Your Company's AI Problem Isn't the AI
A board meeting. A mandate: “We’re an AI company now.”
Six months later, nothing has shipped. The tools are installed. The subscriptions are paid. But the outputs are wrong, the prompts are rambling, and nobody can agree on what they were trying to build.
This is not an AI problem. This is a specification problem.
Daniel Miessler has consulted for hundreds of the world’s largest companies. His conclusion is blunt: the companies winning at AI already know what they want. The ones struggling can’t describe themselves — to AI, to each other, to anyone.
The bottleneck isn’t your model. It’s your spec.
Here are 5 signs of that problem — and how to fix it.
1. The 8-Question Test
Miessler runs a simple test. Can your company quickly answer these?
What problem do you solve for customers? Why does your solution beat alternatives? What are your company goals? What metrics track those goals? What’s blocking you from reaching them? What strategies are you pursuing? What projects are running? Who owns them and at what cost?
Try it. Ask your team today.
A company that knows what it’s doing gives roughly the same answers to these questions across different quarters and years. A struggling company gives different answers every quarter — not because the business changed, but because no one ever agreed on them. People write down whatever makes them look good. The process starts over.
AI cannot help these companies. In fact, it makes things worse. It helps people flail more impressively, with more backflips and better-formatted slides.
AI is an amplifier. Feed it clarity and it compounds clarity. Feed it chaos and it compounds chaos — faster.
2. The Prompt Trap
Most people hear “get better at AI” and go straight to prompts. Better phrasing. More context. Smarter system instructions.
That’s the wrong starting point.
Amazon built a practice around requiring every new product initiative to start with a “working backwards” document — a press release written as if the product already existed and shipped. Not because Amazon was a writing company. Because you cannot build something you cannot describe.
The spec comes before the prompt. Always.
Before you write a single word to an AI, you should be able to answer four things: What does success look like, specifically — a signal a third party could verify? Whose problem is this — name a person, not a department? What are we deliberately not asking for — at least one explicit exclusion? What constraints matter — format, tone, audience?
If you can’t answer those four things, the prompt is not your problem.
The spec is the prompt. The prompt is just transmission.
3. What a Good Spec Looks Like
Here is the same request, twice.
Bad: “Summarize our Q3 strategy for the all-hands.”
Better: “Write a 5-minute spoken script for our all-hands. Audience: 200 employees who know the company but not the detail behind each initiative. Goal: they leave understanding 3 priorities and the logic behind each. Not including: budget, headcount, or anything not yet announced. Constraint: direct and human in tone, not slide-deck language.”
The second version isn’t longer because the AI needs hand-holding. It’s longer because the writer finally knows what they want.
This is the clarify step — a pre-prompt ritual that forces you to move from “I want something about X” to “here is the observable outcome, the owner, the explicit exclusion, the constraint.”
Most people skip it. Then they blame the AI.
The frustration people have with AI not doing what they want is usually them not being able to describe what they want.
4. The Organizational Layer
Here’s where it gets harder.
The clarify step isn’t just a personal productivity habit. It’s a measure of organizational self-knowledge.
A company that can walk into a room and answer the 8 questions clearly — consistently, across teams, across quarters — has already done the hard work that makes AI useful. They’ve compressed ambiguity. They’ve forced agreement on what they’re actually trying to do.
A company that can’t do this can’t brief a new hire well either. Or a contractor. Or a consultant. AI just makes the gap visible faster — and more expensive.
Paul Graham: “Writing doesn’t just communicate ideas; it generates them.” The inverse follows: fuzzy writing is a proxy for unresolved strategy. The prompting problem is an organizational communication problem in disguise.
Miessler puts it directly: “You can’t optimize what you don’t understand. And it’s foolish to scale something that you shouldn’t be doing in the first place.”
If you can’t describe your work clearly to an AI, ask whether you can describe it clearly to yourself.
5. The Asymmetric Advantage
Miessler’s sharpest point isn’t about AI capability. It’s about competitive structure.
Large enterprises have operated for decades without specification clarity. They survived because they were big enough to absorb the waste — and because their competition was just as fuzzy.
That’s changing.
A 5-person company with a tight spec and access to current AI can now build and ship what a 500-person company spent a quarter trying to define. The gap isn’t intelligence. It’s specification clarity.
Stripe’s value proposition was specific enough to fit one sentence. Bloomberg described it as: “all a startup had to do was add seven lines of code to its site to handle payments — what once took weeks was now a cut-and-paste job.” That sentence is load-bearing. Every product decision, every brief handed to a new hire, every prompt given to an AI — flows from it.
Most large companies do not have an equivalent sentence. That used to be fine.
AI widens the gap between the clear and the chaotic. Size no longer protects the chaotic.
Miessler ends with a cold line: “AI is a remarkably small part of who will win or lose in this initial phase.”
He’s right.
The game right now isn’t about models. It’s about who can describe what they want with enough precision that a capable tool — AI or otherwise — can act without guessing.
The companies that will win are not the ones with the biggest AI budgets. They’re the ones who answer the 8 questions. The ones who clarify before they prompt. The ones who’ve made specification a habit.
Reminder to self: before you fix your prompts, fix your specs.