Speak Software

Programming, at the newest
abstraction level: yours.

Twelve hands-on labs where you describe, an agent builds, and the durable human skills — specification, verification, decomposition, taste — are the curriculum. Zero prior programming assumed. Zero apologies for that.

Start Lab 1 → Read the 10-minute essay Play the toys

The frame

You've been handed an engineering department of one — a brilliant, tireless, overconfident intern. It types faster than any human alive, never gets bored, and will confidently hand you the wrong thing with a straight face. These labs teach you to run it: every lab is a client brief, every lab kills one specific misconception, and every lab leaves evidence in a GitHub repo you own. After twelve, your GitHub profile is the portfolio.

Spec → Delta → Critique

The evidence of learning is never a chat transcript. It's what you asked for (the spec), what you overrode and why (the decision log + commits), and what you'd still change (the critique) — the three artifacts that can't be autonomous.

The Replication Test

Hand your spec — alone — to a fresh agent in an empty folder. If it reproduces a product you'd accept, you authored a program in English, and a stranger just executed it. If not, the gap is a precise map of what you failed to say.

The kata

Build it (messily, conversationally). Spec it (fresh folder, Replication Test until it converges). Critique it (would you charge money?). Done when round two converges and round three has opinions you'd defend out loud.

The twelve labs

Act I — wonder & disillusionment

01Hello, Department of One Kills: “programming isn't for me” — one conversation to a live URL on your phone. start heretoy: next-word 02The Demo Illusion Kills: “it works = it's done” — hand it to a human, collect three failures. continues 01toy: demo-illusion 03Say What You Mean Kills: “the AI knows what I meant” — three agents, one brief, three products. standalone tastertoy: three-interpretations

Act II — steering

04The Checklist It Can't Charm Kills: “checking is re-reading” — plain-language checks, automated, then sabotaged. continues 03 05The Flight Recorder Kills: “one wrong move ruins it” — wreck everything on a branch, recover in one command. continues 04 06Teach the Intern Kills: “every chat starts from zero” — the folder is the employee handbook. Prove it. continues 05toy: context-porthole 07Trust, Then Verify Kills: “it sounds right so it is right” — an immaculate paper trail, and a green tick that lies. standalonerepo includedtoy: confidence-dial

Act III — judgement

08The Taste Test Kills: “best is objective” — two working versions, one budget, your reasons in writing. standalonerepos included 09The Last 20% Kills: “the hard part is the first 80%” — polish a near-good app, with receipts. standalonerepo includedtoy: tiny-tweak 10Run the Department Kills: “big things are built in one go” — a paragraph becomes issues; the board drains. continues 09

Act IV — ownership

11The Audit Kills: “the code is the only record” — inherit a stranger's repo; the trail speaks across time. standalonerepo included 12Ship Something True Kills: “I need permission to ship” — your itch, the whole kit, a URL in someone's hand. the capstone

What the labs are, practically

The labs are documents, not software. Each one is a client-framed brief you read on GitHub and carry out in your own repo with your own agent. Nothing to install from us — no lab runner, no environment, no Docker. One-time setup lives inside Lab 1 (a GitHub account, a CLI agent such as pi, model access — honestly billed as the worst half hour of the series). Four labs ship a small product folder to work on (handoff-tiptop, brewcrew-a/b, bookable, stallboard) — every one a single HTML file plus its paper trail, no dependencies.

  1. Get the labs: read them right here on GitHub, or grab everything once — Code → Download ZIP (or clone). The provided folders come with it.
  2. Do them in order — the arc is designed — or jump in: the standalone labs (3, 7, 8, 9, 11) need nothing before them; the continues labs build on a product you made earlier.
  3. Everything you produce lands in your repos, not ours. A lab costs roughly a coffee in model credits; the essay's What it costs section is honest about it.

Environments (Codespaces, Replit, editors), model choices (local vs frontier, subscription vs API), and a curated shelf of other people's interactive explainers: RESOURCES.md.

The toys — one misconception each, ninety seconds, no install

Next Word“it thinks like me” — watch the dice fall, word by word Demo Illusion“the demo is the product” — until real users arrive Three Interpretations“the AI knows what I meant” — three valid products say otherwise Context Porthole“it sees what I see” — it never heard you Confidence Dial“fluent = true” — grade eight claims, meet your score Tiny Tweak“small change, small work” — the inside decides

Going deeper

The labs deliberately live above any programming language. When you want the layer underneath, the same author's book series at books.borck.education is the reference shelf: Conversation, Not Delegation for the thinking-with-AI framework behind the whole series, Think Python, Direct AI when you want the computational-thinking spine with a real language under your fingers, and Ship Python, Orchestrate AI when Labs 4, 5 and 10 make you curious what checks, version control and delivery look like at professional depth.