Multidisciplinary Is the New Specialist

The specialist designer is being squeezed from both sides. AI tools handle execution. AI agents handle implementation. What remains valuable is the connective tissue between disciplines, and that connective tissue finally has a widely-accepted name: design engineering.

This is not a forecast. It is something already happening to the people I work with, the people I follow, and the people teaching me. The squeeze is field-wide, the response is field-wide, and the label is starting to stick.

What follows is an attempt to name the shape, point at the signals that announced it, and show why a multidisciplinary practice is no longer a quirky bonus on a CV but the structural answer to a profession that is being unbundled.

Three signals from 2024–2026

Three (and a half) moments from the last two years point at one structural change in how product work gets made.

On February 2, 2025, Andrej Karpathy posted on X:

"There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists."

The phrase landed because it described something founders and designers had already been doing quietly: shipping software without writing software. The role boundary blurred in public.

A year earlier, Marty Cagan and Bob Baxley at SVPG published The Era of the Product Creator. Their argument was that generative AI is collapsing the cost of producing both interface and code, so strong designers and engineers can now plausibly take on the product-creator role that used to require a dedicated PM in the middle. The handoff chain shortens.

Then in 2024, Vercel made the role institutional in Design Engineering at Vercel:

"Design Engineers at Vercel blend aesthetic sensibility with technical skills, which allows them to deeply understand a problem, then design, build, and ship a solution autonomously."

Note the word autonomously. That is the squeeze, written into a job description.

The fourth beat is closer to home. Luděk Černocký, VP Product Design at Amplify and the lecturer of T1 of this MasterClass, opened his 25.2.2026 keynote with a one-line thesis: "Když je výstup levný, úsudek se stává drahým" — when output is cheap, judgment becomes expensive. He revised himself live, too: his earlier Shape of Design to Come predicted AI would dissolve the junior/senior boundary; in the Q&A he said he was wrong — juniors have it harder now.

Four independent voices, one structural change.

The squeeze, both sides

The pressure is symmetrical, which is what makes it interesting.

On the execution side, the cost of producing a polished interface is approaching zero. v0, Lovable, Bolt.new, and Framer AI turn briefs into shipped surface. A junior with taste outputs more in a weekend than a mid-level studio used to ship in a sprint.

On the implementation side, code is collapsing the same way. Cursor, Claude Code, and Replit Agent close the gap between intent and shipped feature. An engineer with judgment runs an agent loop and produces in a day what a small team used to scope over a week.

What survives the squeeze is none of the artefacts. What survives is ownership of the gap: the loop from research to decision to shipped product, held by one person or one tight pair, with no handoff translation losses in the middle.

T4 of this MasterClass (Design leadership v době recese) frames the same compression from the opposite end: budgets tighten, teams shrink, survivors hold more of the loop. Recession is the multiplier, not the cause. Černocký closes from the speed side: "Pokud rychlost už není problém, je důležité vědět, co nestavět" — when speed stops being the constraint, strategy is knowing what NOT to build.

Design engineering as the label

Once a shape has a name, you can hire for it, teach for it, and argue about it. That is what design engineering now is.

The Vercel post linked above is the cleanest institutional definition: one practitioner, end-to-end, autonomous through the whole loop. It is not a job description bolted together from two existing ones. It is a single integrated practice with its own taste, its own tradeoffs, and its own deliverables.

Maggie Appleton has been cataloguing the role since around 2022. Her directory keeps selecting for the same shape: people who design with materials, not with mockups.

Hayden Bleasel writes about his library Streamdown becoming too complex for a human to maintain and now fully agent-maintained through Claude Code with skills and unit tests. That is not a designer who learned CSS. That is a practitioner who has restructured his work so an agent handles the surface area he used to.

The distinction matters. A designer who knows CSS is a designer with a useful skill. A design engineer is someone whose practice is the integration itself: research, decision, artefact, and shipped product live in one head and one workflow. The skills are a consequence of the shape, not the definition.

What does that integrated practice look like in a concrete product? That is the next section.

Anchor case: the connective-tissue product

For eight months I have been the AI adoption contractor at a fintech client. The work in front of me is an internal full-stack knowledge-sharing portal. It exists because the gap is not the kind a specialist solves.

The official narrative said staff needed AI literacy training. The real gap was different. AI artefacts already existed: prompts, skills, CLI tools, Demo Day recordings, agent setups. They lived in Teams threads, in Confluence pages no one could find, in personal folders. There was no canonical entry point, no way to distinguish official material from a weekend experiment, no way for a tester to find that another tester had solved the same problem three months earlier. Non-technical roles had nowhere to look at all.

The UX decisions fell out of that understanding. A six-type content taxonomy (inspiration, guides, support docs, reusable agent skills, tools, demo recordings) so a marketing manager and a backend engineer both find their entry point. Origin labels so endorsement is legible. A kanban approval queue so the surface stays curated. An MCP endpoint planned in a later phase so the catalogue can be queried from inside Claude Code without opening a browser. That last decision is not one you can make if you do not know what an MCP endpoint is.

The stack was chosen for solo velocity. A server framework, Postgres, a React frontend, embedding-based duplicate detection, nothing requiring a platform team to operate. The trade-off was honest: single-node deployment, fewer features. I could have shipped slower with a richer setup. I would have shipped never.

The hardest part was not the code. It was convincing an organisation with no formal AI maturity strategy that this should exist as a product, not another Confluence page. I wrote the PRD and ran the stakeholder pass under no PM cover.

This product would not have happened if it required a three-role trio. There was no trio. There was me, a repository, a PRD I wrote, and an agent loop. The reason it shipped is that I was the gap.

Two more shapes of the same thing

The fintech portal is one shape. Two smaller ones from the same practice round out the picture.

Pockets. A solo macOS app I designed, coded, and ship daily. Swift in Xcode. AI is real: Claude Code as the agent layer, GPT Vision for capturing receipts from photographs. No team, no handoff. Receipt capture is where Topic 6 of this MasterClass (Design, Ethics & AI) stops being abstract: vision capture raises data-ownership questions a sole designer-engineer cannot offload to an ethics review board.

jholec.com and the SessionEnd hook. My portfolio runs on Framer and auto-documents itself through a bash hook I built for DESB45: on session end, jq parses the transcript, a Haiku call summarises it, and an entry lands in an "Experiences vault" of markdown logs. The post you are reading is itself an artefact of that pipeline. Yesterday's Topic 5 (Speculative Design Normalization) is my current daily build.

Build in public as the cultural layer

Design engineering has institutional definitions, but its cultural infrastructure runs on X and Threads under the build-in-public banner. The movement is what makes the role legible at scale.

Pieter Levels is the founding figure: radical transparency, ship fast, paid tiers from day one. Marc Lou made the meta-move, building in public and then selling the picks-and-shovels (ShipFast boilerplate) to everyone else doing the same. Tony Dinh is the design-tool flavour of the same practice (DevUtils, Black Magic). Steve Schoger and Adam Wathan are the design-craft expression: Refactoring UI and Tailwind grew in public.

The Indie Hackers community has openly asked whether top builders going quiet at scale signals the end of the format. Survivor bias is real. The movement matters, but treat it like a movement, not a manual.

What we're losing

Critical reflection is mandatory here, and the line that most cleanly frames the loss comes from DOC's craft-crisis piece:

"These tools don't make design accessible to everyone. They make design accessible to people who already understand what they're looking at."

The junior pipeline is the first loss. Welcome to the entry-level void (It's Nice That, September 2025) frames the trend; the underlying WEF analysis of Randstad data, 126 million postings, puts entry-level postings down 29 points since January 2024.

Accessibility is the second. Adrian Roselli audited Figma Sites in May 2025; one demo produced over 200 WCAG violations. Generation is not correctness.

Slow craft is the third. Bootcamp's Human After All reads against the squeeze: judgment and empathy are not commoditised. Ask before observing, observe before generating. The faster the tools, the more deliberate slowness has to be.

The lecture concepts, named

Topic 1 (Luděk Černocký), Leading Design in a World Where Anyone Can Build, is the spark this post extends. Černocký names the emerging shape the rise of the product builder — someone who "thinks, decides, builds, and owns it." Same shape as design engineering, different label; this post keeps "design engineering" for its institutional weight (Vercel, Appleton) and credits product-builder as the parallel MasterClass framing. His closing line is the cleanest summary of the thesis: "Když může tvořit kdokoliv, vyhrávají strategie, odpovědnost a úsudek" — when anyone can build, strategy, responsibility, and judgment win. Topic 4 is the recession multiplier named earlier. Topic 5, Speculative Design Normalization, is the daily build in the second panel above. Topic 6, Design, Ethics & AI, is the layer the first panel above cannot offload.

AI transparency appendix

Tools: Claude Code (Opus 4.7, 1M context) as the writing environment, Framer MCP for publishing, my framer-portfolio-blog skill for house style, and a general-purpose research subagent for source discovery.

AI's role was sparring on outline, source curation, and draft editing. The thesis was not generated; it was locked by me in a brainstorming pass with the tool playing devil's advocate. The anchor case and the lecture callbacks are from my own notes and project work. AI surfaced sources for the signals, the label, the build-in-public layer, and the losses.

Sources

People.

• Marty Cagan & Bob Baxley, The Era of the Product Creator, SVPG (2024). Link

• John Maeda, Autodesigners on Autopilot (Design in Tech 2025), Medium (2025). Link

• Maggie Appleton, Collection of Design Engineers, maggieappleton.com. Link

• Hayden Bleasel, personal site. Link

• Pieter Levels, How I build my MVPs, levels.io. Link

• Marc Lou, ShipFast. Link

• Tony Dinh, personal site. Link

• Steve Schoger & Adam Wathan, Refactoring UI. Link

• Julie Zhuo, From managing people to managing AI, Lenny's Newsletter (2024). Link

• Andy Budd, Makers to Curators, Medium (2024). Link

• Luděk Černocký, The Shape of Design to Come, Bootcamp / Medium (2025). Link

• Luděk Černocký, Leading Design in a World Where Anyone Can Build, MasterClass keynote (25.2.2026). Link

Tools.

• Vercel, Design Engineering at Vercel, Vercel blog (2024). Link

• Vercel, v0. Link

• Lovable. Link

• StackBlitz, Bolt.new. Link

• Anysphere, Cursor. Link

• Anthropic, Claude Code. Link

• Replit, Replit Agent. Link

• Framer, Framer AI. Link

Concepts.

• Andrej Karpathy, Vibe coding tweet, X (2025-02-02). Link

• Ted Goas, We'll all be design engineers in a year, UX Collective (2024). Link

Counterpoints.

• DOC, Why AI is exposing design's craft crisis, doc.cc (2025). Link

• Bootcamp, Human After All: Skills AI Can't Replace, Medium (2025). Link

• Simon Willison, Not all AI-assisted programming is vibe coding, simonwillison.net (2025). Link

• It's Nice That, Welcome to the entry-level void, itsnicethat.com (2025). Link