A note on this piece: this is my personal opinion, formed after researching the question "Where is humanity going with AI in the mid to long-term future — human knowledge and expertise used to be locked into humans, but now it can be digitally replicated with AI models, so what are the long-term outlooks for how it will shape human futures?" The trend analysis below draws on published research (cited throughout); the conclusions and framing are mine.
Historically, expertise was organic — passed down through apprenticeships, locked in physical libraries, carried as tacit judgement in individual minds. Turning that expertise into digital, replicable models is a genuine civilisational shift, not just another tool cycle. And it's already reshaping what "being good at your job" means, across four dimensions worth understanding before we get to what to actually do about it.
Right now, generative AI is an extraordinary leveller. MIT Sloan research on contact centre agents found workers with less experience saw the biggest productivity gains from AI assistance — as much as 14% overall, concentrated almost entirely among new or lower-skilled staff. A separate study of software developers found juniors and recent hires increased output by 27–39% with AI coding tools, compared to 8–13% for senior developers. Less experienced people are, in effect, borrowing expertise that took previous generations decades to build.
The open question is what happens next. If AI absorbs the routine, entry-level reps — drafting the first version, writing the basic brief, running the first pass of analysis — junior professionals may never wrestle with the foundational problems that build real judgement. Brookings frames this directly: borrowed expertise might not survive the generation that built it. In twenty years, when today's senior experts retire, the concern is a shortage of people who can look at a confident AI output and know, immediately, that it's wrong.
That's not a reason to panic about AI. It's a reason to be deliberate about which foundational problems you still wrestle with yourself, even when a tool could hand you a shortcut.
Knowing how to write the code, draft the contract, or render the pattern is being commoditised fast. What isn't commoditising is knowing what to solve, why, and for whom.
[Traditional Expert: Execution-Heavy] ──> [Future Expert: Judgement-Heavy] (Knowing HOW to build/write/design) (Knowing WHAT to solve, WHY, and for WHOM)
MIT researchers Isabella Loaiza and Roberto Rigobon mapped this shift formally in a 2026 paper — The EPOCH of AI: Human-Machine Complementarities at Work. Using task-interdependency data across every US occupation, they found that new tasks created in 2024 carry significantly higher scores on five specifically human capabilities than the tasks they replaced, and that jobs built around those capabilities are growing faster, hiring more, and projected to keep doing both through 2034. The expert of the future looks less like a solo craftsperson and more like an editor and director — steering AI-driven work, validating what's ethical, and aligning machine output with what people actually need.
When expert-level knowledge is available to anyone with a prompt, the barrier to entry for complex fields drops close to zero. A single person can now run something that used to need a department — managing build, compliance, and localised marketing at once.
The risk sitting underneath that is homogenisation. AI models are trained on the existing corpus of human output, so they're excellent at reproducing established patterns and mediocre at genuinely new ones. If most professional output leans on the same underlying model weights, cognitive diversity thins out — progress stalls unless people deliberately push past what the models already know how to do well.
An AI can replicate an oncologist's diagnostic pattern-matching. It cannot carry the moral weight of the decision, or offer a patient genuine human presence. That distinction is going to get drawn more sharply, not less — expect "human-verified" and "human-held responsibility" to become both a premium offering and, in some fields, a legal requirement. Simulated alignment was never a moral promise. It's a capability, and capability and accountability aren't the same thing.
The four trends above point at the same conclusion from different angles: the capabilities that don't commoditise are the ones AI structurally can't replicate — not because it isn't good enough yet, but because they aren't the kind of thing a model does. Loaiza and Rigobon's paper gives these a name: EPOCH — Empathy, Presence, Opinion, Creativity, and Hope. Here's what each one actually looks like for a designer, how you build it deliberately, and how you prove you have it — because "I'm empathetic" on a CV convinces nobody.
What it looks like: reading what a stakeholder or user isn't saying — the unstated resistance behind "can we make it pop," the real reason a user abandons a flow that tests fine on paper.
How to build it: run one unscripted conversation a month — a real user, a real stakeholder — with no script and no leading questions. Scripted research trains you to confirm; unscripted conversation trains you to notice.
How to prove it: a case study section that shows the pivot a raw conversation caused, not just the polished outcome. Evidence of listening beats a claim of listening every time.
What it looks like: being genuinely in the room — live, collaborative, responsive — rather than only ever delivering a finished file.
How to build it: push for a real-time working session over an async handoff wherever you can. The habit of thinking out loud with someone, live, is a different muscle to presenting a deck.
How to prove it: testimonials and references that describe working with you, not just receiving what you made.
What it looks like: a defensible point of view on what's right — not just competent execution of what was requested.
How to build it: write your reasoning down before you show the design, every time. If you can't defend a decision in a sentence, you haven't finished making it.
How to prove it: a portfolio that shows the options you rejected and why, not only the one you shipped.
What it looks like: a genuinely novel direction, not a well-executed remix of what's already trending in the same three Figma community files everyone else is using.
How to build it: deliberately source references outside your usual toolkit — different industries, different eras, different disciplines entirely.
How to prove it: work that doesn't look like it came from the model everyone else is prompting. Distinctiveness is now a visible, checkable signal.
What it looks like: the ability to set direction when the brief is vague, the client is nervous about AI disrupting their business, or nobody's sure what "good" looks like yet.
How to build it: practise articulating where something goes in three years, not just what ships this sprint — out loud, in writing, even when nobody's asked.
How to prove it: how you frame a pitch or proposal, not just the deliverable inside it. Vision shows up in the framing before anyone's seen the work.
If you're chasing contract or freelance work in a market that's nervous about AI, EPOCH isn't an abstract framework — it's the actual list of things a prospective client can't get from a prompt, and the reason they'd hire a person instead of generating the deliverable themselves. Execution was always going to be commoditised eventually. What's rare, and what's hireable, is judgement, presence, a real point of view, distinctive creative direction, and the ability to hold a vision when nobody else in the room has one yet.
___________________________________
References:
1. The EPOCH of AI: Human-Machine Complementarities at Work — Isabella Loaiza & Roberto Rigobon, SSRN / MIT, 2026.
2. Borrowed expertise: Why AI's productivity boom may not survive the generation that built it — Brookings, 2026.
3. Workers with less experience gain the most from generative AI — MIT Sloan, contact centre agent study (Li, Raymond, Brynjolfsson).
4. The Effects of Generative AI on High-Skilled Work — junior vs senior software developer productivity gains.
The image at the top of this article was generated from this prompt:
"A bold Constructivist style graphic poster depicting the future of human-AI symbiosis. Interlocking geometric shapes of human profiles and digital circuitry, with the word 'EPOCH' large and integrated into the typography, and the smaller words 'Empathy', 'Presence', 'Creativity', 'Opinion/Judgment', 'Creativity', 'Hope/Vision'. Strong diagonal composition, industrial gears, and abstract network lines. Minimalist color palette of dark teal, solid charcoal, cream, coral, and light gray. (Dark / slide teal - 048BA2, Light teal - C5E6E7, Darkest - 233030, Cherry - EE7C7B, mid gray - dadada, light gray - F3F0F0) Sharp angles, high contrast, graphic design, retro-futuristic USSR avant-garde art style."
Naming a specific art style is one of the best ways to get a genuinely creative result out of an AI image generator — it gives the model a real visual language to work from, instead of defaulting to generic stock-image composition. If you want a shortcut to strong style references like the Constructivist one above, my Art Styles Tool is built for exactly this.
Earn from your design skills
Create & sell patterns for fabric, wallpaper & murals — on Spoonflower & beyond. Client-free, commute-free, creative work you can do from anywhere.
See the book →
Blue-sky thinking — starting from an unconstrained, root-cause view of the problem — is the step most UX process skips. Here's how to use it, with a real health-app example and a way to run it yourself.
My portfolio lived on Webflow for four years. The rebuild — same brand, sharper site — took days, using one free export tool and a documented design system. Here's the order that actually works, and the one thing I'd do differently.
Most multi-brand design system advice quietly stops working around brand five or six. Here's the token architecture, component pattern, and governance structure that holds up at 8+ brands.
Keep your design skills up to date with videos,
AI workflows, tutorials, ways to earn $$$ and free resources.
See past email newsletters.
Unsubscribe anytime.
