Explore
An explorable is one picture that grows a piece per step. At each step you make a guess before the answer appears, then see what can go wrong at that point and how teams handle it. Each one takes about 10 minutes and ends with a short list of things you can say in an interview, plus the words you picked up on the way.
01
How an AI agent works, one piece at a time
Build an AI agent one piece at a time, from a single model call to two agents sharing work, and see the new problem each piece creates and how teams fix it.
About 10 minutes
02
What one AI request costs, and why it's slow
Follow one request from a user to a model and back with a meter running, and see how tokens, context, model size, reasoning, caching and routing move the cost and the wait.
About 10 minutes
03
How agents fail, and how to break one
See how an AI support agent fails loudly and silently, label each failure by where it started, catch it with checks, then attack the agent with prepared attack cards and watch which layer stops each one.
About 10 minutes
04
How software gets built with AI
Walk one change from a task to production with a coding agent: the spec, the sandbox, tests and CI, review, the merge and what happens after, and see where people still decide.
About 10 minutes
05
How AI app builders like Lovable work
Follow one app from a prompt to paying customers in an AI app builder: the preview, a database, login, payments, publishing, edits by chat and credits, and see who runs each piece and which delivery gates go missing.
About 10 minutes
06
How evals work, and when to trust the scores
Turn a real failure into a test case, pick a grader for each check, check the model judge against an expert's labels, set a release gate, try a slice of real traffic, and keep the scores honest when the model or the judge changes.
About 10 minutes
07
How sandboxes contain an agent
Zoom into a coding agent at work and build a box around it, one control per step: the sandbox, the secrets inside, the network out, permission rules and undo, and see what each one stops and what it lets through.
About 10 minutes
08
How an AI assistant works, acting for you across your accounts
Follow an AI assistant acting for one person: what it remembers, what a consent screen grants, what it asks before acting, runs that start while she's away, an email that steers it and how its access ends.
About 10 minutes