Talk

Care made easy

Encoding taste into code. Two years of tools for shipping fast without cutting corners.

Presented by
Matthew Blode
Date
May 2026

Philosophy

You can sense carelessness.

Jony Ive.

Thesis

The bottleneck has changed.

Agents made code cheap, so taste is the bottleneck now.

Done Bear

The list is the smallest part.

Behind the list sit design foundations, developer tools and a platform. Every layer is built on open source.

The system

Introducing the Blode Stack.

  • Design

    Blode Icons, Blode UI and Style Capture.

  • Developer tools

    Agent Skills, AllMD, Commandment, DiffHub and Rubber Duck.

  • Platform

    Blode.md, Convene and Strata Sync.

Strata Sync

With the right abstractions, you get a lot for free.

An open-source sync engine inspired by Linear's. Tick off items and reorder lists instantly, even in aeroplane mode.

  • Works offline
  • No more spinners
  • Always in sync

Blode Icons

3,754

icons with the API agents already know

Same API as Lucide, so agents reach for yours instead.

Judgment agents can use

Put your judgment where the agent will find it.

  • Blode UI: UI you own

    An opinionated shadcn/ui registry of components you can customise, extend and own.

  • Style Capture: point at any UI

    A Chrome extension that gives your agent computed styles and Tailwind mappings.

  • Agent Skills

    Capture what you know and make it repeatable: 23 skills, such as /ui-design.

  • AllMD

    Turns web pages, videos, PDFs, docs and feeds into markdown.

The rest of the toolkit

Tools for reviewing and testing without the wait.

  • Blode.md

    The knowledge layer your AI runs on: GitHub auto-deploy, custom domains, MDX.

  • DiffHub

    Review your branch in cmux. Leave notes, then copy them as a prompt.

  • Spotlight Testing

    Test worktree code in a running dev server. Same Docker, same DB.

  • Commandment

    Voice to text, instantly. Bring your own OpenAI API key.

  • Rubber Duck

    Talk through your code with AI and hear the answers back.

  • Convene

    Work in progress: macOS meeting transcription.

Open source

Build to solve your own problems.

Give it away for free.

At MMW

Put our judgment where agents read it.

  1. 01

    The look lives in @mmw/brand

    Colours, Lexend, radii and the deck palettes live in one package, so no agent picks its own.

  2. 02

    Screens start from the @mmw registry

    New screens start from themed, reviewed components, not from what a generator outputs.

  3. 03

    Repeat corrections become skills

    When we correct an agent twice, the fix becomes a skill or a lint rule, not a PR comment.