Developing professional judgement in the AI age.
A white paper by Alex Young, written as part of the Weekend MBA at Imperial College Business School. Fieldwork is under way and early themes are emerging - publishing November 2026.
Themes emerging
How are mid-market service firms redesigning junior development to preserve expertise and judgement formation in the AI age?
Generative AI is reshaping how service firms operate - drafting documents, summarising research, producing first-pass client deliverables. Most attention has gone to productivity. Far less has gone to what AI means for how expertise and judgement actually develop inside firms.
Junior work has historically done two things at once: produced client output, and developed the judgement and contextual understanding required for senior roles. When AI completes the task, what happens to the developmental work that used to come with it?
The risk is a talent pipeline problem - a generation of technically capable but judgement-thin mid-seniors who struggle with the advisory, client-facing work senior roles demand. For mid-market firms, without enterprise-scale learning infrastructure to fall back on, the risk is particularly acute.
UK mid-market service firms. Two sectors.
UK-headquartered firms selling expertise, advice or relationship-led services. Roughly 50–500 FTE or £10m–£100m revenue.
Management consulting and professional advisory firms; and marketing, creative and communications agencies.
12–15 semi-structured interviews across firm leaders, people & talent leaders, and senior client-facing practitioners. Inductive thematic analysis.
Early findings.
These are early signals from conversations with founders, leaders and practitioners across consulting, advisory, marketing and creative agencies. Names and firms are kept anonymous. Not final conclusions - but the shape of the argument is starting to appear.
The developmental loss is repetition
Junior work was never only output - it was a training set. Seeing volume and variety of examples is what lets someone tell apart tasks that look identical on the surface. AI is not just removing boring work; it is removing the reps that build discrimination.
The verification paradox
Every firm's answer to AI risk is the same: a human checks the output. But the check requires the knowledge that the checked work used to build. Firms are asking juniors to supervise a system that is removing the grounds for their supervision.
Developmental ring-fencing
Firms are starting to protect certain tasks not because AI cannot do them, but because doing them grows the person. Some experiences are being ring-fenced for learning - a different rationale from the usual risk and confidentiality boundaries.
From output to reasoning
Where firms have adapted, the inspection point is shifting. Managers are moving from assessing the final answer to evaluating the reasoning trail. The cheap, transferable practice emerging everywhere: show your working.
AI in the back office, humans in the front
Firms are comfortable automating the administration of development - objectives, one-to-one records, development plans - while explicitly refusing to automate the relationship. AI belongs in the back office of growth; humans stay in the front.
Friction must be manufactured
Struggle is being reframed as a condition of development rather than a cost of it. If AI removes naturally occurring difficulty, someone has to put it back deliberately - through real tasks, safe-to-fail environments, and protected exposure.
What's been said.
AI will always do the boring bit much better. But you still need a controller of that, to be able to navigate it.
It can't just be about taking an output and putting your name to it. It has to be about how detailed a critical analysis you do on everything.
Being in the room, thrown in the deep end, making plenty of mistakes, that's what accelerated me. It's still one of the most formative parts of my career.
Someone who really knows their subject can say, 'I didn't read that' or 'I didn't say that,' they can spot when AI has introduced something that isn't actually there. Without that grounding, would people know enough to catch it?
Month by month.
Six months from question to published paper. Here's the path - and where I am on it right now.
- Phase 01May 2026Setting the question
Sharpen the research question, map the existing literature, design the interview guide and clear ethics. The unglamorous bit that makes everything else possible.
- Phase 02June 2026Finding the rooms
Confirm the sample - founders, MDs, people leaders and senior practitioners across consulting and creative agencies - and start the first interviews.
- Phase 03Jul-Aug 2026Listening hard
Run the bulk of the interviews. Begin coding and thematic analysis. Watch the first patterns - and the surprises - emerge from the transcripts.
- Phase 04Sep 2026Themes emerging
Fieldwork under way. Early signals are landing around where AI is removing junior exposure, what firms are doing in response, and how judgement itself is being redefined.
- Phase 05Oct 2026Drafting
First full draft of the white paper. Build the visuals and the framework. Send it round for honest, unflattering feedback.
- Phase 06Nov 2026Publishing
Edit, tighten, proofread - and ship. Public white paper out into the world, with a launch event and follow-up writing to come.
A practical framework for leaders.
The final paper will be a publicly shareable white paper - written for founders, MDs and people leaders of mid-market service firms. The aim: a clear, usable framework for redesigning junior development in AI-enabled environments, alongside the principles, organisational conditions and leadership behaviours that appear to make it work.
Updates & learnings.
I'll be sharing what I'm reading, who I'm speaking to and the patterns emerging on LinkedIn. Come along for the ride.