The Expert's Paradox
If you need to be an expert to verify what the AI says, you don’t need the AI.
A founder pulls up his laptop in a meeting last winter. Claude says this is the right structure for our financing round, he tells me, sliding a clean three-paragraph summary across the table. Properly capitalised, the right legal terms, the right register. I ask him whether the structure is right.
He pauses. I assume so. That’s why I asked.
The line at the top of this piece is the one I wrote in my notebook on the train home. It is the cleanest way I have found to describe a problem that quietly defines AI use in 2026.
Slop is only catchable by the expert
A slop paragraph (well-formed, plausible, quietly wrong) is invisible to the reader who cannot independently locate the error. The fabricated competitor I logged in Algorithmic Slop only fell apart because I had spent years auditing French SMBs and knew the round did not exist. The phantom GDPR article only fell apart because a general counsel happened to remember Article 47.
Without that prior knowledge, both documents would have travelled. This is the rule, not the exception. The defense against slop is domain expertise, applied at reading time.
The bargain that doesn’t close
The implicit promise of consumer AI is that it lets you operate above your competence level. Ask a question outside your field. Get an answer inside your field. The novice is supposed to gain the leverage of the expert.
Read that sentence again. It is the entire pitch.
But the leverage is only real if the answer is right. And the only person reliably positioned to know whether the answer is right is the expert the novice was trying not to need.
The bargain doesn’t close. The novice pays in time and confidence and gets back text that might be true. The expert pays in time and gets back text she could have written herself, slightly faster.
Who actually gets value
The expert does. Not because she needed the AI. She did not. Because the AI is now her drafter, her devil’s advocate, her stenographer, her quick second opinion. She verifies each output in seconds, because she already holds the map. She delegates production. She keeps direction. She becomes faster at things she could already do.
The novice does not. He is buying a confident voice he cannot grade. He is paying for the appearance of leverage. The cost stays hidden until the day someone forwards his deck to a real expert and the room goes quiet.
This is not a moral story. It is structural. AI amplifies what is already there. Where the expertise is missing, it amplifies the gap.
What this means for an SMB
In my fractional-CIO practice, the paradox is the single most useful question to ask before any AI rollout: who, on this team, can verify the output?
If the answer is “no one”, the rollout is dangerous, regardless of how good the demo looks. You are not adding a junior to the team. You are adding a junior whose mistakes wear a senior’s typography.
The fix is rarely a better model. The fix is putting a domain expert in front of every AI deliverable that leaves the building. Sometimes that expert is internal. Often, on a fractional basis, she isn’t. Either way, the expense the AI was supposed to remove (the senior eye on the output) is the one expense you cannot remove.
If you can, you didn’t need the AI.
We can know more than we can tell. — Michael Polanyi, The Tacit Dimension, 1966