By Barry Buck – Most people prompt AI like they’re writing a Jira ticket. Specs, requirements, acceptance criteria. The output matches the input – competent, narrow, exactly what was asked for and nothing more. Then they wonder why their AI experiments feel underwhelming.
The answer is uncomfortable: the model isn’t the bottleneck. The brief is.
Late one night last week, working on a rebuild of our flagship platform, I caught myself about to write another tidy instructional prompt. Build me X with these specs. Standard fare.
Instead I stopped and wrote what I now call a manifesto – not instructions, but a worldview.
I asked a definitional question (“what is a project, really?”), I answered it with examples that contradicted each other a little (“it’s a git repo, but also AI context, but also documentation”), and then I closed with the lived workflow I’d never written down for anyone, AI or otherwise.
The response broke my product roadmap. In the best possible way. Within minutes the model had reflected my own thinking back to me in cleaner language than I’d used to express it, named a product category I’d been circling for months, and proposed a sequencing change to the roadmap that was correct on first read.
That isn’t a feature you can prompt-engineer your way to with a clever template. It’s what happens when you stop briefing the AI and start co-authoring with it.
Three moves, in order.
First, open with a definitional question, not a build instruction – force the model into the shape of thinking before producing.
Second, answer the question with examples that don’t quite agree with each other – that productive tension is what a list can never capture and what defines a real concept.
Third, close with the lived workflow. Not the sanitised version. The actual sequence of how you do the work, including the bits you’d never put on a slide.
Here’s why this matters for executives, not just engineers. The reason your AI output is mediocre isn’t because the model isn’t good enough. It’s because you’re prompting like you’re briefing a junior contractor instead of describing a worldview to a senior peer.
Manifesto engineering is the difference. It takes longer to write a manifesto than a spec. The output is dramatically better, every single time.
The thing you’ve been struggling to explain to AI is probably the thing you’ve never explained to yourself. Try writing it down this week. The model is waiting for you to mean it.
Barry Buck is the chief technology officer of Saucecode and Roboteur architect
www.saucecode.tech