can demand AI progress and still misunderstand the operational risk.
The next IT leadership gap is not simply tool adoption.
It is judgement under automation pressure.
The risk is not that AI gives bad answers
The more subtle risk is that AI gives plausible answers.
That is harder to manage.
Bad output is easy to challenge when it looks obviously wrong. Plausible output is more dangerous because it appears confident, complete and professionally packaged. For people without enough domain expertise, the output can feel trustworthy before it has been properly tested.
The roundtables highlighted concerns around non-deterministic AI outputs, hallucinations, over-reliance on AI tools and the need for critical thinking skills to evaluate when AI is appropriate. Participants also discussed the importance of training people to understand not only how AI tools work, but why they may provide different answers at different times.
This is a direct vendor signal.
Enterprise buyers will increasingly favour solutions that do more than generate, automate or accelerate. They will ask how the solution supports validation, review, supervision, explainability, auditability and human accountability.
For vendors, “our AI produces better answers” is not enough.
The stronger claim is: “we help your people know when the answer is safe to use.”
Traditional technical skills are becoming more valuable, not less
A contrarian theme emerged in the future-ready IT leadership discussion.
Traditional technical skills are not becoming obsolete. They are becoming the control layer around AI.
One participant stressed that foundational skills such as networking, mathematics and technical knowledge cannot simply be replaced by AI. Another noted that human developers remain essential for verification, domain expertise and guiding business stakeholders through what AI can and cannot do.
That challenges a common enterprise assumption.
If AI can write code, produce documentation, generate analysis and assist with technical decisions, organisations may be tempted to reduce the emphasis on deep technical training. That would be a mistake.
AI can accelerate work, but it cannot fully replace the experience required to understand whether the work is correct, safe, scalable or contextually appropriate.
The leader of the future does not need to write every line of code manually.
But they do need enough technical depth to challenge what AI produces.
They need to know when generated code creates maintainability issues. They need to understand whether a system design fits the architecture. They need to spot when a model recommendation conflicts with business logic. They need to recognise when AI has optimised for the wrong constraint.
This is where AI fluency can become misleading.
Tool fluency without technical depth creates confidence without enough resistance.
Vendor-relevant signals from the roundtables
| Roundtable signal | What it reveals | Vendor implication |
|---|---|---|
| 78 AI licences for 450 users were discussed in one Copilot rollout context. | Demand for AI access is outpacing controlled enablement. | Vendors should help buyers manage usage, training, access and prioritisation, not just deployment. |
| A 20-user Claude Enterprise trial was discussed alongside concerns about usage control. | Buyers are testing multiple AI environments, but still assessing boundaries. | Vendors need to show how their solutions fit inside governed experimentation. |
| One user consuming £100 in a week was used as an example of AI cost exposure. | AI fluency does not automatically mean commercially disciplined usage. | Usage visibility, cost controls and behaviour guardrails are becoming buyer concerns. |
| Participants raised concerns about AI producing credible but incorrect outputs. | The risk is not only accuracy. It is misplaced confidence. | Vendors should build validation, explainability and escalation into the sales narrative. |
| Future IT leadership was repeatedly linked to judgement, emotional intelligence and technical depth. | Buyers are not just buying tools. They are protecting capability. | Vendors that support enablement, supervision and human accountability will stand out. |
| Citizen-developed dashboards and AI-generated outputs were flagged as validation risks. | Business-led creation can move faster than governance and quality assurance. | Vendors should help buyers avoid unmanaged output becoming operational truth. |
| Participants discussed human-in-the-loop approaches and accountability for AI-generated outputs. | AI governance is becoming a leadership behaviour, not only a policy function. | Vendors should show how responsibility stays clear after automation is introduced. |
AI can flatten the wrong learning curve
AI is often praised for shortening the learning curve.
That is partly true.
A junior developer can move faster. An analyst can structure an answer more quickly. A business user can generate a draft without waiting for a specialist. A manager can summarise information without spending hours in documents.
The productivity gain is obvious.
The leadership risk is less obvious.
If AI removes too much friction too early, people may skip the uncomfortable learning moments that build judgement. They may produce outputs without understanding the assumptions behind them. They may rely on generated explanations instead of developing mental models. They may become efficient at asking questions while remaining weak at evaluating answers.
This is not an argument against AI.
It is an argument against confusing output speed with capability development.
Enterprise IT leaders are increasingly aware that AI changes how people learn. The question is not only whether teams can do more work with fewer steps. It is whether they still build the expertise needed to supervise, challenge and improve that work.
For vendors, this creates a different enablement opportunity.
Training should not only show users how to get better outputs. It should teach users how to interrogate outputs.
What evidence supports this?
What assumptions are hidden?
What alternative explanation exists?
What would make this unsafe?
Who needs to review it?
What decision should this not be used for?
The vendors who help buyers build that discipline will be more useful than those who simply promote faster adoption.
Human oversight cannot be a checkbox
Many AI vendors now use the phrase “human in the loop.”
That phrase is becoming dangerously vague.
A human in the loop can mean many things. It can mean a genuine subject matter expert reviewing a high-impact output. It can mean a busy employee clicking approve because the system requires it. It can mean a manager being theoretically accountable for a workflow they do not fully understand.
Enterprise buyers are starting to see the difference.
In the roundtable discussions, human oversight was connected to responsibility, training, validation and accountability. Participants discussed AI as a tool that requires checking, supervision and domain expertise, rather than an autonomous authority.
That should change vendor positioning.
It is not enough to say that humans remain involved.
Vendors need to explain how.
Who reviews the output?
What skills do they need?
What evidence do they see?
What gets escalated?
What gets logged?
What happens when the human disagrees?
How does the organisation know review is meaningful rather than ceremonial?
Human oversight is only valuable when the human has enough expertise, context and authority to challenge the machine.
Future IT leaders need scepticism as a skill
Scepticism is often treated as resistance.
In AI-enabled environments, it becomes a leadership skill.
That does not mean rejecting every AI output or slowing every initiative. It means developing the discipline to ask better questions before trusting the result.
A future IT leader needs to be comfortable asking:
Is this answer technically plausible?
Is it contextually relevant?
Is the data current?
Is the recommendation aligned with policy?
Is the output explainable?
Could the model be biased towards a vendor, platform or default pattern?
What would a domain expert challenge?
What decision should not be based on this?
The roundtable discussions specifically referenced the need to validate outputs, recognise bias and involve domain experts in the oversight of AI-driven decisions.
That matters for enterprise vendors because buyers are not only looking for intelligent systems. They are looking for systems that help their people remain intelligent users.
An AI product that makes users passive is a risk.
An AI product that helps users think more clearly is an asset.
People leadership becomes harder in an AI transition
AI adoption is often framed as a technology change.
It is also a people leadership test.
Some employees are excited by AI because it removes repetitive work. Others worry about job displacement, surveillance, changing expectations or being judged by tool usage. Some senior leaders overestimate what AI can do. Some junior staff overtrust it. Some technical teams resent business pressure to automate before foundations are ready.
The future IT leader has to manage all of that.
The roundtables showed this clearly. Participants discussed generational differences in technology adoption, employee concerns, the importance of in-person collaboration, the need for emotional intelligence and the role of leaders in coaching teams through AI-enhanced workflows.
This is another vendor signal.
If a vendor sells AI as a simple productivity upgrade, they may underestimate the internal politics of adoption.
Enterprise buyers need help with the human operating model. They need adoption plans that recognise fear, capability gaps, governance requirements and leadership accountability.
The strongest vendors will not talk only to the technology sponsor.
They will help the sponsor bring people with them.
The vendor mistake is selling AI as replacement
The weakest vendor message is still common.
AI will reduce headcount.
AI will replace manual work.
AI will remove dependency on scarce skills.
AI will automate complex judgement.
That may sound commercially attractive in some rooms, but it can create resistance elsewhere.
IT leaders know AI can improve efficiency. They also know that overpromising replacement can damage trust, trigger employee anxiety and weaken accountability. In technical environments, replacement language can also sound naive because it ignores verification, domain expertise, architecture, governance and operational context.
A stronger message is more credible.
AI should remove low-value friction while preserving high-value judgement.
That is the balance enterprise buyers are trying to find.
Vendors should show how their tools help teams move routine work faster while keeping responsibility clear. They should show how automation supports experts rather than bypassing them. They should demonstrate where human review remains necessary and where the system reduces cognitive load without reducing oversight.
The buyer does not need another promise that AI will do everything.
They need confidence that AI will not quietly erode the capabilities they still depend on.
What vendors should prove before the conversation goes too far
Enterprise buyers will increasingly test vendors on enablement, not just functionality.
They will want to know whether the tool helps develop stronger working practices or simply produces faster outputs.
That means vendors should prepare to prove:
How users are trained to challenge outputs.
How domain experts remain involved.
How AI-generated work is validated.
How junior users avoid over-reliance.
How governance is embedded into workflows.
How bias, uncertainty and hallucination risk are surfaced.
How managers remain accountable for AI-assisted decisions.
How technical teams can inspect, override or improve outputs.
The more AI becomes part of everyday work, the more buyers will care about the behaviours it creates.
A tool that accelerates weak judgement creates risk.
A tool that strengthens judgement while improving speed creates value.
The leadership benchmark is changing
The future-ready IT leader will not be the person who knows the most prompts.
It will be the person who can lead in an environment where machines sound confident, business pressure is high and technical judgement is harder to observe.
That leader will need AI fluency, but also scepticism.
They will need technical depth, but also people skills.
They will need speed, but also discipline.
They will need innovation instinct, but also governance maturity.
They will need to understand automation, but also know when not to automate.
That is the real opportunity for vendors selling into enterprise IT.
Do not only help buyers adopt AI.
Help them protect the human capability required to govern it.