T Shape vs Robots: What David Epstein Gets Right About Staying Valuable in the AI Age

Why this issue matters

• AI automates narrow, repeatable tasks. Your moat is problem framing, cross-domain pattern matching, and decision leadership

• Range is trainable. You don’t need a new degree, you need better inputs and deliberate synthesis

• Generalist ≠ shallow. It’s depth in one arena plus the ability to bridge to others on demand (T Shape

You’ve seen it this week. A prompt spits out a half-decent draft. Someone ships it. Leaders clap because time was saved. Quiet panic follows: if output is this easy, where’s my edge?

It’s here: breadth.

Today we’re pulling from David Epstein’s book, Range: How Generalists Triumph in a Specialized World.

Epstein argues that in complex, fast-changing environments, the people who win aren’t the narrowest experts, they’re the ones who connect ideas across domains and make non-obvious analogies.

In other words, when AI compresses execution, range raises your value.

This issue turns that thesis into a practical playbook for analysts and managers under AI pressure.

Think: T-Shape skills.

Why the Squiggly CV Will Save Your Career
(And the Straight Line Won't)

AI doesn't care about your dashboard skills. It's likely to be better at them than you are soon.

ChatGPT can write your SQL, and is getting better all the time. Claude can critique your data model with an eye that seasoned practitioners posess. Cursor can scaffold your entire analytics pipeline while you're still opening Power BI.

The vertical bar of your T-shaped skills - that deep technical expertise you spent five years building? Being automated at a pace that should genuinely scare you.

It scares me a little.

But here's what still breaks every LLM on the market: knowing which problem to solve in the first place.​

That requires context. Business history. Political landmines.

The unwritten rules that shaped why your data looks the way it does. And most importantly, it requires pattern recognition across domains that AI simply doesn't have.​

The horizontal bar of the T - your breadth - is one moat that is still left.

What Breadth Actually Looks Like

Not "I dabble in Python and Tableau" That's just tool hopping.

Real breadth is accumulated scar tissue from solving structurally similar problems in different contexts. It's recognizing that marketing attribution and inventory forecasting are the similar problem: allocating credit under uncertainty with delayed feedback.​

When someone asks for a "new AI powered customer scoring model" analysts with breadth don't reach for another vendor tool. They pull from capital budgeting frameworks they saw in finance. They borrow decision thresholds from medical device projects.

They know what worked, what didn't, and why.

Across contexts that AI has never connected.​

That's not mystical "range". That's professional pattern matching. And it's one skill AI can't commoditize yet.​

Five Breadth Habits That Matter

1. Learn one adjacent domain per quarter

Not by taking a course. By shadowing someone for a week.

If you're in marketing analytics, sit with the finance team during month-end close. Ask stupid questions. Watch what decisions actually change when the numbers move. You'll discover that half of their forecasting problems are identical to your attribution mess (just with different labels and slightly scarier stakeholders).​

This isn't about becoming a finance expert. It's about expanding your mental library of problem structures so you recognize them faster than someone who can prompt an AI.​

Charlie Munger uses this mentality - having onhand a bunch of mental models to help navigate problems.

2. Build a 3-question intake ritual

Before you touch any tool, before you even open a ticket, force yourself to write down:

• What decision changes if this number moves?

• Who loses if we're wrong?

• Have we solved this category of problem somewhere else?

Share these answers with the requester. Half the time, they'll realize they asked for the wrong thing. They might just realise they have framed a problem AI would have happily solved incorrectly.​

This discipline is what separates "built a dashboard" from "prevented a bad decision"

Only one of those survives automation.​

3. Practice cross-domain translation

Take any analytics problem you're working on right now. Rewrite it as three different problems:

• As a finance problem

• As an operations problem

• As a product problem

Example: Customer churn prediction becomes credit default risk becomes defect rate forecasting becomes feature adoption likelihood. Same math, different industries, completely different political constraints.​

When you can translate fluently between contexts, you'll spot that half the "revolutionary new AI tools" are just rebranded logistic regression from another field. (Just a silly example)

And you'll know which ones actually matter.​

4. Document what AI can't see

After each project, write out the anomalies that AI probably would never have picked out: "The dataset didn't capture X, so we had to talk to Y, which revealed Z"

That's the breadth muscle. The legacy vendor relationship that explains the weird schema. The policy change from 2019 that nobody documented. The fact that the VP of Sales rounds all his forecasts to the nearest $50K because he hates decimals.​

AI hallucinates confidently through all of this.

You won't if you've built the habit of noticing context.​

5. Own one business outcome

Pick something scary: retention rate, margin improvement, lead conversion. Make it yours for two quarters. Don't just model it, own whether it moves.​

You'll be forced to learn pricing. Product roadmaps. Sales comp structures. The messy, cross-functional reality that makes analytics valuable instead of decorative.​

This is how you go from "I built a model" to "I know why the model didn't matter until we fixed the incentive structure."

Only the second sentence keeps you employed.​

The (Real?) Story About Range

Analytics leaders who are actually hiring right now don't always ask about your Alteryx certifications. They want to be able to ask "Can you walk into a pricing problem and recognize it's structurally identical to the capacity planning mess we solved last quarter?"​

That's not about being smart. It's about having enough context from different problem domains that you see the patterns before AI gets (re)prompted five times to maybe stumble on them.​

The T-shape isn't about making your CV look interesting. It's about building a library of business contexts that lets you solve adjacent problems faster than an LLM can be steered toward the right answer.​

What This Means Monday Morning

If you've got a career path that looks scattered, start narrating it differently.

Not: "I've bounced around between retail, healthcare, and supply chain."

Try: "I've solved flow constraints in three industries, which means I recognize demand signals faster than specialists and frame problems better than generalists."​

That's a story that survives automation. Not because you're more technical than AI, because you're not. But because you've built pattern recognition across contexts that AI doesn't have access to.​

The straight-line CV optimizes for 2019. The squiggle optimizes for 2027.

Pick accordingly. Squiggle~!!~

Your Range Ledger

Build this once, update monthly.

1. Range Map: three columns
   • Core depth: your strongest domain or toolset
   • Adjacent domains you can speak and translate
   • Transferable patterns you’ve used across contexts (constraints, incentives, queues, diffusion)

2. Analogy Bank: five mini-cases
   • “We cut onboarding time 28% by treating it like a manufacturing cell, not a help article.”
   • “We fixed out-of-stock alerts by borrowing from SRE incident response.”

3. Decision Dossier: recent calls you shaped
   • Decision, alternatives considered, principle used, outcome, lesson to reuse

4. Next 30-Day Inputs
   • One book or longform outside your lane
   • One shadow session with another function
   • One problem you’ll reframe with an analogy

Your Next Move (this week)

• Pick a live request. Rewrite it as a decision, not a deliverable. State the decision owner, the trade-off, and the minimum data needed.

• Add one structural analogy in your brief. “Treat this like capacity planning” or “this is portfolio risk.”

• Schedule a 20-minute cross-function chat. Ask for the constraint you never see from your side.

Hit reply and tell me which analogy you used and what changed.

Best,

Tom.

Disclaimer: Some of the articles and excerpts referenced in this issue may be copyrighted material. They are included here strictly for review, commentary and educational purposes. We believe this constitutes fair use (or “fair dealing” in some jurisdictions) under applicable copyright laws. If you wish to use any copyrighted material from this newsletter for purposes beyond your personal use, please obtain permission from the copyright owner.

The information in this newsletter is provided for general educational purposes only. It does not constitute professional, financial, or legal advice. You use this material entirely at your own risk. No guarantees, warranties, or representations are made about accuracy, completeness, or fitness for purpose. Always observe all laws, statutory obligations, and regulatory requirements in your jurisdiction. Neither the author nor EchelonIQ Pty Ltd accepts any liability for loss, damage, or consequences arising from reliance on this content.