Earlier in this series, we argued that a business needs a colleague with a position. The research note examined why agreement alone is a poor test of good advice. This final piece is about the product: what a wit brings to a conversation, how advice differs from a decision, and what a record can help you check.
They have a point of view before they meet you
Morgan, our Skeptical CFO wit, has a stance you can read before starting a conversation: “Cash is a fact; everything else is a story.” Morgan's profile names what they challenge, when they speak up and where they defer. Those are concrete commitments: challenge “Spending ahead of signal”; speak up when “Optimism is substituting for evidence”; defer on “Product taste and creative judgment.”
These lines come from Morgan's authored profile. Morgan is a fictional colleague written by Zazuria Labs, not a real CFO whose identity we have copied. The profile gives the conversation a starting position. It is something you can compare a reply against, rather than a promise that every reply will live up to it.
A wit's name and portrait make the relationship recognizable. Their stance and way of working give you something more demanding to recognize: the judgment you came to them for.
Each wit brings their own perspective
Convene Morgan, Sloane and Rafe around one question and you are asking for finance, growth and sales perspectives. The Boardroom generates each wit's contribution separately, with that wit's authored profile and the context available to them. The conversation lets them respond to one another.
Separate contributions do not, by themselves, prove independent judgment. Wits use foundation models, and can share those models' weaknesses. The point of seating different perspectives is to make disagreement possible and visible; how well they sustain it is something to evaluate, not assume.
What they recommended is kept apart from what you decided
The record distinguishes a wit's recommendation from your decision. You can see recorded advice alongside whether you adopted it, declined it, partly took it, or have not ruled on it yet.
Imagine Morgan advising against a price increase and you going ahead anyway. That is an illustrative situation, not a reported customer conversation. A useful record keeps the two positions separate: Morgan recommended one course; you chose another. Your decision should not be rewritten as the advice Morgan gave.
That distinction also tells you how to read the record. It contains the recommendations and decisions that were captured. It is not a guarantee that every remark was extracted correctly, or that the wit can recall every detail of a conversation perfectly.
Changing their mind, on the record
A colleague who never changes their mind is not much help when the facts change. The record supports revisions with dated history and a reason for the change, rather than replacing the old position without a trace.
“I recommended holding the price in March; with the supplier's new terms, I'd raise it in June” is the kind of change that should be explainable. The dates here are examples. The useful question is whether the new evidence justifies the new advice. A saved history gives you something to inspect; it cannot answer that question for you.
The disagreement matters
When you choose a different course from a wit's recommendation, that declined advice can remain alongside your ruling. Boardroom contributions are also attributed to their speakers. The memory system is designed to carry relevant parts of a sitting into each wit's own record.
We do not claim that every objection is automatically recalled by every wit afterward. Keeping the original conversation and distinguishing its speakers matter precisely because summaries and later recall can be incomplete. The standard is a disagreement you can revisit, not a consensus invented after the event.
A track record is not an accuracy score
The track-record view brings together recorded recommendations and what you did with them. Its adoption measure tells you how often you took the advice. It does not tell you whether that advice was right.
To judge that, you still need to connect the decision with what happened afterward, including circumstances the wit could not have known. A confident answer is not evidence of correctness; neither is a high adoption rate. The record's value is that it gives you something specific to question.
You're in the chair
You bring the question, decide what context to share and make the call. Research from another AI app can be part of that context through a supported connection or material you provide. That does not mean every AI app is connected, or that sharing happens without your involvement.
The wit's role is to offer counsel from a recognizable position. Yours is to weigh it against the facts, your constraints and the other people affected.
What we check, and what still needs evidence
On the path where Idiosynt generates a wit's reply, the system checks the candidate against the authored persona and checks references to recalled memories. A failed check can lead to another attempt; an unsupported memory reference can be removed, and a reply can be held. These checks are safeguards, not a guarantee that every statement is true or every judgment sound.
There is a separate mode where a host AI app writes the answer. Idiosynt cannot claim to have checked words it does not receive. That distinction matters when comparing experiences across apps.
The harder question remains how reliably each wit holds a defensible position under pressure, and changes it when the evidence warrants it. We do not yet have a completed comparative result to publish here. This article describes the reviewed product behavior and its limits as of October 7, 2026; it does not report a new live conversation test or a measured advantage over an unmodified model.
That is the standard we want a wit to earn: a position you can understand, advice you can examine, and a record that helps you decide how much trust to place in the next answer.