99 Strategic · a hands-on demo

One big ask vs. a real plan

Turns out that pickup-point comment was real — a whole shipping route got delayed, and 50 customers need to hear from you. Watch what happens when you ask for it all at once versus asking for a plan first.

Picking up the story: the carrier confirms it — a routing problem delayed 50 orders on one route, including hers. You need to reach all of them today.

1. One big ask — toggle to see the difference

Off: a single vague prompt trying to do everything at once. On: the same request, but asking for a plan before asking for the output.

off = "just handle all 50 customers" in one shot
You
Write something to send to all 50 customers whose orders got delayed.
AI's response

Nothing here calls a real AI model — it's a scripted simulation so the difference is instant and obvious. The real thing works the same way: one enormous ask tends to produce one enormous, mediocre wall of text; the same ask broken into a sequence produces something you can actually check and use at each step.

2. Now predict it yourself no peeking

You got the 4-step plan above. But you're in a hurry, so you tell it to skip straight to sending all 50 messages before the "flag who needs a personal note" step is done. What's the actual risk?
None — the plan already guarantees a good outcome no matter what order you use it in.
You lose exactly what the plan was for — the people who needed a personal note get the generic template instead.
The AI will simply refuse the request outright.
A plan only helps if you actually follow its order. Asking for the steps and then skipping straight to the end just recreates the same big blind action as before, with an unused plan sitting next to it. The value was always in the sequence itself, never in the plan merely existing.

One reply doesn't fit the template

You send the batch template to 49 people and personally follow up with the handful the plan flagged as needing extra care. One of those replies comes back fast — it's her. Same customer as every lesson so far. This time, though, it's good news.

The part that generalizes past this one example

"Break it into steps" is the same idea as every lesson before this one, just applied to size instead of specificity — and it happens to also fix shipping delays. A vague prompt (lesson 1) got vague output. No standing rules (lesson 2) meant retyping the same instructions. No real data (lesson 3) meant guessing. No memory (lesson 4) meant forgetting. No tool use (lesson 5) meant inventing facts. This one's the same pattern: asking for everything in one shot means getting one undifferentiated blob back, no matter how good the underlying tool is.

Giving it a structure to work inside is always the actual fix — a specific prompt, a rules file, real data, real memory, a real tool, or, here, a real plan. "The AI needs to be smarter" was never it. Same lesson, six times over.

Next: Bringing It All Together →