The Intern Test

Here is a simple test for whether your team’s AI use is ready for client work. Imagine the same output came from a bright intern on their third day: fast, articulate, confident, and completely new to your clients, your standards, and the consequences of being wrong. Would you send their memo to a client without a senior person reading it first? Nobody would. Yet teams forward AI output every day with less scrutiny than they would give that intern.

The comparison is not an insult to the AI. It is a calibration. A capable intern produces genuinely useful work, saves the team real time, and still needs review, because usefulness and reliability are different properties. The same is true of AI output, with one important difference: the intern knows when they are unsure and says so. The AI sounds equally confident when it is right and when it is not, which makes the review more necessary, not less.

What we do for interns that we skip for AI

Professional teams have well-worn habits for junior work. A senior person reads the draft before it goes out. The numbers get checked against the source. The claims that sound too neat get questioned. The final version carries the reviewer’s name as well as the author’s, because the firm knows the client holds the firm responsible, not the person who happened to type it.

None of these habits are controversial. They are just what professional quality control looks like, and they exist because experience taught every firm the same lesson: confident drafts from smart people still contain errors, and the errors that reach clients cost more than the review that would have caught them. The strange thing about the AI era is not that the lesson changed. It is that many teams quietly stopped applying it the moment the smart drafter became a machine.

Why the habit slipped

Part of it is polish. Junior work usually looks junior, which invites scrutiny. AI output arrives formatted, structured, and fluent, which signals finished even when it is not. Part of it is volume: AI produces more drafts than any intern, and informal review does not scale with it. And part of it is a category error, the sense that a system this capable must have already checked itself, which is precisely what it has not done.

The intern test cuts through all three. It does not ask whether the AI is impressive. It asks whether this specific piece of work, heading to this specific client, has had the review the firm would require from any other junior source. If the answer is no, the polish is irrelevant. Unreviewed work is unreviewed work.

Making the test operational

A thought experiment only helps if it changes the workflow. Qonera makes the intern test structural rather than aspirational: answers are grounded in your Evidence Base and cited claim by claim, outputs can be stress-tested across independent models, and nothing important leaves without a named reviewer signing off through the review and approval workflow, with the whole path recorded in a tamper evident audit trail. The senior read-through that every intern’s memo gets is exactly what the workflow enforces for AI-assisted work.

The teams that pass the intern test are not the ones that trust AI least. They are the ones that treat it the way they treat every talented junior contributor: real work, taken seriously, reviewed before it ships, with a named person standing behind the final version. That standard built every professional firm’s reputation long before AI arrived, and it is the same standard that will protect that reputation now.