Workflow & productivity: 20 ways AI gets it wrong
Forgotten constraints, self-contradiction and 'almost right' deliverables. Each failure mode below is phrased as the question people actually ask, with what it looks like in real work — and the layer of the Trust Stack that catches it.
Does AI guess instead of asking for what it's missing?
A deliverable is produced on a guessed value instead of asking for the real one.
Why does AI ask questions it could have answered itself?
It stalls with clarifying questions when it had enough to proceed.
Can AI overcomplicate a simple task?
A one-line answer is turned into a multi-step process no one needed.
Does AI oversimplify a complex task?
A genuinely hard problem is reduced to a checklist that ignores the hard parts.
Why does AI lose track of earlier instructions?
A constraint given at the start is quietly dropped by the end.
Can AI forget the constraints it was given?
The output violates a rule that was clearly stated up front.
Does AI contradict itself across responses?
Today's answer flatly disagrees with what it said earlier.
Why is AI formatting inconsistent?
Headings, lists, and styles change shape from one section to the next.
Can AI ignore the format it was asked for?
A table was requested and prose was delivered instead.
Does AI ignore word and length limits?
A 200-word brief comes back at 800 words.
Why does AI add sections nobody asked for?
Extra headings appear that the brief never requested.
Can AI drop sections that mattered?
A required section is missing from the final output.
Does AI fail to prioritize what matters most?
Everything is presented at the same weight with no sense of what's critical.
Why doesn't AI explain the trade-offs?
An option is recommended with no mention of what you give up to choose it.
Can AI fail to produce something actually usable?
The output looks complete but can't be used without rebuilding it.
Does AI create busywork instead of progress?
The work generates more tasks to manage than value it delivers.
Why does AI force the user to check everything by hand?
Nothing can be trusted, so every line has to be re-verified manually.
Can AI produce "almost right" work that still needs heavy editing?
The draft is 80% there but the remaining 20% takes longer than starting over.
Does AI reliably judge the quality of its own work?
It pronounces a flawed answer "complete and correct."
Why does AI give a polished answer before actually thinking?
A confident, well-formatted reply arrives before the real reasoning was done.
More from the library
One model can’t reliably catch its own mistakes. A council of independent minds can.
Run your work through the councilAll 250 failure modes · See also: the Trust Stack