Ep #22: Sara Vienna (Metalab): The Trust Gap, Tools for Everyone, and Why Nobody Reads Case Studies

Tom and Jonny sit down with Cassie Groos, design systems lead at Unily, to learn what happens to a design system once AI starts reading it. They dig into the prose documents that brief her agents, unpack a token architecture built for an infinite number of brands, and go behind the scenes of her first big stage at Hatch.

Tom and Jonny sit down with Christine Røde, designer at The Browser Company and the first guest under our new name, Mostly Working. They go behind the scenes of Dia’s Morning Brief, unpack why people now hand their data to AI tools so readily, and dig into the craft a designer keeps when she can ship her own fixes. In this episode: Dia’s Morning Brief exists because most people stall at a blank text box, so the browser sends the prompt for you and opens a tab that briefs you on your day. Christine came back from two weeks’ holiday to a useful prototype that looked like ‘pure slop’, and rebuilt it around hundreds of hand-picked public-domain paintings from the National Gallery of Art. She designed Dia’s memory flows with lock icons everywhere, then watched user testing show that people were happy to tick the box. Arc shipped something new every week and ended up with around 35 branded features, whereas Dia builds and tests plenty and ships only what adds enough to the product. You get eight or nine nitpicks before an engineer has had enough of you, and Christine now ships those fixes herself the week after launch. A year ago she was asking her head of design whether she’d still have a job, and today she’s enjoying the work more than she has in years. Chapters: 0:08 — Welcome to Mostly Working, and introducing Christine Røde 0:46 — From a Mostly Working meetup talk to Dia’s Morning Brief 1:21 — What Dia is, and how it has changed as AI has 4:08 — The blank text box problem, and a brief that arrives on its own 5:57 — The first prototype: beige, serifs and ‘pure slop’ 7:00 — Finding the imagery: SVGs, big type or public-domain paintings 8:32 — Tom on what design adds to a prototype that already works 11:04 — Signal over volume: why the brief stays short 14:54 — Jonny on promising two or three things, and design reacting to engineering 16:37 — Why AI quality comes before design fidelity 19:01 — Onboarding, and persuading people to connect their tools 20:42 — From WhatsApp-era privacy to ‘you can all just read my email’ 24:08 — Trust, and design’s part in earning it 26:45 — Designing Dia’s memory feature for users who weren’t worried 28:53 — What faster shipping does to the quality bar 30:09 — Why Dia ships less often than Arc did 32:01 — Designers shipping their own polish 34:12 — Working in seasons, and the hectic first month 38:24 — Custom tools for curating paintings and sketching interactions 42:08 — Real art as a brand decision, and hiring a painter 45:45 — What it means to be a designer now 48:17 — Quick-fire: free time after AGI, and Lime bikes 50:54 — The part of her craft Christine won’t hand to AI 52:09 — A rooftop garden, and the last 20% of vibe coding 53:31 — Final thoughts, and where to find Christine

Tom and Jonny announce that the Near Future Podcast is becoming Mostly Working. They explain why they're closing the consultancy to back the podcast, unveil the Mostly Working Summit at Monzo on 22 October, and go behind the scenes of a new brand built with Claude Code. In this episode: The Near Future consultancy is closing, and the podcast carries on every week as Mostly Working. We explain the three tests behind the decision: founder fit, how hard the work was to sell, and how much energy it gave us. At design + AI in Brighton, 30 senior designers built their own design tools in about an hour, from a Crazy Eights racing game to a skin analyser. The Mostly Working Summit runs at Monzo on 22 October: a free morning for 100 design leaders, then our final Build Your Own Design Tool workshop. Apply at mostlyworking.co. Tom walks through building the new brand, sketching the mark in Figma and having Claude Code generate palette tools, motion principles and even a jingle. Chapters: 0:04 — The Near Future Podcast is becoming Mostly Working 1:03 — How the podcast started, 19 weeks in 2:17 — Winding down the Near Future consultancy 3:02 — Where the Mostly Working name comes from 4:30 — Founder fit, ease of sale and energy: why the consultancy - stopped 6:53 — Jonny on wanting to build, and joining a small tech company 11:09 — Tom's test: follow the most interesting conversations 13:21 — Thank you to Near Future clients 14:30 — What to expect from Mostly Working 17:24 — Build Your Own Design Tools at design + AI in Brighton 20:43 — Why the workshop builds keep getting better 23:15 — Hatch in Berlin: Paula Scher and Bob Baxley 26:33 — Announcing the Mostly Working Summit 27:57 — How the Mostly Working meetup grew from 70 to 160 sign-ups 31:45 — Summit details: Monzo, 22 October, fireside chats, roundtables and the workshop 36:13 — mostlyworking.co and the new brand 37:20 — Designing the M/W mark in Figma 39:09 — Claude Code generates palette tweakers, motion principles and a logo jingle 43:37 — Building brand tools by voice note from your phone 45:29 — Flora versus building your own image generators 47:22 — Tote bags, feedback and wrap-up Mentioned in the episode Mostly Working The Mostly Working Summit: AI x Design 2027 — Thursday 22 October 2026, Monzo, London Apply for the summit or book the workshop Mostly Working meetup Events design + AI, Brighton, organised by Product Unleashed Hatch Conference, Berlin People Tom Harman Jonny Burch Simon Amor, co-founder of Morse and co-host of the Mostly Working meetup Paula Scher, partner at Pentagram Bob Baxley, designer and former head of product design at Pinterest Domingo Widen, Fin Ini, Fin James Storer, Monzo Companies Morse (formerly Sling Money) Monzo Fin Pentagram Tools and products Claude Code Claude Opus Figma FLORA Suno OODA About the pod Mostly Working (formerly the Near Future Podcast) is a weekly conversation with design leaders about what they're building with AI. Send feedback to [email protected], and find everything at mostlyworking.co.

Tom and Jonny sit down with Gabriella Hachem, co-founder of Dessn and the podcast’s first design tool founder. They dig into why Dessn runs your codebase in the cloud so you can design on production, unpack the read-only decision that shaped the product, and go behind the scenes of building a design tool while the models caught up. Key takeaways Dessn runs your codebase in the cloud so you can design on top of the real thing. No local setup and no coding, and the page you pull onto the canvas is your production homepage rather than a representation of it. Gabriella has waited her whole career to open the real homepage and work on it. Read-only access to the repo was a focus decision before it was a security one. The first version of Dessn opened pull requests. They dropped it, because opening a PR isn’t the hard part anymore and it was never how they were going to win. Handoff now runs through MCP straight into Cursor or Claude Code. Leaving the engineering gate where it is smooths adoption. Workflows are already changing, and change is heavy even when it’s powerful. Keeping one familiar checkpoint untouched means a design team only has to renegotiate the part that actually matters. Dessn’s ideal customer profile turned out to be a culture, which is a very hard thing to qualify for. Every usual deterministic signal failed: vertical, size, stage. The proxy Gabriella watches for is whether a design engineer role already exists on the team. An excited user counts for nothing if leadership isn’t bought in. Gabriella has watched a deal die at that gate with the user still cheering for it. Nim is the one who says no, and she’s learning to, because you only have so much time and where you spend it matters. The handoff only ever existed because designers couldn’t reach the code. Remove the skill barrier and you remove the reason for the relay. Fewer people between the decision-maker and the executor gets you higher quality output and more speed. Roles were built on skills, and skills are the part that’s collapsing. You knew Figma so you were a designer, you knew how to code so you were a dev. Gabriella never became a Figma expert and still considers herself a good designer, because the thinking layer was always the job. Dessn’s learning curve is a curve of power. It’s mostly prompt-based, so there’s no tool to master, but there is a blank canvas problem: what can I even do in here? Ask it for a dropdown of every state a user could be in and it pulls them straight from your code. The models caught up to the vision, and that’s what conviction buys you. It didn’t work, it didn’t work, it didn’t work, and then it worked. Dessn never rode the hype wave, because the product couldn’t deliver on the promises yet, and now it can. The industry has gone from FOMO to fatigue after a slew of broken promises. Gabriella wants the headline to be exactly that: no broken promises. She’s blunt about competitors who answer the source-of-truth problem by creating another source of truth. The designer’s job is heading towards curation, and that was always the high-leverage part. Deciding this is the right thing to do was always the leverage. Her worry, one year out, is how much Dessn builds for the requests arriving today versus the version of design work she thinks is coming. “When did a feature become synonymous with a role?” An investor asked Gabriella where the visual editor was, given she builds for designers. Direct manipulation comes up constantly, and she thinks the industry should reopen that question rather than assume the answer. Chapters 0:04 — Our first design tool founder: Gabriella Hachem and Dessn 1:18 — Montreal, first jobs, and never fitting the boxed roles 2:22 — She didn’t quit to build a design tool. She quit to solve a problem 3:34 — Design and prototype in your codebase, no local setup, no coding 6:42 — Jonny built a Temu version of Dessn for a hackathon 8:38 — Read access or write access to the repo, and why Dessn went read-only 11:

Tom and Jonny sit down with Frank Bach, a product designer at Anthropic who built his own design tool during onboarding week. They dig into why he prototypes in HTML instead of Figma, unpack his Swiss Army knife career from Red Bull to Headspace to Instagram, and get his advice for designers feeling the AI squeeze. Key takeaways Build the tool while you build the work. Frank made his design tool feature by feature during his first project at Anthropic, adding each part the moment he hit the need for it. A share function when he wanted to show his PM, then a Loom-style recording feature the next day, camera feed and all, in about three prompts. Load a control and a variant side by side and click them in sync. The most unusual thing he built steps two prototypes through the same screens on a single click. He is hoping Figma steals the idea. Most people inside a company have never seen the screens they own. Everyone at Headspace got the app free, so nobody had laid eyes on an upsell screen in years. A live side-by-side of current state and proposed state is what gave leadership the confidence to green-light the work. Begin anywhere. Frank has turned John Cage’s line into a working design principle. What matters is having something to click, see and feel, and the details get figured out along the way. Variety compounds differently to mastery. Energy drinks at Red Bull, meditation at Headspace, then social media at Instagram: every jump was chosen on purpose. He calls himself a Swiss Army knife of design, and reckons it is what makes someone stand out in an interview panel. Learning the tools on the company dime is a real benefit worth asking for. DoorDash set a goal it called Design 2028, where every designer becomes a design engineer, and gave people the time and space to get there. Frank admits he complained about it at first and is now grateful he did not have to learn it in his own hours. A lot of people working hard on something is a reason for empathy, and the metric still decides whether it worked. Frank got pushback from a former colleague for publicly critiquing a feature. His read: designers who want to talk business have to be able to talk about failures like adults, because there is not a lot of emotion in business. Do the next project’s pre-work before you get to it. Frank queues up a foundation on the Tuesday so that Monday’s project starts from something rather than a blank page. Half of it may be throwaway, and a PM running agents can now come back in a morning with what used to take a two-week sprint. Solve one of your own problems and it becomes the standout project in your portfolio. For anyone in an organisation with no AI budget and no guidance from above, that is Frank’s route through. A grocery list app or a plugin for a digital guitar amp both count, imperfect included. Old, slightly bad technology is a good counterweight to this pace. Frank keeps a five-inch black-and-white Quasar countertop television on his desk, one channel and a radio dial. He calls it poorly designed and perfectly designed at the same time, the same pull that took people back to vinyl. Close the 100 tabs and learn one thing today. His answer to feeling overwhelmed in an AI-first company. Nobody has landed on a perfect process yet, which means the process is still yours to shape, and he found that anxiety and excitement flipped over into each other almost overnight. Chapters 0:04 — Bureau, Edenspiekermann, Headspace, Instagram, DoorDash: the run-up to Anthropic 2:18 — A day at Anthropic, where a 12-month roadmap makes no sense 4:32 — The LinkedIn post that started all this 5:03 — A quiet onboarding week and an enterprise Claude account 7:46 — “How do I share this?” A share function, then a Loom feature, in three prompts 9:14 — Shortcut keys, screen glimmers and museum wallpapers 10:18 — Accidentally rebuilding Claude Design, backwards 11:24 — Sunshine Shop, WooCommerce and total cowboy mode 12:33 — Will the Frank tool still be around i