Michael Gardon · CHIEF AI Accelerator
Accelerator / Phase 1 / 1.2 Description
Phase 1 · Module 1.2

Description: scope what you actually want in detail

Good inputs, good outputs. The 4 C's are how you give the AI enough to actually get what you want.

~7 min · Phase 1 · The first D
The principle

Good inputs, good outputs

Garbage in, garbage out. It's an old line, and it's still true. The quality of what you get back is set by the quality of what you put in.

Most AI problems come down to one thing: we don't actually know how to describe what we want. We hold it clearly in our own heads and assume a few words will carry it across. They don't. The failures from the last lesson (generic, wrong length, wrong format, off tone) are almost always a thin input, not a dumb tool.

And most problems are Description problems. Even a task you hand off that flops is usually a description failure underneath: the scope was vague, or the ask was too big to do in one shot. Get good at describing and everything else gets easier.

That's why the 4 C's exist: a repeatable way to scope what you want in enough detail to actually get it.

Why it matters

This isn't prompt engineering

You're not learning prompt tricks. You're learning the semantics of working with AI: how to say what you mean clearly enough that a machine can act on it. That's the base skill for everything that comes later. Once you have it, you can:

Encode your thinking, processes, and taste into reference files, so you don't have to re-explain yourself.
Build skills, workflows, and agents that execute for you.
Learn by saving and updating your thinking, your approach, and how the work gets done, so the system keeps getting better.

Get describing right now, and every ring after this gets easier.

The method

The 4 C's

C1
Context

Who it's for, your role, the situation, what you've already tried, and the material the AI should work from. The biggest lever. More on it below.

C2
Charge

The outcome you want, and the mode you want the AI working in. Sometimes that's a deliverable: write, summarize, draft, compare. Sometimes it's a way of working: "just help me think this through," "blue sky, don't critique yet," "don't build anything, we're still scoping." Naming the mode keeps the AI from red-teaming your brainstorm, or running off to build before you're ready.

C3
Constraints

Length, tone, what to avoid, what's non-negotiable.

C4
Cut

The shape and form of the output. An email, memo, table, bullets, one paragraph. It can also be a deliverable type: a knowledge document, a process, or even a task ("capture this as a task").

The biggest C

Context is the biggest one

You are rarely short on context. You leave it in your head, or sitting in a file on your computer. The fix is to actually give it to the AI. It comes from four places:

Context what the AI works from 01 Prompt directions What you type into the chat 02 Encoded knowledge Saved files that load every time 03 XLS PPT Work reference files PowerPoint, Excel, Word, PDF 04 URL Examples Screenshots, links, samples

Before writing a long prompt, ask: what do I already have that the AI could just read? Then paste it, attach it, or point to it.

The context you supply over and over does not have to be retyped. Any time you type the same thing twice, that's a signal to save it. Encode it once into the documents and workflows the AI references on its own, and it pays off every time after. That compounding is the real unlock, and it's where we're headed.
Worked examples

See it: before and after

Example A: the delay email

Thin promptWrite an email about the project delay.

You get a generic, hollow note that could be from anyone, about any project.

The 4 C's promptContext: I'm the COO. I need to tell our largest enterprise client the integration is slipping two weeks. This is the second slip, and they've been patient.
Charge: Write the email.
Constraints: Professional but honest. Own the delay without groveling. Three short paragraphs.
Cut: A ready-to-send email with a subject line.

You get a note to that client, about that situation, in your voice, that you can almost send as-is.

Example B: the board update

Thin promptSummarize Q1 for the board.

You get the numbers restated. The board already has the numbers.

The 4 C's promptContext: Q1 update on an underperforming portfolio company. The board already knows revenue missed. What they don't know is the turnaround plan.
Charge: Write the update that frames the miss and makes the case for the plan.
Constraints: One page. Direct. No spin.
Cut: A short narrative up top, then a table of the key metrics.

The thin version restates what they already have. The 4 C's version drives the decision you actually need from them.

Map

Which C fixes which failure

When this goes wrongThe C that fixes it
GenericContext + Charge
Wrong lengthConstraints
Off toneConstraints
Wrong formatCut
Reuse

A skeleton you can steal

Paste this and fill in the blanks:

Prompt skeletonContext: I'm [role] at [kind of business]. [The situation, what I've tried.]
Charge: [What you want, and the mode: build it, draft it, or just think it through.]
Constraints: [Length, tone, what to avoid.]
Cut: [Format or deliverable type: email, doc, process, task.]
Your turn

Try it

What's next
Up next
Discernment

The 4 C's get you a good draft. Knowing whether that draft is actually right is the next lever.

Module 1.3 · Discernment →

Go deeper on prompting: Anthropic's prompt engineering guide →