Chaining: Big Projects, Small Prompts
10 min read
Somewhere around your third good prompt, ambition kicks in. If AI can write one email, why not the whole launch? So you type the mega-prompt — "create my complete spring launch plan with packages, pricing, emails, and social posts" — and get back twelve pages that look finished and help nobody.
Why the giant prompt fails
Two reasons. First, the model spreads its attention across everything you asked for at once, so every section comes out shallow — the average answer problem from chapter 1, multiplied by every section you crammed in. Second, and worse: you get no checkpoints. If the pricing logic in section two is off, every email and post built on top of it inherits the mistake, and you discover that at the end. Or your customers do.
The fix is chaining: running the big job as a sequence of small prompts, where each one builds on output you've already reviewed and approved. Outline first. Approve it. Then draft piece by piece, with your judgment between every step.
Maya's spring launch, as a chain
Maya wants a launch plan for spring services: cleanup packages, prices, an announcement email, social posts. One project, four prompts, all in the same chat.
Prompt 1 — Research questions:
I own a 14-person landscaping company in Calgary and I'm planning
our spring services launch. Before we build anything, ask me the
10 questions you'd need answered to do this well — about my
customers, my crew capacity, and last year's numbers.
Prompt 2 — Offer outline (after she answers):
Based on my answers above, outline 3 spring cleanup packages: a
name, what's included, and who each one is for. No prices yet.
Prompt 3 — Pricing table (after she edits and approves the packages):
Here are the final packages: [PASTE APPROVED OUTLINE]. Last year
our average cleanup invoice was $480 and each crew did 4 jobs a
day. Build a pricing table with a one-line rationale per price,
and flag any price you think the market will push back on.
Prompt 4 — The pieces (after she approves the pricing):
Using the approved packages and prices above, write the
announcement email to past customers (under 200 words) and 5
social post options I can choose from.Notice that prompt 1 produces nothing. It asks. Making the AI interview you first is the cheapest quality upgrade in the whole chain, because it pulls out context you'd never think to volunteer. From there, every prompt stands on ground Maya has already approved — and each one stays small enough to hold to the standard of The Five-Part Prompt.
Between steps two and three, she killed a package that would have wrecked her crew schedule in May. Thirty seconds at a checkpoint, versus rewriting an entire launch after the emails went out. That's what the checkpoints are for.
You've already built one of these
If this structure feels familiar, it should. It's the same logic as the rain-day decision tree Maya built back in Operations: From Chaos to Checklists: take one big judgment call and break it into small, explicit steps with a decision point between each. The decision tree does it so an employee doesn't have to call you. The chain does it so the AI doesn't have to guess. Same skill, different target — and founders already have it, because delegating a big project to a person works exactly the same way.
When to chain, when not to
Not everything deserves the ceremony. A single email, a review reply, a job posting — that's one prompt and a follow-up or two. Chain when the deliverable has parts that depend on each other: a launch plan, a hiring process, a new service line, a price increase with all its supporting messages. Here's the tell: if a mistake in one section would force you to redo another section, chain it.
Try it now
Choose the project you've been avoiding because it's too big to start. Don't ask the AI to do it. Ask it for the 10 questions it would need answered before it could do it well, then answer them. That's step one of a chain — and the project officially isn't unstarted anymore.