(this isn’t a play on words with Dr. Sheldon Cooper’s Fun With Flags)
One of my absolute favourite things to do when meeting a new friend is to sit down and talk nonsense with them and listen to their opinions on some of the industry’s hot button topics. Little did I know when I met Roger (a fellow NTC I hadn’t met until today!) he’d level up my AI knowledge from his experience on how I’m doing it wrong 🙂
Alas, it was not a sit down over a coffee though, Julien had had a horrible 3 days of travelling with cancelled flights, missed connections and now luggage still in Germany so we embarked on a dual purpose walk to a Target store. Primarily to get him some essentials until his luggage arrives, but also to keep us awake from dark o’clock traveling.
So off we went on a walk (more like a trek) – Julien, Roger, and I towards a Target on Google Maps and initially started with my usual baseline on politics (quickly put to bed), but then swiftly onto AI specifically the ethics, and how Roger uses it in his day to day life.
I was stunned to be honest, as someone who does like to treat CoPilot like a posh search engine to find information I know I’ve seen before, and to sense check ideas.
Roger and Julien took it to a whole new level with how they use ‘Reverse Prompting‘ to improve accuracy and quality of the end goal, and using ‘spec-driven development‘ to allow rapid iteration of design specifications without reinventing the wheel with a full regeneration of the end result.
An hour of walking through the streets of San Jose listening to these two guys speaking with such passion about how they do so much more than just use it as a posh search engine left me with some new tricks to try in my life.
(Thanks CoPilot) – Reverse prompting is an AI technique where, instead of starting with carefully engineered instructions, you give the AI an example of the outcome you want and ask it to work backwards to identify the prompt, requirements, structure, or process needed to reproduce it.
For executives, its value is that it shifts AI use from “learning how to write perfect prompts” to showing AI what good looks like.
For example, you might provide a strong board report and ask, “What prompt would consistently produce an executive summary in this style, with the same level of detail and decision focus?”
The AI can then derive a reusable prompt or template.
This makes reverse prompting particularly useful for standardising reports, presentations, communications and analysis across an organisation, while still requiring human review to ensure the AI has correctly understood what made the original output effective.
(Thanks again CoPilot) – Spec-based development with AI is an approach where you define what the software should do, how it should behave, and what success looks like before asking AI to change the code.
Rather than simply telling the AI “change this colour”, you provide a small specification with clear requirements and acceptance criteria.
For example, instead of “change the button from blue to green”, the specification might state: “Change the primary action button from blue (#0078D4) to green (#107C10). Apply the change to all primary buttons, preserve hover and disabled states, make no other visual or functional changes, and verify that the resulting colour meets our accessibility requirements.”
The AI can then inspect the application, determine which code or styles need changing, implement the change and check its work against the specification.
For executives, the important distinction is that the specification becomes the contract with the AI: people define the required outcome and constraints, while AI handles more of the implementation, making AI-assisted development more controlled, repeatable and auditable than relying on conversational prompts alone.
Why this wording works: it keeps the concept executive-level while using the blue-to-green change to make the difference between an informal prompt and a measurable specification immediately tangible.
Hearing both of these approaches, I could immediately see where I’m wasting tokens (token anxiety is at thing!) reinventing the wheel for simple changes, and getting angry because the AI tool didn’t deliver the outcome I wanted because my prompt wasn’t explicit enough, or didn’t provide all of the necessary inputs to obtain a good outcome.
I’m looking forward to further discussions in the coming days.