Andrew Hogan and Shane Johnston of Figma join Dan and Nik to unpack Figma's 2026 AI report. The headline is about AI shifting from individual productivity to team dynamics: 41% now say AI meaningfully changes how their teams work together, up from 7% two years ago. Shane and Andrew deliberately interrogate the "10x everybody" narrative: output is measurably up (lines of code, App Store submissions), but outcomes aren't following, because AI is exposing the communication and alignment seams that were always there. Handoffs can still be an issue. Prototypes are trivial to generate but very hard to dissect. And perhaps unsurprisingly, decision-making turns out to be the real bottleneck.

The conversation also covers the blurring of roles (designers shipping fully functional AI-built prototypes jumped from ~10% to 61% in a year; developers doing design work is up sharply) and why developers are now among the loudest voices arguing design matters more. Andrew and Shane reframe role overlap as tax reduction rather than job-stealing — the designer who fixes the visual bug instead of filing the ticket — and make the case for the canvas as shared context, for design systems as the thing "orchestrating relationships" between functions, and for the four shapes of AI adoption (the nascent group has roughly halved to 18%).

LINKS

Figma's 2026 AI Report — blog post

Full report / download

Report microsite

Andrew Hogan on LinkedIn

Andrew Hogan, "AI Fluency Isn't the Finish Line"

Shane Johnston on LinkedIn

Everyone in design is yelling the same two words right now: taste and judgment. But how do you get them? David Hoang's "Training Design Senses" does, and this episode starts there: how to train the eye, design dexterity, fluency of aesthetics, and rhetoric, with a detour into what copying from masters means when the copying happens through an AI. Then Christine Zhu on systems thinking for agentic products, which asks PMs to stop shipping features and start gardening: decompose for agent experience, define contracts between the parts, build feedback loops that capture correction signals. Our last bit of news is LLM Counter-Defaults from Nadia Piet and AIxDESIGN, a slider-based tool that exposes the settings your LLM shipped with and writes the instructions to override them.

Our guest is Dan Maccarone, founder and CEO of Charming Robot, who spent a year rebuilding his studio's entire design process around AI after a project went sideways. He walks us through the experience brief (a PRD with a friendlier name), the whiteboard/review/approved branch structure, and what he calls high-speed mediocrity. He also explains why the prototype's 300,000 lines of code are never shipping anywhere, why documentation matters more now rather than less, and why his team still starts with sketching so they're building from their imagination instead of Claude's.

LINKS

"Training design senses"
https://www.proofofconcept.pub/p/training-design-senses

”what systems thinking looks like for PM'ing AI products”
https://open.substack.com/pub/christinezhu/p/grow-systems-build-features?utm_campaign=post-expanded-share&utm_medium=web

LLM Counter-Defaults by Nadia Piet / AIxDESIGN
https://counterdefaults.aixdesign.co/

Thinking in Systems book
https://amzn.to/4wuNA6o

How Buildings Learn book
https://amzn.to/4g2Y99V

"Never mind the prompts, here's the thinking"
https://uxdesign.cc/never-mind-the-prompts-heres-the-thinking-f4af4b92155f

Charming Robot
https://charmingrobot.com

Story in a Bottle podcast
https://storyinabottle.charmingrobot.com/

The Barstool MBA (Audible)
https://amzn.to/4pGRZ3q

Fool's Gold, NYC
https://www.foolsgoldnyc.com/

MHCI Capstone sponsorship
https://hcii.cmu.edu/academics/mhci/capstone

Dan and Nik dig into a trio of OpenAI stories: the launch of GPT 5.6 Sol and its jump to the top of the Design Arena leaderboard, notably by suppressing common AI design anti-patterns (goodbye, purple gradients) and combining known templates with customized outputs — followed by Figma Make's OpenAI integration and what it might signal about the shifting model landscape. They also geek out over Codex Micro, OpenAI's slick new macro keyboard designed by Work Louder, which uses RGB-lit keys to let developers monitor and control multiple coding agents without touching a mouse.

The episode's guest segment features Jason Hong and Sauvik Das, CMU HCII faculty and co-founders of fuguUX, an AI-powered usability auditing tool. Jason and Sauvik explain how frontier language models can now scale subjective design judgment — evaluating concepts like information scent that were previously impossible to codify — and walk through their two offerings: synthetic user testing (with a diverse cast of twenty AI agents built from Census microdata) and automated heuristic audits. The conversation touches on why B2B software remains so dreadful, the testing pyramid, and the long-term vision of continuously self-improving websites.

LINKS

Design Arena
https://www.designarena.ai/

GPT 5.6 Sol topping the design leaderboard
https://notes.designarena.ai/how-openais-sol-finally-learned-design-taste/

Codex Micro
https://openai.com/supply/co-lab/work-louder/

Figma Make + OpenAI integration announcement
https://www.figma.com/blog/gpt-5-6-is-now-available-in-figma-make/

fuguUX
https://www.fuguux.com/

Dan and Nik pause at the halfway mark of 2026 to reflect on 25 episodes, 10 guests, and a wild six months in AI and design. Fresh off a week of building apps with Fable — Dan's ambient bird-listening app for AirPods, Nik's long-running multi-agent builds — they revisit the year's biggest themes: the collapse of fidelity and whether wireframes still matter, taste as the designer's (maybe) moat, cognitive surrender and the strange compulsion to keep feeding the machine, canvas versus code as Figma and Claude Design converge, designing for agents when your next user isn't human, the rise of markdown skills (and the ick of scraped design systems), the path for junior designers, and why user research may be the most undervalued discipline of the AI era.

LINKS
Scraped Styles MD repository
https://styles.refero.design/

Reverse Centaurs Are the Answer to the AI Paradox
https://pluralistic.net/2025/09/11/vulgar-thatcherism/

Aaron Draplin Takes On a Logo Design Challenge
https://www.linkedin.com/learning/aaron-draplin-takes-on-a-logo-design-challenge

How tech workers are feeling in 2026
https://www.lennysnewsletter.com/p/how-tech-workers-are-feeling-in-2026

Why using AI leads to Brain Fry
https://hbr.org/2026/03/when-using-ai-leads-to-brain-fry

Nik and Dan catch up on the news out of Figma Config 2026 — where generative plugins stole the show for Nik (build them with tokens, run them for free) — before digging into Anthropic's big Claude Design 2.0 overhaul, with canvas editing, design-system imports, a two-way round-trip with Claude Code, and a fix for its token-burning problem. From there: MC Dean asks who actually owns the quality bar when anyone can ship, Emily Campbell reimagines the layers of UX for probabilistic AI systems (goblins included), and Perplexity's new Brain points toward self-improving memory for agents. Plus a detour into AI memory, forgetting, and one very confused Claude that thought the show got renamed. Stick around for news on Dan's forthcoming sci-fi novel.

Show notes:

• Anthropic ships Claude Design 2.0 (VentureBeat): https://venturebeat.com/technology/anthropic-ships-major-claude-design-overhaul-with-design-system-imports-code-round-trips-and-a-fix-for-its-token-burning-problem
• MC Dean, "Who Sets the Quality Bar?": https://marieclairedean.substack.com/p/who-sets-the-quality-bar
• Emily Campbell, "Layers of AI Experience": https://emilycampbell.co/writing/layers-of-ai-experience
• Perplexity, "Self-Improving Memory for Agents" (Brain): https://www.perplexity.ai/hub/blog/self-improving-memory-for-agents
• Dave Song, "Control over Context" — user-led memory management concepts: https://medium.com/ui-for-ai/control-over-context-44d975e92efc
• Dan's upcoming novel, The Restoration (D.S. Duffy): https://dsduffy.com/

Dan reports live from Figma's Config 2026 conference in San Francisco, talking with seven design leaders — Peter Merholz, Jesse James Garrett, Greg Petrov, Ryan Freitas, Kim Lennox, Alice Quan, and Emily Campbell — about the keynote, major announcements like Code Layers and Weave, and the vibe on the floor. The conversations circle a few big tensions: Figma's push into Adobe's territory (motion, brand, marketing) versus doubling down on AI-to-code workflows where competitors like Claude Design are gaining ground, the shrinking of design teams, and how designers can stay strategic by leaning into craft, taste, and curiosity. Off-stage, the mood was one of cautious adaptation — less panic than last year, more pragmatic grappling with what roles, tools, and identities look like now.

A year ago the consensus was that AI would hurt Apple. Last week at WWDC they showed up with a real Siri AI, the privacy story nobody else can match, and the $$$ hardware to run it. The "AI for the rest of us" play might just work. Dan and Nik dig into what Apple actually shipped, why App Intents matter for designers, and the gut-punch reality that your year-old iPhone won't run it.

Also this episode: Airbnb CEO Brian Chesky is building a design-led AI lab, and nobody can quite figure out what it's selling. Sarah Gibbons maps the four design jobs AI has created so far. And Perplexity's research with Harvard Business School claims agents can cut task time by up to 92% — with enough caveats to fill a show.

LINKS
Secret AI Lab News
https://www.bloomberg.com/news/articles/2026-06-04/airbnb-ceo-brian-chesky-plans-to-start-a-new-ai-company

The Four Design Jobs AI Created (So Far)
https://www.nngroup.com/articles/design-jobs-ai-created/

Perplexity/HBS Research
https://arxiv.org/abs/2606.07489

Thinking Machines Demo Video
https://youtu.be/A12AVongNN4?si=ODvSCf9IKTzwBiiU

Dan and Nik dig into Microsoft's sweeping redesign of 365 Copilot , transforming the prompt box into a task-aware workspace, leaning into progressive disclosure, and positioning AI as an embedded layer across Word, Excel, and Outlook rather than a bolt-on feature. They also discuss Jeff Gothelf's argument that "done" means something fundamentally different for AI products, where probabilistic outputs require teams to write acceptance criteria as distributions and build failure triage into launch rather than after. Rounding out the news: Amazon's Proteus warehouse robots can now take natural-language direction from human workers, and a Salesforce designer shares how her team is mastering AI tools through peer-to-peer learning, a dynamic that gives Dan flashbacks to the early days of the web.

Then Nik sits down with Mark Boyes-Smith, Head of AI Design at Miro, for a deep look at how one of the world's largest collaboration platforms is integrating AI at every layer. Mark walks through Miro's new Sidekicks and Flows features, including voice interaction and custom connectors, and talks candidly about how his distributed design team rapid-prototypes in code, branches and remixes ideas, and builds conviction before shipping. He also shares his thinking on designing for long-running agentic tasks, the coming challenge of multi-agent orchestration on an infinite canvas, and what resilience and "product taste" mean for designers navigating an era of relentless change.

LINKS
Microsoft Copilot 365 Redesign
https://www.microsoft.com/en-us/microsoft-365/blog/2026/05/28/introducing-a-new-design-for-microsoft-365-copilot/

What “done” means when you’re shipping AI features
https://jeffgothelf.com/blog/what-done-means-when-youre-shipping-ai-features/

Amazon's new Proteus warehouse robot

https://www.engadget.com/2187338/amazons-new-proteus-warehouse-robot-is-fully-autonomous/

Designing how designers master AI

https://uxdesign.cc/designing-how-designers-master-ai-642d8751d945

New Miro Features at Canvas26

https://www.youtube.com/live/qM-KxCr4I5I?si=HGtSokg95B9kO5k9