Leading Through the Messy Middle of AI

Most leaders are stuck choosing between two extremes when it comes to AI at work. Quick wins that save a few minutes here and there but don't move the needle. Or big transformation programs that promise efficiency and often cost jobs.

Andrew Pope, a Dear Thrive advisor specialising in productivity in distributed teams, calls the space between these two extremes the messy middle. It's the everyday workflows, processes and habits that make up how a team works, and Andrew argues it's where leaders have the most influence over how AI gets adopted well.

He shared this framework at a recent SPARK session, Dear Thrive's speaker series for leaders navigating change. Here's what he covered, and what it means for anyone leading a team through AI change management right now.

The pressure to "use AI" is real, and it's not helping

Andrew opened by naming something most leaders feel but don't say out loud: there's a real expectation that leaders should be visibly using AI, and it's coming from more than just internal pressure. Technology vendors have a financial interest in adoption, pricing is climbing after a period of cheap access, and the result is a constant low hum of pressure to be across every new tool, not just AI.

That pressure creates a second, quieter problem.

Watch for "botshitting"

Andrew introduced a term that's gained traction: botshitting. It's different from misusing AI. Botshitting is what happens when people stop checking AI output altogether because they're too stretched to review it before sending it out.

It's becoming one of the most common failure points inside teams, not because people don't care about quality, but because reviewing AI output takes time nobody feels they have. For leaders thinking about AI strategy consulting or rolling out new tools, this is the risk that sits underneath the rollout: adoption without review isn't adoption, it's exposure.

Set guardrails, not policies

Andrew's first practical recommendation for the messy middle is clarity, and clarity doesn't need to come from a lengthy policy document. A simple traffic light system does more work than most 40-page guidelines:

• Low risk - use of AI that's fine to just get on with
• Medium risk - output that needs a second look before it goes anywhere
• High risk - work that needs approval or a defined review process before use

This is the kind of lightweight governance that change management consultants often recommend, but it works just as well built internally by a team that understands its own workflows.

Map the workflow, not the prompt

The second recommendation is a shift in where leaders start. Andrew's view is that starting with prompts means starting with the solution, not the problem.

Instead, map the actual workflow that's causing friction. What's the trigger? What's the source of the pain? What does a good outcome look like? Common candidates include triaging incoming work, stakeholder reporting, or onboarding new starters, the repeated, painful processes every team has but rarely stops to examine.

This is a subtle but important distinction for anyone leading a team through AI change management: the tool comes second. The workflow comes first.

Make it social

AI has largely been treated as an individual habit, something people do at their own desk with Claude or Copilot. Andrew's argument is that the real value comes from talking about it as a team, because the conversation isn't really about AI. It's about the work.

This includes properly listening to the skeptics in the room. In the change management workshops Andrew has run recently, skeptics consistently turned out to be some of the most valuable voices in the process, precisely because they don't trust AI outputs by default. That scepticism makes the critical thinking sharper and the final output better, provided leaders bring skeptics into the process rather than working around them.

Set norms, don't rely on assumptions

A recurring mistake Andrew has seen is leaders assuming people already know what's expected of them, or will simply do the right thing without being told. The fix is to build shared norms deliberately:

• What should we do together, and what stays individual?
• When do we check in on AI-assisted work?
• Who's responsible for reviewing outputs?
• What does our traffic light system actually mean in practice?

Dear Thrive's own research into the digital employee experience found that a lack of shared working norms was one of the biggest drivers of a poor experience at work, well before AI entered the picture. The same principle applies here: consistency and clarity beat good intentions.

Where to start

Andrew's closing advice was to keep the first step small:

1. Find one workflow that causes friction
2. Rate it using a simple traffic light
3. Decide whether it needs checking, or whether it's fine to just go
4. Run a session with the team to talk it through together

The goal isn't for leaders to become AI experts. It's to bring clarity to how AI fits into a specific team, department or division, and to lead people through that process together rather than leaving them to work it out alone.

The takeaway for leaders

The messy middle is uncomfortable because it doesn't come with a template. But it's also where leadership coaching and hands-on AI consulting tend to have the most practical impact, in the day-to-day workflows a team actually runs on, not in a policy binder or a one-off training session.

If you're leading a team and trying to work out where AI actually belongs in how you work, 1:1 session with Andrew through Dear Thrive..

Frequently Asked Questions

What is the "messy middle" of AI adoption? The messy middle is the space between small, low-value AI quick wins and large transformation programs. It refers to the everyday workflows, processes and habits a team uses to do its work, and it's where leaders have the most direct influence over how well AI gets adopted.

What does "botshitting" mean? Botshitting is when someone stops reviewing AI-generated output before using or sending it, usually because they feel too busy to check it. It's a growing failure point in teams using AI tools, distinct from simply misusing AI.

How do you create AI guardrails without writing a long policy? A simple three-tier traffic light system works well: low risk (fine to use as is), medium risk (needs a second look), and high risk (needs approval before use). This gives teams clarity without requiring a lengthy governance document.

Should leaders start with AI prompts or with workflows? Workflows first. Starting with a prompt means starting with a solution before the problem is defined. Mapping a painful, repeated workflow (like onboarding, reporting or triage) first makes it clear where AI can actually help.

Why does team skepticism about AI matter? Skeptics tend to be strong critical thinkers who don't trust AI output by default. Involving them in the process, rather than working around them, generally improves the quality of the final output.

Who is Andrew Pope? Andrew Pope is a Dear Thrive advisor specialising in organisational design and distributed teams collaboration. He spoke on this topic, "Leading Through the Messy Middle of AI," at a recent Dear Thrive SPARK event.

#leadingthroughchange #coaching #professionaldevelopment #leadership #AIadvisor