ai goodies ✨
Posts
how designers actually work in 2026 / 30+ AI-infused workflows

how designers actually work in 2026 / 30+ AI-infused workflows

real tools, real prompts, real agent architectures / what my community shared when I asked how they actually work now plus the patterns hiding inside all of it.

Ioana from AI Goodies
August 06, 2026

✋ Before we get started, I need your help. AI Goodies is still a small, independent effort, and it needs your help to grow 🌱

so if you enjoy this read + my vibe, please share the article on your platform of choice, send it to friends, recommend others to subscribe ✔️

two community projects you should know about 👇

🚀 the AI × Product Design Accelerator > a four-week live program for designers ready to build and ship in the AI era. we’ll also be shipping your state-of-the-art portfolio. 😎

🌡️ the Design Wellbeing Index > my independent annual study of the emotional and economic reality of design work. my most important work. Survey fields this September, sign up to participate and / or get the results.

also, a full-circle moment happened while I was putting this issue together 🥹 reading through the workflow responses, I spotted Paper listed in someone's stack > and I almost replied "they sponsor my newsletter!" haha 👾

not gonna lie, that felt amazing. and it tracks with why I partnered with them in the first place: I do believe Paper points at the future of design tooling. watching it show up organically in the wild, inside real workflows, next to Claude Code and Figma > that's the strongest evidence there is.

yay, thanks + let's get to it.

Last week I posted a question on Linkedin: how do you actually work now? Yeah, I wasn’t looking for the aspirational workflow, the ideal, post-perfect path, but for the real, messy, Tuesday-afternoon version of how people design in 2026.

The comments turned into one of the most valuable threads I've ever hosted. Over 30 designers shared their actual setups / tools, prompts, agent architectures, the parts they automated and the parts they refuse to. I promised I'd document it all in an article, so here it is: a consolidated map of how designers really work in 2026, built from the community's own words plus the industry data to frame it.

And before we dive in, the backdrop matters. The AI in Design Report 2026 found that 91% of designers now use AI weekly, and three out of four use it daily / a jump that usually takes years and happened in twelve months. Figma's State of the Designer 2026 found 72% using generative AI with 98% increasing usage year over year, and / this one stopped me / the number of designers participating in development doubled to 41% in a single year.

So the workflow didn't just change > it fractured, expanded, and rebuilt itself.

Here's the new architecture, pattern by pattern.

pattern one / the terminal became the studio

The single most consistent thread across every workflow shared: Claude Code is the new center of gravity. And I mean as the operating environment for the entire design practice, way beyond a coding tool. 📟

Christopher Starr, who builds Pirch, describes it best: "my whole agency is a folder of markdown files it reads every session / clients, project dossiers, every Figma Make prompt ever sent and what it produced / so I never re-explain context. AI does all the production: screens, code, drafts, ops. I keep the judgment: which screen should exist and when the generated thing is confidently wrong." He's shipped an App Store app and a full product this way. And he doesn't write code.

Michał Idziak runs most of his workflow through Claude Code and Cursor "with all necessary MCPs connected / Figma, Atlassian, Miro, Gitlab, Google." Martin Rud controls everything through Claude Code in VS Code, with agents that command camera angles, styles, scenes and light across his image and video tools. Alexandra Pituru runs Claude Code + GitLab with / and I love the precision of this metric / a 100% merge rate, using Figma "exclusively for illustrations or marketing-related subjects, otherwise I'm kinda out."

The terminal / the most engineer-coded surface in software… has become a design studio. That would have sounded kinda’ absurd three years ago, right?

pattern two / Figma got a new job description

Nobody abandoned Figma (and Design is NOT Dead). But almost everyone renegotiated their relationship with it. The demotion from "the center of everything" to "a specific tool for specific moments" came through in nearly every comment, each with its own flavor:

Kyle Shepherd: "At this point Figma is essentially a massive playground and scratchpad for me." He designs in code for most work, and keeps Figma for the abstract territory / branding, narrative, experimentation.

Nikos Antonogiannis: "Figma is still open. It just stopped being the centre." He doesn't start on a canvas at all: "I write the reasoning down, argue with the model about it, then go to functional code. The working prototype is what users see. Static screens only when I need to align stakeholders."

Claudia Bogun: "Figma is the structural record now, not where the values live" / her canonical values live in tokens.css.

Alexandru Cristian Popescu, Senior UX/UI Designer at Ubisoft: "Figma comes after, not to explore but to fix the thing forever and build the design system."

Erik Ladňák: "Figma stays, but the weight moved to the spec."

Notice the pattern inside the pattern: Figma survived by specializing. Playground. Structural record. Stakeholder alignment. Design system archive. The exploratory and production work migrated elsewhere / and what remained is what Figma is genuinely best at. 👾

(Worth noting the infrastructure that enabled this: Figma's MCP server launched in mid-2025, and in February 2026 the bidirectional Claude Code integration arrived / Design to Code builds from selected frames, Code to Canvas reconstructs a running UI back into editable Figma layers. The two worlds now talk to each other natively, which is exactly why designers can move fluidly between them.)

pattern three / the chats are disposable, the brief is the institution

This might be my favorite discovery from the whole thread: the real infrastructure of the 2026 design practice isn't any tool: it's persistent context > the second brain that survives between sessions.

Alexandru coined the line I can't stop repeating: "As I learned, the chats are disposable, the brief is the institution." He fixes rules, caveats, and direction in a Claude Code project, and everything else flows from that.

Tatiana Sukhova built what she calls a Brand DNA: "our knowledge, skills, and assets packaged so me or anyone on my team can build anything on-brand, in any agent, in minutes. It's agent-agnostic by design, but we live in Claude." She prototypes complex experiences "20x faster / not faster mockups, faster thinking."

Christopher's folder of markdown files / one fact per file, every prompt logged before it's sent / is the same principle. So is Shayla Khezi's reference library, where every reference is saved with a note on what problem it fixes, so when a product has a confusing pricing page, her agent pulls references that solved confusing pricing.

Claudia takes it further with process discipline: one session per kind of decision, each closed with a handover document. "Nothing crosses strands silently. A copy decision surfacing in a build session gets logged and handed over, not fixed on the spot."

And Allen D. names the meta-realization: "I didn't end up building a workflow, I ended up building the garage the workflows live in." His model shifted from Design → Build → Ship to Observe → Preserve → Relate → Design → Build → Reinforce / systems that accumulate understanding over time.

This is the actual skill divide emerging in 2026: who can build context infrastructure that compounds. Prompting is table stakes / the institution you build around the prompts is the differentiator.

pattern four / designers became team leads of agents

The second structural change: designers are no longer using one AI. They're orchestrating several, with deliberate role assignments.

Stanley Vaganov treats his stack like a staffed studio: "Replit is my dev, Claude is my designer, GPT is my project manager. I will start with GPT outline then ask it to write cohesive prompts for Replit to develop and for Claude to design."

Alexandru describes his org chart: "a higher model as brain and coordinator, the Tower, holding the main brief, with Opus and Sonnet as the hands. And skills, skills, skills, always handcrafted, the red team one being especially useful."

Merilly J Ruglas "built an entire support team of AI agents" across Claude, ChatGPT and Gemini for research, concept validation, editing, and concept generation.

And Shayla's redesign pipeline is a full agentic production line: the agent researches the target and writes a brief → captures the current state in screenshots → builds a moodboard from her annotated reference library → drafts the redesign prompt → runs the exploration itself in Claude Design or her terminal → sends the chosen direction to her Figma file through the MCP for manual refinement. Four to five hours per redesign, down to 1.5.

The job title still says designer. The day-to-day increasingly resembles creative direction of a small, tireless, occasionally overconfident team.

pattern five / the universal split: AI produces, humans judge + decide

If there was one sentence structure that repeated across almost every single comment, it was this: "AI replaced X. What I still do myself is judge."

Paul Boag, 30+ years in UX: "AI has completely taken over the mundane end of things. What I still insist on doing myself is the strategic layer. AI is hopeless at that because it lacks the organizational context."

Michał: "The skills surface; I judge. AI isn't replacing design taste / it's removing everything between me and the moment I use it."

Aakif Umar: "AI generates layouts, but it can't judge whether something's right."

Sara Darbi: "The parts I still insist on doing myself are the human ones / asking the right questions, making sense of what I'm seeing, and using judgment to decide what to do next."

Christopher, again, with the line I'd frame and hang on a wall: "the design skill didn't get replaced / it became the spec."

This maps exactly onto what I've been writing all year / the deciding versus the making, the curatorial turn, design at a higher altitude. The community has independently converged on the same conclusion. The value relocated from production to judgment. Everyone can feel it. 👾

but / the uncomfortable part

I want to honor the critical voices in the thread too, because they raised something important that the enthusiasm can drown out.

Bozena Bulat asked the question that haunted me: "Anyone else feel like AI sped up everything except the thinking?" She still does the messy understanding part by hand, "especially in regulated flows where one fast wrong turn costs a lot."

Cem Kurtaraner was ruthlessly honest about his own behavior: "It's faster, but I catch myself exploring less and selecting more." And he named an enterprise pattern I've started seeing everywhere: "Teams increasingly bring us AI-generated mini-products instead of user problems. Because these prototypes already look like solutions, people become attached to them." His summary is the sharpest sentence in the entire thread: "We've lowered the cost of proposing solutions, but not the cost of judging them."

Claudia defended the sketch against the prompt: "A prompt hands me something plausible in ten seconds and I have decided nothing. A sketch stays undecided until I decide." Sit with that one. It's a whole design philosophy in two sentences.

And Nikos pushed the critique to its logical end: "This whole thread agrees that AI produces and we judge, and nobody is checking the judging." So he built a check / SprintZero, a two-agent system where an adversarial QA agent attacks the research output before it reaches him, "including the conclusions I would have agreed with." His context: 17 years designing regulated financial platforms, "where a fluent wrong answer is expensive."

This is the frontier question of 2026, and I don't think we've answered it: if judgment is our remaining value, what's our quality control for judgment itself? Nikos is one of the only people I've seen actually build an answer.

the consolidated map / what the 2026 workflow looks like

Zooming out from 30+ workflows, here's the shape that emerges:

the reasoning layer / the work starts in language now, not on a canvas. Writing the brief, arguing with the model, fixing constraints before anything visual exists. Manuela Paun's opening prompt teaches constraints before delegating anything: palette, icon style, lighting / "confirm you can reproduce this, then wait for my first request."

the context layer / the second brain: markdown files, Brand DNA, reference libraries, handover documents, design.md files. The institution that survives disposable chats.

the production layer / Claude Code, Cursor, Claude Design, Figma Make, Replit, Framer / where agents generate screens, states, prototypes, and code, wired to real design systems through MCPs.

the judgment layer / human, still, entirely. Which screen deserves to exist. Whether the fluent answer is wrong. When to reset the question from "how do we build this?" back to "what problem are we solving?"

the refinement layer / and here Figma returns, for what it does best: fixing the thing forever, design systems, stakeholder alignment, the last pass of craft.

The tools will keep churning. But I suspect this five-layer shape / reason, contextualize, produce, judge, refine / is the durable architecture of the practice for the next few years.

how to start building yours

If your workflow still looks like 2023 and this all feels overwhelming, here's the distilled advice from the people in the thread:

start with dictation and dialogue, not automation. Kyle works with Claude his entire workday, but a lot of it is thinking: "brainstorming, verbally dumping the design changes I want to see, clarifying my own thinking, articulating what I actually want built before code gets written." That's an accessible entry point / use the model as a thinking partner before you use it as a production engine.

build your brief before your stack. The people getting the most out of these tools all invested in persistent context first. One document with your rules, constraints, style, and direction. The chats are disposable, remember. The brief is the institution.

teach constraints before delegating. Steal Manuela's move: define the palette, the style, the rules, ask the agent to confirm / then start. Otherwise, as she puts it, "he'll design whatever he imagines."

automate the audit, keep the verdict. Michał's UX audit skill / grounded in real customer success signals, outputting prioritized annotated flows / is the template: let AI surface, you judge.

and protect your thinking time. Bozena's warning is the one to tape to your monitor. Speed is seductive. Exploration collapsing into selection is the failure mode. Sometimes the sketch that stays undecided is doing more work than the prompt that decides too fast.

thank you

To everyone who shared their workflow in that thread: this article exists because of your generosity. 🙏🏼 This is what I love most about this community / the willingness to open up the messy, half-baked, honest parts of how we actually work, so we can all figure out this transformation together.

The full thread with all 30+ workflows is here / go meet these people, their setups are worth studying in detail.

🔥 let's collaborate 🤝 get featured in my newsletter

We have the coolest audience that you can get yourself in front of.
Reach out at [email protected] if interested 💋

✨ for more goodies, follow me on:

UX Goodies on Instagram

Ioana Teleanu on Linkedin