Nobody's role will look the same in a year, including whoever wrote this. What a designer, a researcher or a strategist actually does shifts as the agents make more of the work, so this is us teaching each other, without pretending anyone knows where it ends up. The work our clients get and the careers we build here both depend on it.
Learning
AI has left us with more questions than answers, so we unlearn and relearn together.
The level everyone at Made by Many reaches, whatever their discipline.
5 lessons
Five short lessons, an afternoon in all. The four habits everyone at the studio holds, the handful of words that make the rest of the programme legible, and one thing plugged in so the words have something to point at.
Most bad output is a bad brief: state the goal, the constraints and the audience.
A plausible answer and a correct one read the same, so the check is against a source, never against confidence.
Some material never touches a model, and anything machine-made is said so before anyone asks.
A model, a harness, an agent and a skill are four different things, and the difference is what each one can see and do.
A connector or an MCP gives an agent a tool, and the setup is a sign-in rather than a build.
Set up your machine and learn the parts every project is made of: version control, pull requests, packages, and what a design system looks like in code.
6 lessons
Six lessons, about a day. Nothing here is design yet. It is the ground every project stands on, learned by doing each thing once, so that when the agent does it for you in the next course you can see what just happened.
Claude Code, a terminal and the playbook skills, installed once.
Git is checkpoints: every change is recorded and undo is always available.
A pull request is how a change is proposed, looked at and merged, and it is the unit everything else moves in.
Most of a project is other people's code, pulled in by name and pinned by version.
A framework gives a project pages, routes and components, and changing one component changes every screen that uses it.
Colour and spacing come from tokens rather than typed-in values, which is what a design system is once it is code.
Design and build a real thing with an agent: brief it, set up the design system, explore the options, and make the components and screens.
10 lessons
Ten lessons, and the course most people are here for. Designing and making are one activity in it: you start a project, give it a design system, explore what it could be, and make it. Nothing deploys and nothing has a backend. It is still a real thing.
A real project, scaffolded by the suite and running in your browser inside an hour.
A design prompt says what the thing is for, who it is for, and what it must not be.
From a brand colour to a full token set, so everything made after this is on-brand by construction.
Refusing the first plausible direction is the skill: ask for several distinct options and name the axis each should vary on.
One component, with its three states, real content and a visible focus ring.
Compose components into a screen, then a second screen that shares them.
Taste stays in charge: a generated screen is a starting point you direct until it is yours.
A model flatters by default, so critique has to be asked for in a way that cannot be.
Contrast lives in the tokens and the keyboard pass runs on the working prototype, so accessibility is checked during rather than after.
What we generate, what we do not, and how every asset records where it came from.
Take the thing you made and make it work in the world: deploy it, give it memory, connect it to real tools, put a model inside it, and own it in production.
8 lessons
Eight lessons that turn the thing from the last course into something other people can use. Each one adds a capability the last did not have, in the order the questions come up: where is it, what does it remember, what is it connected to, what does it do on its own.
A push becomes a build becomes a URL someone else can open.
A page that has to remember something needs somewhere to put it, and that is what a database is.
A CMS or an API gives the thing real content, and the key that unlocks it never goes in the code.
A model behind a feature of the thing itself, reached through the Vercel AI Gateway.
Scheduled jobs and serverless functions do work while nobody is watching, and nobody looks after a server.
The audits say what is wrong with the thing before a person does.
Nobody hands it over: the person who made it promotes it and owns what is live.
A skill is a way of working written down where the machine can follow it, and you now have one.
The practical guide to turning an intent into a spec with an agent, and keeping that spec alive while the build corrects it.
2 lessons
Two lessons for now. The rest wait until the way of planning has settled enough to teach.
An intent becomes a spec covering the problem, the scope, the riskiest assumptions and the success criteria.
A planning session is a working conversation that ends with a drafted spec, not a meeting that ends with notes.
The practical guide to using AI in research without mistaking a plausible account for something anyone said.
No lessons yet