The Deck and the Gap Analysis
24 Jul 2026Our AI Steering Committee met this week and, among other things, decided we should stop using Claude to build internal-facing slide decks. If the audience is internal, make it a memo instead.
The same week, I pointed one of our agents at a VP’s roadmap, had it cross-reference our meeting notes, our project tracker, and Slack, and it handed back three workstreams that had slipped off the list. One of them was flagged company-wide as one of the biggest blockers to our onboarding numbers. I checked the gaps. They were real, so I sent them over.
I don’t think those two facts contradict each other. I think they rhyme, in a way that should bother anyone trying to govern AI use from a dashboard.
Why the committee pulled back
AI spend is up 551% month over month, and the return on it is still theoretical. Given a cost curve going vertical and no ROI story at scale yet, pulling back on the lowest-value use is the reasonable move. A slide deck someone skims for ninety seconds is a strange place to watch that kind of increase land. The committee saw a use case producing something disposable and cut it. On those terms, I’d have voted for it too.
This isn’t abstract for me. One of my goals this quarter is delivering ROI-positive outcomes from AI through training and tool rollout, which puts me on the hook to know which uses actually pay. A cut made by artifact type doesn’t answer that, even when the specific call is a reasonable one.
The wrong axis
The decision measured value by the artifact. Internal presentations are low-value output, so cut the tool that makes them. But the artifact was never where the value lived or didn’t. A deck generated in ten seconds saves you nothing when it took no thought to make. The same model, aimed at the roadmap, did a piece of thinking I couldn’t have done by hand that week and wouldn’t have found the time to try.
Same tool, opposite value, and the difference has nothing to do with whether the output is a deck or a memo. The line that actually separates the two uses is whether you’re using the tool to skip the thinking or to do more of it than you had time for.
A deck no one had to think to make is the first kind. A gap analysis across five sources you’d never hold in your head at once is the second. The format they arrive in is the wrong thing to watch: a memo can be exactly as thoughtless as a deck, and the analysis is worth the same whether you present it or write it up. Govern the format and you’ve touched none of what made the spend worth paying.
The gap I couldn’t have found by hand
The roadmap lived in one doc. The real status of each item lived somewhere else: standup notes, the tracker, a dozen Slack threads. Cross-referencing all of that is exactly the work that never gets done, because holding five sources in your head for an afternoon is the part there’s never room for. The agent read all of them and answered one question: what’s on this roadmap that stopped moving? Three answers came back. One was an onboarding blocker the whole company was already worried about.
If you want to know whether a use is worth the spend, look at whether it’s producing polish or recall. Polish is the deck, the reformatted memo, the meeting summary that reads nicely and changed no decision. Recall is the tool holding more context than you can and handing back the piece you’d have missed. The value was never in the output looking finished; it was in surfacing something true you didn’t have the bandwidth to reach. Aim the tools at that.
The enforcement problem
The objection to my own line is the committee’s best defense: you can’t police intent. “No internal slide decks” is a rule you can write down and check. “Use it to think, not to skip thinking” is a poster, not a policy. A committee can govern surfaces because surfaces are visible; it can’t sit at each person’s keyboard and score whether the thinking was real.
That’s true, and it’s why the line can’t be a committee rule at all. It has to be the question each person asks before they reach for the tool: would this have happened without me, or am I doing something I couldn’t do alone? A format rule is enforceable but blind. It can’t tell a thoughtless deck from one that presents real work, and it says nothing about the thoughtless memo that takes the deck’s place. The committee governed the one thing it can see from where it sits. The distinction that actually decides value is only visible at the keyboard, in the moment someone chooses what to ask the tool for.
What the spend chart can’t see
The 551% on the committee’s dashboard is one number, and it can’t tell the two uses apart. It would read identically if every dollar went to disposable decks or if every dollar went to work no one had time to do otherwise. The metric that triggered the whole decision is blind to the only distinction that decides whether the money was well spent.
The part a policy can’t reach
That judgment can’t come from the dashboard, and a policy can’t supply it either. It comes down to whether the person reaching for the tool can tell the difference between skipping their thinking and extending it. That difference can be taught, and teaching it is most of what I think AI enablement actually is: not handing people a tool and a target, but building the judgment to point it at the work that only happens because the tool held more than they could.
It’s also the harder job, and the one the spend chart will eventually be measured against. A ban on decks moves a number down. Getting a team to aim these tools at the gap analysis there was never time for is what moves the return from theoretical to real, and no policy can do that part for them. It comes from people who’ve been shown the difference and made responsible for it.
The committee can take the decks away. It can’t make anyone think. That part belongs to the person who opens the tool, every time they do.