← All episodes Podcast

Ep #24: Alex Smilansky (Mayku): Desktop Factories, a Company Brain, and Building the Right Stuff

Jonny ·
Also listen onSpotifyApple PodcastsAmazon MusicDeezerYouTube

Tom and Jonny sit down with Alex Smilansky, co-founder and CEO of Mayku, the company shrinking factory machines down to the desktop. They hear how a Kickstarter that nearly flopped sold 1,300 machines in a month, unpack the Mayku Brain and the agents that now build its website, and dig into why building the right thing matters more than ever.

Key takeaways

  1. Mayku shrinks factory machines down to the desktop, as the desktop computer once did for computing. At Mint Digital, Alex watched ideas reach real users weeks after someone had them on a Monday, while getting a physical product into production could take close to a million pounds and 18 months. Mayku began as Ben Redford’s Goldsmiths project on putting the means of production in normal people’s hands.

  2. Treat Kickstarter pre-orders as a debt. Mayku aimed for 50 grand, about 100 machines, and sold 1,300 in a month: roughly 600 grand in the bank, and machines owed to 72 countries, including an island in the middle of the Atlantic. ‘On your balance sheet, Kickstarter sales are a liability,’ Alex says, and when people ask him whether they should run one, he asks whether they have any other option.

  3. A Kickstarter can turn on a single share. The campaign funded within an hour but got no press, and the team went out for sushi to commiserate. The next morning Golem13, a niche French geek page on Facebook, shared the campaign video, in which someone carves a potato into the shape of a cactus pot and makes 100 of them at the kitchen table. It reached 65 million views and brought in around 300 grand of orders that day.

  4. Follow the customers who need the machine most. One tiny business (two people and two parrots) was making 80 grand a month selling World of Warcraft bath bombs from four FormBoxes, but researchers making microfluidic chips and Dyson’s prototyping team pointed to a bigger market. Mayku built the Multiplier, a pressure former that works to sub-micron precision on sheets up to 8mm thick. It now makes truck parts, thousands of microfluidic chips a month, and clear aligners in about 10% of the UK labs that produce them.

  5. A company brain can be a GitHub repo, a Claude plugin and a Kanban board. The Mayku Brain pushes shared skills to everyone through a plugin. Suggested changes land on a board in Coda as proposed improvements, and once Alex approves one, an agent opens a pull request. Mayku Data, an MCP over a database refreshed hourly with machine usage, sales and web traffic, keeps the context current.

  6. Start with the repetitive admin the team already does. A sales quote, with an on-brand PDF at the latest pricing and a draft email, has gone from 20 minutes to one. Packaging labels for a new product now come from a skill. The first improvement from the team came from Ben: stop CC’ing Alex on quotes Ben has sent himself. Alex wants these to feel like the team’s tools, so contributing has to be easy.

  7. A relay of agents can run a website for a team with no in-house designers or developers. A CMO agent holds Mayku’s messaging for each vertical (a dental lab hears something different from a design studio) and writes the page spec. A design agent works from the design system in Claude Design, a CTO agent writes the build plan, and a cheaper model deploys to staging and production, with a human check along the way.

  8. Many agents can live in one well-organised folder. After trying Paperclip and cloud deployments, the setup that has worked best for Alex is Claude Code and a carefully structured folder, following Jake Van Clief’s Interpretable Context Methodology (ICM). A map sits at the top, each agent gets a workspace with its own skills and outputs, and requests are routed to the right one. Because it’s all files, it versions in Git like anything else.

  9. Storytelling is the fundamental AI can’t supply. AI ‘could be much cleverer than you, but it can’t know the story you wanna tell’. Alex uses it to spend less time on production and more on the outcome he wants, and to see many versions quickly so he can find out what he wants.

  10. Keep breakthrough ideas on paper for longer. A design-school tutor warned Alex never to put a rendering next to a Post-it sketch, because the polished one looks more evolved without being better. A minute of Claude can crystallise a direction that is then hard to change. For a clear deck, Alex scribbles the story and iterates with Claude; for a problem nobody has solved yet, pen and paper can be the more productive place to stay.

  11. Cheaper production raises the stakes on building the right thing. Alex recalls an Ezra Klein piece on why the internet brought no measurable productivity boom: email saved time sending mail and cost it reading more of it. Teams under pressure can produce more of the wrong stuff, so his question is the fastest, cheapest way to build the right thing.

  12. After AGI, Alex will be baking, cooking and making things out of wood. The product he wishes he’d made is anything by Teenage Engineering, the studio behind the OP-1, which also designed Mayku’s Multiplier.

Chapters

  • 0:08 — Welcome, and meeting Alex at Mint Digital more than 10 years ago
  • 1:14 — What Mayku does: factory machines for the desktop
  • 2:06 — A free prototype from Wales, and 1,300 machines on Kickstarter
  • 3:09 — From World of Warcraft bath bombs to industrial customers
  • 4:37 — How the FormBox works, Henry Hoover included
  • 5:41 — Microfluidic chips and the lab on a chip
  • 6:47 — Dyson, the Multiplier, and clear aligners
  • 8:45 — Mint Digital, and London’s Internet of Things moment
  • 10:30 — From design engineering to a digital studio
  • 11:32 — StickyGram, and an idea in production within weeks
  • 12:22 — Why a physical product takes 18 months and close to a million pounds
  • 13:45 — Launch day on Kickstarter, and a sushi dinner to commiserate
  • 16:07 — Golem13, and the video that hit 65 million views
  • 17:48 — Why Kickstarter sales are a liability
  • 18:57 — Jonny on Matt Webb’s Poem/1, and suffering the problem you set out to solve
  • 20:08 — Fast-forward to AI
  • 20:54 — Sceptical of AI note takers
  • 21:22 — December 2025, and building the Mayku Brain
  • 24:20 — Mayku Data, and context that stays fresh
  • 25:08 — Tom on the AI x Design meetups and Alex’s isometric illustrations
  • 26:19 — Quotes in a minute, labels from a skill, and Ben’s first improvement
  • 28:31 — A small, partner-based team
  • 29:11 — Why the website was always a template
  • 30:29 — CMO, design and CTO agents that ship new pages
  • 31:48 — A landing page for rubber duck makers
  • 32:25 — Agent harnesses, Paperclip, and fundamentals that stay the same
  • 33:45 — Jake Van Clief’s ICM: many agents in one folder
  • 35:45 — Jonny on OpenClaw, and Markdown all the way down
  • 36:26 — Fundamentals versus experiments
  • 37:20 — Storytelling as the fundamental AI can’t supply
  • 39:28 — Tom on letting Claude sequence his slides
  • 41:23 — Asking AI for more, and the Post-it next to the render
  • 44:12 — Jonny on half-baked work that shipped because it looked ready
  • 47:27 — The fastest, cheapest way to build the right thing
  • 48:22 — Ezra Klein, email, and the missing productivity boom
  • 51:22 — Quick-fire: life after AGI
  • 52:20 — Quick-fire: the product Alex wishes he’d made
  • 53:44 — Where to find Alex

Mentioned in the episode

Alex Smilansky

Mayku

Companies and places

People

Tools and products

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 hello@mostlyworking.co, and find everything at mostlyworking.co.