The one fix that moves the readiness needle fastest.
Most businesses try to fix everything a readiness check flags at once. The faster route is finding the single fix that's genuinely blocking an AI agent, and doing that one first.
4 min read
Most of the list doesn't matter yet
An independent bookshop gets its readiness results back: eleven separate issues, from missing stock sync to vague delivery estimates to an unclear returns page. Facing a list like that, the natural instinct is to start at the top and work down, or to hand the whole list to whoever's free. Neither is usually the fastest route to an agent being able to buy from the shop, because eleven issues rarely carry equal weight.
Of the eleven issues, most are refinements an agent would only need once it's already past the first real blocker. A vague returns policy matters once an agent is deciding whether to recommend the shop for a gift purchase, but it's irrelevant if the shop's stock figures are wrong often enough that an agent can't trust a "yes" on availability in the first place. Fixing the returns page first, while the stock problem sits underneath it, spends real effort on something that isn't yet the thing stopping a sale.
Find the first thing an agent gets stuck on
The single most useful test isn't reading the list, it's watching where an agent genuinely stops. Ask an AI agent to check stock on three real, specific titles and see exactly where it hesitates or gives a wrong answer. That's the live blocker, not whichever issue happens to sound most serious on paper. For the bookshop, this might turn out to be the stock sync rather than anything on the readiness report's more prominent early items.
The specific titles matter more than they might seem to. A vague test, "do you have any crime novels," lets an agent improvise its way to a passable answer even with real gaps underneath, because there's room to be roughly right. A specific test, a named title, a named edition, a real customer's actual question, doesn't leave that room, which is exactly why it surfaces the blocker a generic question would have quietly papered over.
One fix, measured, beats five fixes, guessed
Committing to a single fix, then genuinely checking whether it changed the outcome, teaches the business something a long to-do list never does: whether that fix was the real blocker or just a plausible-looking one. If stock accuracy turns out not to be the answer, that's still useful information, because it rules something out and points toward whatever's next in line, rather than leaving a dozen half-finished improvements with no way to tell which one, if any, helped.
The list still matters, just not all at once
The other ten issues on the bookshop's list still matter, just not yet: they're queued behind whichever one is currently stopping a transaction cold, and tackling them in that order gets a shop to its first agent-completed sale faster than spreading the same effort evenly across all eleven from day one.
What Selfe adds here is exactly that visibility: a live check against where an agent genuinely gets stuck, rather than a static list that treats every issue as equally urgent.
How do we know which issue is the real blocker?
Run a real request through an agent and watch exactly where it fails or hesitates. The live result is more reliable than ranking the list by how serious each issue sounds.
What if fixing the top issue doesn't change anything?
That's still useful. It rules that issue out as the blocker and points to whichever one's next, rather than a guess with no way to check it worked.
The readiness score, explained: what each number actually measures.
A readiness score isn't measuring whether an AI agent likes your business. It's measuring three separate, specific things, and knowing which one is low is worth more than the number itself.
What a failed readiness check actually looks like.
A failed readiness check rarely looks dramatic. It usually looks like an AI agent quietly giving up or guessing, which is harder to notice than an obvious error.
What has to be true before an agent presses confirm.
An AI agent completing a booking on someone's behalf needs more certainty than a person browsing does, because nobody is there to notice if something's wrong before it's confirmed.