Performance Review Framing — by an AI council
Frame difficult feedback so it is honest, fair and actually heard. Four rival AIs debate it and hand back: A structured, fair review with feedback that lands and a path forward.
A structured, fair review with feedback that lands and a path forward.
Signed off by an independent Final QA audit — with a “verify before you rely on this” list, never a black box.
Why this team
Difficult feedback fails at delivery: too soft and the message dissolves, too blunt and the listener spends the meeting defending instead of hearing. The Executive Coach structures the conversation so change is actually possible, the People Lead keeps it fair, evidence-based and procedurally sound, and the Brutally Honest Friend performs the essential service of stating the undiluted message first — so everyone can see what must survive the softening. The Negotiation Expert reads the other side of the table: how this specific person will hear it, and what response keeps the conversation alive.
Each seat runs on a different frontier model — GPT, Claude, Gemini, Grok — so it’s a genuine cross-model review, not one AI in costumes. How a run works →
What the debate sounds like
An illustrative excerpt — A team lead must tell a talented senior developer — the team’s strongest technical contributor — that his dismissiveness in code reviews is driving two colleagues toward resignation.
The undiluted version, so we know what we are protecting: “Your code reviews are making people want to quit. Two of them. You are the strongest engineer here and the biggest reason the team is fragile.” Every softening pass from here must preserve that content — the failure mode of this conversation is walking out having discussed “communication styles” in the abstract while he heard a compliment.
Fairness audit before delivery: is this pattern documented with examples, or is it two complaints and an impression? He will — reasonably — ask “when?”. The lead needs three specific review threads with the actual comments, and dates. Also check the system’s contribution: if he is the only senior reviewer for a large team, the volume pressure is real and shared responsibility, which does not excuse the tone but belongs in the room.
Model his likely first move: technical justification — “my comments were correct”. Do not contest it; concede it instantly, because it is true and it is not the point. “The comments were right. The delivery is costing us people.” Separating correctness from impact removes his best defensive position without humiliating him — and keeps the conversation on the only contested ground, which is behaviour.
Questions
What if the person cries, argues, or shuts down?
The plan includes the likely-reaction branches for your specific situation — the rehearsal covers the hard moment, not just the opening. The general principles: do not retract the message to end the discomfort, do allow silence, and separate the emotion (acknowledged) from the content (unchanged). The one thing to avoid is softening the message in real time to buy relief.
How do I give hard feedback to someone more senior or more skilled than me?
Anchor on impact, not hierarchy: you are not judging their craft — often you concede it explicitly — you are reporting consequences they cannot see from inside their own behaviour. That framing survives the skill gap. The council rehearses the exact concession-plus-redirect for your case, because it is the move that most people fumble live.
Is this for annual reviews or for one difficult conversation?
Both — the structure (evidence, one clear message, likely reactions, owned forward path) is the same whether it lands in a review cycle or a Tuesday. What changes is the paperwork: for formal reviews the People Lead seat adds the documentation and consistency checks that make the feedback fair and defensible.
Your material is used only to run your review — never to train public models. Encrypted in transit and at rest. Security & privacy →
Want full control — pick your own minds, set the depth? Open the full council →