When an AI project runs late, almost nobody names the real reason. People talk about the model, about integration, about priorities. The reason is usually duller than that: the contract is a scanned PDF, the discount rule lives in two people's heads, and the latest version of the price list is in an email from February.
No model can read what your company never wrote down anywhere.
Entropy, not intelligence
a16z named it in its big themes for 2026: the limiting factor for AI companies has become data entropy — the steady loss of freshness, structure and truth inside the unstructured space where most corporate knowledge lives. That is not a sentence about technology. It is a sentence about your shared drive.
How it shows up in practice
Two sources say different things. The report and the system disagree about the month's revenue, and the company decided which one to believe out of habit. An AI has no such habit and will surface the discrepancy on day one.
The rule exists, but it is not written down. "For a long-standing customer we let it through." Whoever decides that today can explain it in thirty seconds. Nobody has ever asked them to write it down.
The document is a photograph. An invoice photographed, a contract scanned crooked, a spreadsheet exported as an image. All of that can be read today, but it costs money and needs checking — and that belongs in the budget, not in the surprises.
The good news
Fixing it is not a two-year project, because you do not have to fix everything. You have to fix the path of one decision. Pick one: credit approval, order release, issuing a duplicate. List where each piece of information behind that decision lives today and who keeps it up to date. Whatever is left with no owner is exactly the work, and it is usually smaller than the meeting suggests.
It is also the one front that pays for itself even if you never do any AI: organized data means faster reports, a shorter close and a shorter argument.
Source: a16z, "Big Ideas 2026: Part 1" (Jennifer Li, on data entropy).
Your mess has a size. It is worth measuring before you hire anyone.
That is the first thing the diagnostic does, and it ends with a written plan.