Sprint 1 · ~11 min · includes a hands-on task on your own work
What vibe coding & AI agents actually are in 2026
One operator can now scaffold a working app in an afternoon — if you know what you're actually wielding, and where its edges are.
The idea
Vibe coding means building software in plain language: you describe the outcome, the model writes and wires the code. That's real leverage, but it has a ceiling. Learn the ladder before you build anything. An ASSISTANT (Claude, ChatGPT) drafts, reasons, and codes with you — it needs you in the loop. An AUTOMATION (Zapier, Make, n8n) runs fixed, triggered steps the same way every time — trigger to action, no judgment. An AGENT plans its own steps, picks tools, and works toward a goal with verification — powerful, but it costs far more tokens and adds failure surface, so most ideas need an assistant or an automation, not an agent. The durable truth underneath all three: AI generates fast and confidently, but generation is cheap and correctness is not. You are the orchestrator. The skill isn't typing speed — it's judgment about what to build and how to check it.
Walkthrough
- 1Write one paragraph: the thing you want to build, exactly who it's for, and the one painful problem it kills.
- 2For each part of that idea, label which rung you actually need — assistant, automation, or agent.
- 3Circle the simplest rung that solves the core problem — default down a rung whenever you're unsure.
- 4Note one place the build will produce something you'll have to verify (data, money, a customer-facing action).
Worked example
A solo founder who runs a small bookkeeping service wants to stop manually triaging the intake forms that hit their inbox.
The prompt
I'm a solo founder. I want to build: a tool that reads new client-intake form submissions, pulls out company name, service needed, and budget, and drops a one-line summary into a Slack channel so I can triage fast. Walk me through which of these I actually need and why: (a) an AI assistant I prompt by hand, (b) a fixed automation that runs on each new form, (c) an autonomous agent. Recommend the SIMPLEST option that solves it, name a concrete tool for it, and list anything in this flow I'd still need to verify by hand.
What good output looks like
Recommended: (b) a fixed automation — the steps never change (new form -> extract 3 fields -> post to Slack), so an agent would be overkill and burn tokens for no benefit. Tool: Zapier or Make, with one AI step to extract the fields. Verify by hand: the AI extraction can mis-read budget or service type, so post it as a DRAFT summary you skim — don't let it auto-route or auto-reply to the client.
Where it burns you
- Reaching for an 'AI agent' because it sounds impressive when a 3-step automation does the job cheaper and more reliably.
- Treating a slick demo as a finished product — the gap between demo and shippable is almost entirely error-handling and the things the model silently skipped.
- Believing the leverage stats literally ('one person does the work of ten'). They're directional and come from selected, AI-augmented founders — treat them as motivation, not a promise.
Do it now
Write the one-paragraph build idea you'll carry through this whole track, and label which rung — assistant, automation, or agent — its core actually needs.
Before it touches a price — verify
- I can state my build idea, its user, and its one core problem in a paragraph.
- I picked the simplest rung (assistant / automation / agent) that actually solves it.
- I named at least one output I'll have to verify rather than trust blindly.
You should now have: A one-paragraph build idea plus a clear-eyed read on whether it needs an assistant, an automation, or a true agent.
Guardrail: Leverage claims ('one person does the work of many') are directional and drawn from selected founders — motivation, not a guarantee. Pick the simplest rung that works.