Issue #8 - The Thinking Layer
Subscriber Edition
A prompt says: "Write me a marketing email." A framework says: "Every email has one job. The subject line is a promise. The body delivers on that promise. The CTA is the one action that matters. Permission is a relationship, not a transaction." Same AI. Completely different output.
Field Note
Last year, I ran a test. I asked AI to write a client proposal for a web redesign project. No framework. Just a clear, well-written prompt with the client details, the scope, and the desired outcome.
The result was fine. Professional tone. Correct structure. It hit every point I mentioned. Any consultant would look at it and say, "This is competent."
Then I wrote ten lines. Not a better prompt. Not more detailed instructions. Ten lines that described how an expert thinks about proposals. Principles, not procedures.
Something like: "A proposal is not a description of work. It's a demonstration of understanding. The client already knows what they want built. What they're buying is evidence that you understand why it matters. Lead with their problem, not your solution."
And: "Price anchoring happens before the number appears. If the proposal has established clear value, the price feels like a natural conclusion. If the proposal has described activities, the price feels like a negotiation."
Ten lines. No formatting requirements. No template structure. No word counts. Just principles about how expert proposal-writing actually works.
The output was unrecognizable from the first attempt. Same AI. Same client details. Same scope. But the proposal now opened with the client's actual business problem instead of the agency's credentials. The pricing section felt like a recommendation instead of a line item. The timeline framed deliverables as milestones the client would experience, not tasks the team would perform.
The prompt had told AI what to write. The ten lines told AI how to think about writing.
That's the difference between an instruction and a thinking layer.
Case Slice Teardown
Role: Senior recruiter at a mid-size technology company
Situation: Writing 8-12 job descriptions per week. Using AI to draft them, but every output read the same: generic, interchangeable, full of phrases like "fast-paced environment" and "passionate self-starter." Candidates weren't applying. The few who did couldn't articulate why they'd chosen this company over competitors.
Constraint: Couldn't spend more than 20 minutes per job post. Volume was non-negotiable. The problem wasn't time, it was that every AI-generated post felt like every other AI-generated post.
Intervention: Built a 10-line thinking layer about job description psychology. Core principles: "The job post is the first work sample of your company's communication quality. If the post is generic, candidates assume the company is generic." And: "Describe the problem the role solves, not the tasks the role performs. Engineers don't want to 'maintain systems.' They want to solve the problem that makes maintenance necessary."
Outcome: Application quality increased measurably within two weeks. Multiple candidates referenced specific lines from the job descriptions in their cover letters. The recruiter's time per post actually decreased because the thinking layer eliminated the revision cycle. The first draft was consistently close to final.
What's notable here: The recruiter didn't become a better writer. The AI didn't get an upgrade. The ten lines changed what "good" meant in the context of the task. The prompt had been asking for a job description. The framework asked for a demonstration of what it's like to work at this company. Same request. Different thinking layer. Different result.
The Deep Dive
The entire prompt engineering industry is built on one assumption: if you give AI better instructions, you get better output. More specific. More detailed. More constrained. Better results.
This is true, up to a point. And then it stops working.
A well-crafted prompt can specify format, tone, length, audience, and structure. It can include examples. It can describe what good looks like. This is instruction-level optimization, and it produces incrementally better results.
But instructions operate on a fixed plane. They tell the AI what to produce. They don't change how the AI approaches the production.
A thinking layer operates on a different plane entirely. It changes the reasoning that precedes the output. It doesn't say "include a value proposition in paragraph two." It says "the reader is comparing you to three alternatives they found this morning, and your job is to make the comparison irrelevant by reframing what they should be comparing on."
That's not an instruction. That's a way of thinking.
Consider the difference in practice:
Instruction: "Write a blog post about project management. Include tips for staying organized. Use a professional tone. 800 words."
Thinking layer: "Every project fails at transitions, not tasks. The gap between 'assigned' and 'started' is where 90% of delays hide. The most valuable project management insight is that the plan is never the bottleneck; the handoff between plan and execution is the bottleneck. Write from this premise."
The instruction produces a competent blog post that reads like a hundred other competent blog posts. The thinking layer produces something that makes readers stop and reconsider their own experience.
This is what frameworks do. They don't make AI more capable. They make AI's existing capability more precisely directed.
A 10-line framework has a specific anatomy. Not rules to follow, but a structure that consistently works:
Lines 1-2: The reframe. What does an expert see that a novice doesn't? "A budget isn't a spending plan. It's a statement of priorities made visible." This reframe changes every decision that follows.
Lines 3-5: The principles. Three operating rules that govern quality in this domain. Not best practices. Principles. The difference: a best practice says what to do. A principle says what's true regardless of what you do.
Lines 6-8: The failure modes. What does bad look like, and why does it happen? Experts don't just know what works. They know exactly how things fail. Including failure modes in a thinking layer prevents the most common errors.
Lines 9-10: The quality test. How would an expert evaluate the output? "If you removed the company name, could this proposal belong to anyone? If yes, it's not done." The quality test gives AI a self-evaluation standard.
Ten lines. Not a prompt template. A thinking layer.
Framework Quality Check
1. Does it change reasoning, or just formatting?
If your ten lines specify structure (headers, bullet points, paragraph order), you've written a template. If they change how the AI thinks about the problem before producing output, you've written a framework. Templates produce consistency. Frameworks produce insight.
2. Would an expert in this domain nod or shrug?
Read your ten lines to someone who's been doing this work for twenty years. If they say "obviously," you've captured principles that are genuinely true. If they shrug, you've written generic advice. The bar is: would this change the output of someone who already knows what they're doing?
3. Can it survive a different task?
A prompt about email marketing only works for email marketing. A thinking layer about permission-based communication works for emails, onboarding flows, push notifications, and customer support scripts. Frameworks transfer across tasks. Prompts are single-use.
4. Does it include what fails?
The most useful part of expert knowledge isn't what works. It's knowing how things break. If your framework only describes good output, it's incomplete. The failure modes are what separate a thinking layer from a wish list.
Write your first 10-line thinking layer
Pick one task you do repeatedly.
Client emails. Project estimates. Status updates. Hiring evaluations. Something you've done enough times to know the difference between adequate and excellent.
Write the reframe (2 lines).
Complete this sentence: "Most people think this task is about ___. But what it's actually about is ___." That gap between the common view and the expert view is your reframe. It's the most valuable part of the entire framework.
Write three principles (3 lines).
What's true about doing this well, regardless of the specific situation? Not tips. Truths. "The first paragraph of a status update determines whether anyone reads the rest." That's a principle.
Write the failure modes (3 lines).
How does this task go wrong? Not "it could be bad." Specifically how. "Status updates fail when they describe activities instead of outcomes. The reader doesn't care what you did. They care what changed."
Write the quality test (2 lines).
How would you evaluate the output in ten seconds? "Read the first sentence. If it's about the project, it's working. If it's about the team, start over."
You now have a 10-line framework. Test it. Give it to AI alongside your next prompt for that task. Compare the output to what you'd get without it. The difference is the value of a thinking layer.
What's one task where you know the difference between adequate and excellent?
Reply with the task and one principle you'd put in a thinking layer for it. The best ones get featured (with credit) in future issues.
A 10-line framework is where it starts. The Strategic Thinking Academy teaches the complete methodology: how to extract frameworks from any domain expertise, test them systematically, and build compound intelligence where multiple frameworks work together.
Explore Strategic Thinking Academyor
Start with the foundation
Minimum Viable Intelligence ($297)
Strategic Thinking Weekly - Published every week
Subscribe for full access to all issues
Unsubscribe anytime - Tampa, FL