Why does the pitch promise more than the build can deliver?
Short answer: incentives. Sales is paid when you sign, engineering is judged months later, and generative AI demos are uniquely good at hiding the distance between those two moments. The translator below decodes the deck; the fix is one question: which engineer will own our integration?
Gartner predicted at least 30 percent of generative AI projects abandoned after proof of concept, and calls the vendor side "agent washing": thousands claim it, about 130 are real (Gartner, 2024-2025).
Sales gets paid when you sign. Engineering gets judged in month three. AI projects break in the gap between the two.
// the sales translator
what the deck says · what it means
Run the next deck you receive through this. Every phrase below is real, common, and technically true. The question next to it is the one that separates a build from a pitch.
usually means: Often a thin layer on the same rented model your competitors use. The label says nothing about what, if anything, they built on top.
ask them: What did you build on top of the model, and which parts are yours?
usually means: The demo is deployed in weeks. The integration into your systems, your data and your workflow is the actual project.
ask them: Weeks to a demo, or weeks to production on our data?
usually means: Usually retrieval over the documents you upload. Nothing "learns" unless someone builds a feedback loop, and that is rarely in the quote.
ask them: What exactly updates when we correct a wrong answer?
usually means: SSO, a log page and a higher price tier. Evals, audit trails and data residency are the real enterprise checklist.
ask them: Show us the eval suite and the audit trail.
usually means: On their benchmark, with their data. Your documents, your formats and your edge cases were not in the test.
ask them: What was the test set, and can we run the same test on ours?
usually means: Frequently last year's automation with this year's label. Gartner counts about 130 real ones among thousands who claim it.
ask them: Show us one autonomous run, end to end, unedited.
usually means: An API exists. Someone still has to wire it into your systems and keep it wired, and that someone is usually you.
ask them: Who does the wiring, what does it cost, and who maintains it?
// the deep dive
Nobody is exactly lying. "Deployed in weeks" means the demo. "It learns your business" means retrieval over the documents you upload. "Enterprise-ready" means SSO and a log page. "99 percent accurate" means on their benchmark, not on your data. The account executive is not the person who will wire the system into your ERP, clean ten years of your files, or sit in the eval meetings, and by the time the gap surfaces the commission has cleared. AI made this old software pattern worse, because generative systems are at their most impressive in the first five minutes and at their most fragile in month three. Two habits protect you. First, run every phrase in the deck through the translator above. Second, do the thing a sales team cannot fake: ask for the engineer who will own your integration to join the next call, and ask them what will be hard. If no engineer shows up before the contract does, that is your answer.
// chapter faq
How do I pressure-test an AI vendor before signing?
Ask for a production system, not a demo. Ask for the three-year cost, not the licence. Ask what you own the day you leave. And above all: ask the engineer who will own your integration to join the meeting, and ask them what will be hard.
Why do AI demos look better than the product?
A demo is the best case: curated data, a rehearsed path, no edge cases. Your reality is the worst case: ten years of messy files, formats the vendor never saw, and users who did not attend the rehearsal. Generative systems are most impressive in the first five minutes and most fragile in month three.
Is the salesperson lying to me?
Usually not. Every phrase is technically true about the demo. The gap is structural: the person paid on the signature is not the person who has to make it survive contact with your data, so the incentive is to leave the hard parts undiscussed. Your job is to make them discussed.
Every figure in this chapter is sourced. The full source list lives on the main map. Open the map
This is one chapter of ten. The whole course is free.
The full map has the interactive tools, the 8 minute audio edition, the live layer and every source. And if you want it run against your own reality, that call is free too.