Gen AI Innovation
Remote training brings its own set of challenges. Four sessions, fully online, with a team based in Kenya — no room in a room, so to speak. No body language to read, no chance to catch confusion before it settled in. We worked through it anyway. What came out the other end made the whole thing worthwhile.
The final assignment was simple on paper: take a real task from your own work, use more than one AI tool to get it done, and show your working. No hiding behind a polished output — I wanted to see the process, not just the result.
The Task
One delegate was leading a vendor evaluation for a new organisation management system. Three vendors. Three-hour technical presentations for each. Twenty evaluators sitting through all of it. And at the end of it, a decision that would shape how an organisation serving thousands of members runs its operations for years to come.
It's the kind of task that looks nothing like a training exercise and everything like real work — which is exactly the point. This is where you see whether the framework actually holds up, or whether it was just a nice idea in a classroom.
Watching her work through it, you could see all four D's mapped out clearly.
Delegation
The structural work went to AI: the first draft prompts, the formula logic, the scaffolding of the evaluation workbook itself. The parts that needed her judgement — the evaluation criteria, and the final call on what actually mattered to her organisation — stayed firmly hers.
That split is the whole point of Delegation. It's not "let AI do it," it's knowing precisely which parts of the task are yours to own and which parts are safe to hand over.
Description
Her first attempt at a prompt was thorough — maybe too thorough. She asked the AI to act as a senior enterprise architect with twenty years of experience in procurement and digital transformation.
The output that came back looked impressive. Confident. Structured. Full of credentials she'd never actually given it — the model had built that expertise for itself, filling in a gap she'd left open.
This is the part people underestimate about prompting. A vague or overly generous brief doesn't just produce vague output — sometimes it produces confident output that's quietly wrong for the job. It read like something written by an expert. It just wasn't written for her twenty non-technical evaluators.
Discernment
She checked it anyway — and that's the step that saved the whole exercise. She recognised the mismatch: the tool needed something simple enough for twenty people with no technical background to use on presentation day, and what she'd been handed was built for someone who didn't need the help in the first place.
So she stripped the role back to what it should have been from the start: technical lead. Real context. Real constraints. No invented expertise doing the thinking for her.
Discernment is the D that has nothing to do with the AI at all. It's entirely about whether you're paying attention to what you asked for versus what you actually needed.
Diligence
The second version was sharper in every way that mattered. A full workbook: instructions, a master evaluation matrix, individual scorecards for all twenty evaluators, and a consolidated results sheet that pulled every score into one ranking dashboard automatically — charts included. No formulas to fix by hand. No copying numbers between sheets on the day.
But before any of it went near the evaluation team, she went through it line by line and took ownership of every number in it. That's Diligence — not trusting the output because it looks finished, but verifying it because you're the one accountable for what it says.
What It Replaced
When she submitted the assignment, she told me: "For the first time, we didn't have to do manual consolidation after evaluating vendors."
That's the line that stayed with me. Not the workbook itself — impressive as it was — but what it replaced. Hours of pulling scores together by hand, chasing down evaluator sheets, checking totals. All of it done the moment the last evaluator hit submit.
The Bigger Point
That's the 4D Framework working in one real task, not as four separate skills practiced in isolation, but as one decision feeding into the next — Delegation setting the boundary, Description shaping the ask, Discernment catching what needed catching, Diligence closing the loop. At every stage, the human stayed in charge of the process. The AI moved fast. She made the calls.
That's what AI fluency actually looks like in practice. Not speed for its own sake — speed directed.