Back to The Messy Middle
Part 3 of 3
The Human Side
People, change, and the AI transition. A practical guide to bringing your team along without breaking them.
Last updated: April 2026
People, change, and the AI transition. A practical guide to bringing your team along without breaking them.
Last updated: April 2026
AI adoption is not a technology problem. It is a people problem.
Every conference keynote, every vendor pitch, every breathless LinkedIn post frames AI as a question of tools and platforms. Which model? Which vendor? Which integration? And those questions matter. But they are not the reason most AI projects fail. The reason most AI projects fail is that the people who need to use the tools do not want to use them, do not trust them, or do not understand why they should.
The technology is ready. The question is whether your organisation is. And that question has almost nothing to do with software. It has to do with trust, communication, fear, ego, identity, and the deeply human experience of being told that the way you have done your job for the last ten years is about to change.
If you are a CEO, MD, or senior leader reading this, you probably feel the urgency. You can see what AI is capable of. You can see your competitors moving. You want your business to move faster. But between your ambition and the actual change sits your entire workforce, each person carrying their own mix of curiosity, scepticism, and fear.
This is the human side of the messy middle. And getting it right matters more than choosing the right AI vendor.
There are two groups of people in every organisation when it comes to AI. They look at the same technology and see completely different things.
CEOs, board members, investors, senior leadership
Typical response: "We need to move faster. Why are we not using AI across every department already?"
Team leads, managers, senior staff, specialists
Typical response: "This is just another initiative that will create more work before it creates less."
Neither group is wrong. The decision-maker is right that AI will reshape their industry and that moving slowly carries real risk. The operator is right that change is disruptive, that new tools create short-term pain, and that their years of expertise should not be casually dismissed.
The problem is that these two groups rarely talk to each other honestly. Decision-makers present AI as pure opportunity, glossing over the anxiety. Operators express resistance through passive compliance or quiet scepticism. The result is a gap between what leadership announces and what actually happens on the ground.
Aggregated from McKinsey, PwC, and Gallup surveys, 2025-2026
The gap is not between your business and AI. The gap is between the people who set the direction and the people who have to walk it. Close that gap first. The technology part is straightforward.
When someone resists AI adoption, the temptation is to label them as stuck in their ways or afraid of technology. That is lazy thinking. Most resistance comes from legitimate concerns that deserve honest answers, not dismissal. Here are the four fears you will encounter in every organisation, and the truth about each.
This is the big one, and it deserves a direct answer. Some roles will change significantly. Data entry clerks, basic report writers, first-line customer support agents handling routine queries, junior researchers doing literature reviews. These roles are already being transformed. Not in five years. Now.
But "transformed" is not the same as "eliminated." The data entry clerk becomes the person who validates and corrects AI outputs. The report writer becomes the person who interprets and presents insights. The customer support agent handles the complex cases that AI escalates, and those cases require more skill, not less.
The honest answer is: most roles will be augmented, not replaced. A 2026 McKinsey analysis of 800 occupations found that less than 5% of jobs can be fully automated with current AI. But roughly 60% of jobs have at least 30% of their tasks that can be automated. That means the job changes shape. It does not disappear.
Verdict: Partially justified. Be honest about which tasks are changing. Do not pretend nothing will change. But also do not overstate the threat. Most people's jobs will get better, not disappear.
The tools are changing faster than people can learn them. This is not perception. It is measurable. In 2024, the standard AI coding tool was GitHub Copilot, an autocomplete for developers. By mid-2025, it was Cursor and Windsurf, full-file AI editors. By early 2026, it was Claude Code and OpenAI Codex, autonomous agents that can build entire features. Three distinct generations of tooling in 18 months.
This is happening across every profession, not just software. Accountants who learned Xero two years ago now need to understand AI categorisation. Marketers who mastered Google Ads now need to understand AI-generated creative. HR managers who built interview frameworks now need to understand AI screening tools.
The pace is genuinely unprecedented. And for someone who has spent years mastering a particular toolset, the idea that they need to start learning again can feel demoralising.
Verdict: Justified. This fear is real. The answer is not to minimise it but to invest in continuous learning and make AI training a normal part of work, not a one-off event. People who learn the new tools alongside their domain expertise become more valuable, not less.
"I spent 10 years learning this system and now AI can do it in 10 seconds." This is a real feeling, expressed by real people, in every industry. The accountant who spent years learning tax legislation. The developer who spent a decade mastering a programming language. The procurement manager who built supplier relationships over 15 years.
When AI can draft a tax return, write a function, or analyse supplier pricing in seconds, it can feel like those years of learning were wasted. Like the expertise has been commoditised.
Here is what that feeling misses: AI can produce a first draft faster than any human. But it cannot tell you whether the draft is correct. It cannot apply judgement honed by a decade of edge cases. It cannot read a client's unspoken concern or navigate the politics of a procurement decision. The expertise is not in the output. It is in knowing whether the output is right.
Verdict: Emotionally valid, but mostly unfounded. Expertise becomes more valuable with AI, not less. The people who know the domain deeply are the only ones who can validate what AI produces. They move from doing the work to directing and quality-checking the work. That is a promotion, not a demotion.
This one is rarely spoken aloud. Nobody wants to admit they feel left behind. But it is perhaps the most common fear, especially among mid-career professionals who have been competent and confident in their roles for years and now feel like beginners again.
The AI landscape moves so fast that even people who are genuinely interested struggle to keep up. A new tool launches every week. The terminology keeps shifting. Last year it was "prompts." This year it is "agents" and "MCP" and "tool use." Next year it will be something else. For someone who just wants to do their job well, the constant barrage of new concepts can be paralysing.
They open ChatGPT, try a few prompts, get mediocre results, and conclude that either AI is overhyped or they are not smart enough to use it. Both conclusions are wrong. The tool just needs better context, and the person just needs better guidance.
Verdict: Very common, very understandable. The fix is structured onboarding, not a link to a YouTube playlist. Show people one tool, one workflow, one improvement. Build confidence through small wins. Nobody needs to understand the entire AI landscape. They need to understand how AI helps them do their specific job better.
Something remarkable has happened in the last 18 months. Tools like Bolt, Lovable, Replit, and ChatGPT have made it possible for anyone to build things that previously required professional skills. A marketing manager can create a landing page. A project manager can build a basic internal tool. A founder with no technical background can prototype an application.
This democratisation is real, and it is valuable. It lowers barriers, speeds up experimentation, and lets people test ideas without waiting for a development team. But it also creates a dangerous illusion: the idea that because anyone can use the tools, anyone can produce professional-grade results.
They cannot. And the gap between what an amateur produces and what a professional produces is not small. It is enormous.
A project manager using Claude Code is not the same as a senior developer using Claude Code. The tool is the same. The output is not. AI amplifies what you already know. It does not replace what you do not know.
Consider software development. A non-technical person can prompt Claude to build a working application. It will look right. It might even function correctly in a demo. But under the surface, the architecture may be fragile. Security vulnerabilities may exist that the person cannot identify. Performance bottlenecks may be invisible until the application has real users. Error handling may be minimal. The database schema may not scale.
A senior developer using the same tool produces fundamentally different output. They know which questions to ask. They know which patterns to apply. They know what "good" looks like, so they can steer the AI towards it. They know what can go wrong, so they test for it. The AI does the typing. The developer provides the thinking.
The same pattern holds across every profession. A junior marketer using AI to write copy produces passable content. A senior marketer using the same tools produces content informed by years of understanding audience psychology, brand voice, and conversion patterns. The AI writes faster. The expert writes better.
This matters for leadership because it reframes the entire AI conversation. The right question is not "can anyone use these tools?" The answer is yes. The right question is "should they, for production work?" And the answer is: it depends on the stakes.
For internal exploration and low-stakes experimentation, let everyone use AI freely. Encourage it. Celebrate it. For anything that touches customers, handles money, or has lasting consequences, pair AI tools with domain expertise. The combination of professional knowledge and AI capability is the most powerful force in business right now. Neither one alone comes close.
Most AI rollouts fail because they follow the wrong playbook. Here is what works, based on what we see in businesses that actually get their teams using AI.
Do not launch AI across the entire company. Pick one team that has a clear, repetitive pain point. Customer support is often ideal: high volume, repetitive queries, measurable outcomes. Or accounts payable: manual data entry, invoice matching, approval chasing. Choose the workflow where the pain is obvious and the improvement will be visible.
The goal is not to "do AI." The goal is to make one team's life measurably better. When that team talks about the improvement in the break room, you have done more for AI adoption than any all-hands presentation ever could.
The worst thing you can do is present slides about AI's potential. Nobody cares about potential. They care about their own workday. Instead, build the improvement first, then show it working. "This report that took you four hours now takes 20 minutes. Here, try it." That is more convincing than any McKinsey deck.
Demonstrations beat presentations. Every time. A 15-minute live demo where someone does their actual job faster with AI will convert more sceptics than a 90-minute strategy session. Make it real. Make it their workflow. Make it undeniable.
The fastest way to kill AI adoption is to have IT or leadership dictate which tools every department uses. The finance team knows their pain points better than anyone. The sales team knows where their process breaks. The operations team knows which manual steps are wasting hours.
Give teams a budget and a framework. Let them identify their own AI opportunities, propose solutions, and lead the implementation within their domain. Set guardrails around data security, vendor approval, and integration standards. But within those guardrails, let the operators drive. People adopt tools they chose. They resist tools that were chosen for them.
Every job has tasks that people dread. The weekly status report that nobody reads. The data entry that takes three hours every Monday. The formatting of proposals that should be standard but never is. The manual reconciliation of spreadsheets.
When you introduce AI, lead with those tasks. "We are automating the part of your job that you hate." Not "we are making your role more efficient." Efficiency is a boardroom word. "Removing the boring stuff" is a human one. People do not fear losing the parts of their job they dislike. They fear losing the parts they are proud of. Keep the pride. Automate the pain.
An external AI consultant can set up the tools and build the integrations. That is valuable. But they leave. The person who actually drives long-term adoption is the team member who uses the tools daily, finds new applications, and helps colleagues when they get stuck.
Identify these people early. They are usually mid-career, technically curious, respected by their peers, and frustrated by inefficiency. Give them time, resources, and recognition. Make "AI champion" a formal role, not a side project. These people are worth more to your AI transition than any vendor partnership.
Let us talk about the thing that leadership is thinking but not always saying: headcount. AI will affect the number of people some businesses need. Pretending otherwise is dishonest and ultimately more damaging than the truth.
Yes, AI will reduce headcount in some areas. Routine administrative work, basic data processing, simple customer queries, standard report generation. These functions will require fewer people. That is not speculation. It is already happening. Klarna reduced its customer service workforce by 700 agents through AI chatbots in 2024. Financial institutions are cutting back-office roles. Accounting firms are restructuring around AI-augmented workflows.
Yes, it will create new roles in others. AI trainers, prompt engineers, integration architects, data quality managers, AI ethics officers, human-AI workflow designers. These roles barely existed two years ago. Some of the people in the reduced roles will move into these new ones, but only if you invest in the transition.
The net effect is positive for businesses that manage the transition well. The net effect is destructive for businesses that handle it badly. The difference is not the technology. It is the leadership.
Estimated impact based on current AI capabilities, April 2026
Low displacement. AI assists with research and preparation but the human judgement is irreplaceable.
AI accelerates execution but creative direction still requires human taste and judgement.
Junior roles shrinking. Senior roles amplified. One experienced developer with AI does what a team of three did a year ago.
Volume content roles declining fast. Strategy and performance marketing roles remain strong.
Routine, rule-based work is the first to be automated. Reskilling into data quality and validation roles is the path forward.
AI chatbots now handle 60-80% of tier-1 queries. Remaining support roles focus on complex, high-empathy cases.
What separates the businesses that thrive from the ones that suffer through this transition? Three things.
First, they are honest. They tell their teams what is changing and why. They do not sugarcoat the impact on roles, but they also do not catastrophise. They lay out the plan: which roles are evolving, what new roles are being created, and how people can transition.
Second, they invest in people before they invest in tools. Training budgets, time for learning, mentorship programmes, clear career paths that incorporate AI skills. The cost of reskilling an existing employee is a fraction of the cost of hiring a new one, and you keep the institutional knowledge.
Third, they move fast but not recklessly. They start small, prove the value, expand deliberately, and check in with their teams at every stage. Not because they are being cautious. Because they know that sustainable change requires buy-in, and buy-in requires trust.
If you are reading this as a senior leader who wants to start the AI transition properly, here is a concrete timeline.
Foundation
Proof of concept
Scaling
p0stman helps businesses navigate the human side of AI adoption. We build the tools, train the champions, and set up the workflows that make the transition feel like progress, not upheaval. If you are a senior leader who wants to move fast without losing your team, let us talk.
Written by Paul Gosnell, founder of p0stman. 20 years building digital products across healthcare, finance, hospitality, and manufacturing. Now building with AI every day and helping businesses bring their teams through the transition. This briefing is part of The Messy Middle, a three-part series on navigating AI adoption practically.
Read more about Paul