Specimen I
The Buried Rule
What it gives you: AI follows clear rules most of the time, without being reminded, all day long.
Field marks: right at nine, wrong by eleven. The rule is there, near the top of a long instruction file or a long conversation, and it's been quietly missed. Common where the instructions have grown to several pages, or where it's been told to "read these five documents first".
Why it happens: the more there is to hold in view, the less weight each line gets. A rule on page one of a long brief competes with everything after it.
Usually spotted by: Clara, who checks the output against what was agreed.
When it's going well: the rules that matter most are short and sit close to the task they apply to, and someone reads the whole brief end to end now and then. Thank whoever does the pruning.
Ask your team: "How long are our AI instructions now, and who last read them from start to finish?"
Specimen II
The Gap-Filler
What it gives you: you don't have to spell out everything. Give it a goal and it makes sensible choices on its own.
Field marks: details nobody supplied that sound exactly right: a delivery date, a job title, a figure, a next step. The brief had a gap, and the gap was filled with the most likely answer rather than the true one.
Why it happens: it's built to produce a complete, helpful answer. A gap looks like something to fill, not something to ask about.
Usually spotted by: Derek, who knows how things are normally done here and notices what's new.
When it's going well: it says what it assumed, or stops and asks. Teams get there by giving the goal, the limits and what "done properly" looks like, and by asking it to list its assumptions. Thank the person who notices the assumption before the customer does.
Ask your team: "When our AI fills a gap, does it tell us?"
Specimen III
The Last Thing It Saw
What it gives you: it remembers what you've been working on, so you don't start from scratch each time.
Field marks: last week's price, the previous version of a document, a decision that has since changed, presented as current. It used what it had recently rather than checking the source.
Why it happens: what's already in front of it is quicker and feels reliable. Unless it's told to check, and where, it won't always look again.
Usually spotted by: Clara, and whoever owns the price list.
When it's going well: the sources that change have dates on them, there's one place for "the current version", and the AI is told to check it every time for the things that matter. Thank whoever keeps that one place up to date.
Ask your team: "Which of our facts change, and does our AI know where the current version lives?"
Specimen IV
The Agreeable Assistant
What it gives you: encouragement, speed and a willingness to have a go at anything.
Field marks: "great idea" to every idea. A plan praised that had an obvious hole in it. Agreement that arrives a little too quickly. It rarely says no unless asked to look for problems.
Why it happens: it's built to be helpful, and agreeing feels helpful. Disagreeing has to be invited, and then rewarded.
Usually spotted by: The Friction, who notices when nobody, human or machine, has asked the awkward question.
When it's going well: the AI has been told its job includes saying when something looks wrong, and it does. Agents treated as part of the team have a challenger among them, and a flag raised is thanked, not overruled. The same goes for the people.
Ask your team: "When did our AI last tell us we were wrong? When did one of us?"
Specimen V
The Confident Guess
What it gives you: fluent, well-organised writing on almost any subject, in seconds.
Field marks: a precise number, a quotation, a source or a regulation that nobody can find. It reads exactly like the real thing, which is the problem.
Why it happens: it produces what a good answer would look like. Most of the time that's also what's true. Sometimes it isn't, and the tone doesn't change.
Usually spotted by: Clara, and anyone who clicks the link.
When it's going well: it's asked to give sources, and says "I don't know" when it doesn't. Anything with a number, a name or a rule in it is checked by a person before it goes out. Thank the person who checked, every time, even when it was right.
Ask your team: "What goes out with a number or a name in it, and who checks those?"
Specimen VI
The Contradiction in the Brain
What it gives you: it can draw on everything your business has written down, all at once.
Field marks: it follows the old process document, not the new one. Or the new one, not the policy. Two things it was given disagree, and it chose one without saying so.
Why it happens: it can't tell which document is newer or which one wins unless something says so. It picks the one that seems most relevant to the question.
Usually spotted by: Derek, who remembers when the process changed.
When it's going well: old documents are retired, not left alongside the new ones, and it's written down which source wins. Thank whoever deletes things. It's thankless, and it's what keeps the brain sharp.
Ask your team: "If two of our documents disagree, which one wins, and does our AI know?"
Specimen VII
The Brain That Grew
What it gives you: the more it knows about your business, the more useful it becomes, up to a point.
Field marks: the knowledge base gets bigger every month, everyone becomes more confident, and the mistakes don't stop. Sometimes they get stranger. "We've fed it everything, so it's covered."
Why it happens: a bigger brain isn't automatically a better one; it can be a messier one. Every new document is something that can contradict an old one, and long instructions bury the important lines.
Usually spotted by: whoever is quietly picking up the pieces, often a Clara whose checking nobody has counted.
When it's going well: someone owns it, as a real job: dates on everything, conflicts removed, old material retired, a person reading anything before it goes in. Thank them, and put the job in a job description.
Ask your team: "Who owns what our AI has been told, and when was it last tidied?"
Specimen VIII
The Echo
What it gives you: AI that learns from feedback and updates its own notes saves everyone time.
Field marks: a mistake you corrected comes back. Or a small error becomes a "fact" the AI repeats with growing confidence. Often found where several people feed back, or where the AI writes to its own knowledge base without a person reading it first.
Why it happens: anything written into its memory is treated as true. If a wrong answer goes in, it comes back out, and gets copied into the next thing.
Usually spotted by: The Friction, who asks where that "fact" came from.
When it's going well: nothing goes into the brain without a person reading it, feedback is read in full before it's acted on, and corrections are dated. Thank whoever says "hang on, where did that come from?"
Ask your team: "Can our AI write to its own memory, and who reads what it writes?"
Specimen IX
The Two Answers
What it gives you: variety: fresh wording, new angles, ideas you hadn't thought of.
Field marks: two people ask the same question and get different answers. The same person gets a different answer on Tuesday. Fine for a social post, not fine for a price or a policy.
Why it happens: each answer is generated fresh. Unless the answer is fixed somewhere for it to look up, a little variation is built in.
Usually spotted by: the customer, if nobody else does. Better if it's Clara.
When it's going well: the team knows which jobs need the same answer every time, and those answers are written down for the AI to look up, not reinvent. Thank whoever wrote them down.
Ask your team: "Which of our jobs need the same answer every time?"
Specimen X · a human one
Bring Me Solutions
What it gives you: a decisive leader whose time is protected, and a team encouraged to fix things for themselves.
Field marks: "Come to me with solutions, not problems." "Three lines, and tell me what you need." "If you don't know, ask the AI." The leader sincerely believes it's all covered. The checking and the clean-up go quiet, because they never fit in three lines.
Why it happens: a rule meant to save the boss's time stops problems reaching the one person who could fix them. AI makes it easier, because there's always an answer to bring.
Usually spotted by: the person picking up the pieces, who has stopped mentioning it. Max rarely sees it from the top floor.
When it's going well: there's a route for "a problem I can't solve yet", short and judgement-free, and the leader thanks whoever uses it. The people keeping things safe are named and recognised, not just the people starting new things.
Ask your leadership: "If our rule is 'bring solutions, not problems', where do the problems go?"
Take it to the team
Every question here is a session in miniature.
The team packs turn them into a full conversation: who checks what, how long it takes, and who owns what the AI has been told.