I was there, with a team where we needed to ship something in about a month, hard deadline, no option to extend.
We got a brief. We analyzed it. We wrote a technical plan: risks, opportunities, open questions. These questions weren’t vague, they were super specific config values only the vendor could return.
Silence… we followed up for two or three weeks while we worked at other things. We knew it was important but we weren’t able to do more, we didn’t have direct contact with the vendor. Everything was communicated through product and management.
Just a week before the deadline, they came and asked “is it ready?”
We explained that the plan had been shared almost a month ago and that several vendor answers were still missing. Until we got a response, we couldn’t finish the decision, the implementation, or the configuration.
It was clear that neither product nor management had read the documentation we delivered.
Then the manager pulled out a large document the AI had generated for her.
It wasn’t checked against our architecture. It wasn’t aligned with what the customer actually used. It was the internet’s idea of “how to do the thing.”
No context. Confident tone. Useless.
Their homework was to read our docs and get the answers. By the deadline they were still missing them.
You could see it in the room: they didn’t get why “no” was the answer. Not only ignorance but no interest in understanding. That awkward silence after you explain something, where you don’t know if you ask again. She already believed she was right. She answered herself.
The model wasn’t wrong.
The brief — and the understanding behind it — was empty.
“We’re waiting on the vendor’s config so we can close the plan.”
“Can’t we just ask the AI for that config?”
“No — it’s private and specific to them.”
“Exactly. So ask the AI.”
“……”
What AI actually multiplies
AI makes clear systems clearer — and messy systems messier.
Same tool. Different outcome. The difference is almost never the model.
People think the failure mode is bad output.
The failure mode is plausible output that hides the fact that nobody decided what mattered.
That doc looked finished. So the room stopped arguing. That’s when the damage starts.
Flipping the coin
Recently I joined a greenfield service. The brief was good yet incomplete.
I read it. I understood the current implementation, what they wanted, their pain points, their customer’s pain points, and more than that, what they needed.
From all the information I was able to gather and was given, I built a phased action plan. Of course there were unknowns.
When I presented the plan, they corrected decisions that existed in people’s heads but not in the documents; they answered questions I couldn’t have invented. We updated the plan, we all were satisfied.
Execution started.
Taking the time to understand, and with a clear plan, things moved faster than they expected. Scoped for about a month; in roughly five days we were around 80%, validating outcomes and correcting course where needed.
Clarity shows up when people share what they don’t know and argue it through, not when they hoard it.
Clarity up front. Speed after.
“Good that you’ve been asking all this. The plan landed well — it helped us pick up things we’d left undefined.”
This isn’t new. It’s just faster.
Clarity problems showed up long before chat windows.
A standup where everyone reports activity and nobody can say what moved. Reporting tasks instead of outcomes.
A ticket titled “fix it” with a screenshot and no expected behavior. No description, reproduction steps, nothing.
A roadmap full of initiatives and empty of tradeoffs.
We used to absorb that cost slowly: meetings, rework, quiet frustration.
Now you can generate a week of work (or wrong work) from a fuzzy sentence.
The cost compounds in an afternoon.
What “behind on AI” often actually is
I keep seeing the same pattern in teams that feel behind.
They buy tools. They run demos. They write prompts.
They still can’t answer, in one sentence, what problem they’re trying to solve.
The tool becomes another place to put unfinished thinking.
What AI still can’t do
AI can generate solutions.
It cannot decide what matters.
It cannot hold the tension between “fast” and “right” when those conflict.
It cannot notice that the team is optimizing the wrong thing and “feeling productive” while doing it.
All this is still human.
The uncomfortable version
If AI keeps giving answers you don’t trust, the prompt isn’t the problem. Your clarity is.
If your team moves faster every week while outcomes stay flat, you don’t have an execution problem. You have a definition problem.
Clarity is a multiplier
Not a nice-to-have. Not a soft skill you schedule after the “real” work.
It’s the thing that decides whether every tool you add, AI included, makes you sharper or just louder.
So the question is:
“Am I clear enough for AI to actually help me?”
If the answer is no, then that isn’t a tooling gap.
That’s the work.