The agent in the boardroom

Kirin Holdings CoreMate

In July 2025, Kirin Holdings made room for a new participant in its Group Executive Committee: CoreMate. It was not a thirteenth decision-maker. It was an AI adviser built to test proposals and surface better questions before executives made the call.

Worked case · public-source reconstruction Autonomy: advise

Why this case matters

In a conservative setting, the role came first.

A weaker version of the story would be “Kirin put AI in the boardroom.” The stronger reading is an operating-design choice: CoreMate was given a role, not authority. What appears to have mattered before it arrived were the decisions about what it would touch, what it would say, and where human judgement would stay in charge.

Introduced
July 2025
Architecture
12 AI personalities
Knowledge base
10 years of board and committee records
2026 status
Rolled out to operating companies

The case

A thirteenth chair, not a thirteenth vote.

In an executive committee, the valuable work often happens before anyone decides: a proposal is stress-tested, hidden assumptions surface, and weak options fall away. CoreMate is positioned in that moment. Proposal owners can spar with it before the meeting; executives can see selected discussion points during the meeting.

Kirin says CoreMate was built from ten years of Board and Group Executive Committee data, other internal materials, and the latest external information. Those materials are expressed through 12 AI personalities, each bringing a different management perspective.

That is why the case works as a canvas example. The model is not the story. The design of the role is the story: what CoreMate may see, what it may say, what it may never decide, and who remains accountable when its questions enter the room.

Why it matters

The canvas turns a headline into an operating model.

Without the canvas, CoreMate can sound like a quirky corporate AI story. With the canvas, the design question becomes sharper: what kind of participant are we creating, and where does that participant stop?

The operating model then snaps into view. CoreMate has cognition through the model, control through the meeting process and AI policy, and reach through the knowledge sources and display channels it can use. Its power comes from the combination, not from the model alone.

The filled canvas

Turning the story into nine design decisions.

The entries below are intentionally compact. A canvas should not read like a requirements document; it should make the load-bearing choices visible enough for a team to argue about them.

Worked example

Kirin Holdings CoreMate

Agentic AI Design Canvas
designtheagent.com

North Star

What business outcome should this agent's work ultimately contribute to?

Improve the quality and pace of Kirin Group executive decisions by adding source-grounded challenge before and during senior meetings.

Target Workflow & Agent Role

Which workflow is it part of, and what contribution is the agent responsible for making within it?

Workflow: Group Executive Committee preparation and live deliberation. Role: strengthen executive deliberation by acting as a structured challenger, broadening and pressure-testing the discussion without making the decision.

01

Users & Stakeholders

Who uses it directly, and who else is affected by what it does?

Direct users: executives, proposal owners, facilitators, and secretariat. Affected: operating companies, employees, consumers, partners, shareholders, and society.

02

Success Measures & Standards

What measures will show that the workflow improved, and what performance standards must the agent meet?

Measures: proposal readiness, new perspectives surfaced, long-term discussion, decision cycle time. Standards: relevant, source-grounded, concise, low-noise.

03

Autonomy by Action

For each consequential action in its role, how independently may the agent act?

Autonomy: advise. It challenges proposals before meetings and surfaces overlooked perspectives during them. Nothing sits higher: executives hold every decision.

04

Tools & Action Channels

Which tools, systems or channels may it use to get information, make changes, communicate or take action?

Reads knowledge sources, supports pre-meeting sparring, and displays selected discussion points in meetings. No autonomous external action appears in public sources.

05

Rules & Boundaries

What must it always do, what must it never do, and when must it stop, escalate or hand off?

Always follow Kirin AI Policy: human-centric, safe, fair, private, secure, transparent. Never present AI output as a binding decision.

06

Context & Knowledge

What information may it use, and which sources should take priority when they disagree?

Trusted context: ten years of Board and Group Executive Committee data, internal materials, management philosophy, proposal materials, and current external signals.

07

Memory & Learning

What should it remember across interactions, what must it forget, and how should it learn and improve over time?

Remember institutional records and approved feedback. Update external signals. Define what confidential meeting material is retained or forgotten.

08

Ownership & Oversight

Who owns the workflow outcome, who is accountable for how the agent operates, and how will it be reviewed?

Executives own decisions. Proposal owners and facilitators decide what enters the room. Governance sits with Kirin policy and management oversight.

09
Job · what it's for Authority · what it may do Knowledge · what it may know Accountability · who answers

What to notice

Four lessons travel beyond Kirin.

Cells 01 & 04

The role is the product.

CoreMate's most important design choice is not the model. It is the pairing: a role defined as structured challenger, held at the rung “advise”, while executives keep decision rights.

Cells 03 & 05

Executive attention is scarce.

Twelve personalities are useful only if the system filters them ruthlessly. The design problem is selecting the few questions that deserve the room.

Cell 07

Knowledge makes it specific.

The generic version is “an LLM in a meeting.” The valuable version is an LLM grounded in Kirin's history, proposal context, internal materials, and external signals.

Cells 08 & 09

Learning is a governance decision.

Post-meeting feedback can make CoreMate better, but it also raises what should carry forward. Memory, access, and correction are design choices, not afterthoughts.

Coach audit

We ran this canvas through the coach.

Not as an outside fact-checker saying “be careful,” but as an executive assistant asking the one question that matters: if CoreMate is going into the next senior meeting, what would make its questions worthy of the room? Here is what came back.

Coach review · round one

Minor gaps

What holds together

  • One coherent story at the chosen autonomy: an advise-level role, read-only tools, and human decision ownership all pull the same way.
  • The role is disciplined. CoreMate challenges proposals and surfaces perspectives; executives hold every decision.
  • The knowledge base is concrete enough to design around: ten years of senior records, internal materials, and current external signals.

What pulls apart

  • Rules & Boundaries (Cell 06). The agent's real power is choosing what to surface and what to filter, and no rule or measure guards against a one-sided selection. A challenger that quietly filters out valid counter-points can distort a decision while looking helpful.
  • Success Measures & Standards (Cell 03). The standards are aspirational words like “relevant” and “low-noise” with nothing measurable behind them, so there is no way to tell whether it is doing its job well.
  • Context & Knowledge (Cell 07). Good sources are named, but nothing says which one wins when ten years of board data and a current external signal point different ways.
  • Ownership & Oversight (Cell 09). Who owns the decisions is named. Who owns how the agent operates day to day, and how it will be reviewed, is not.
  • Memory & Learning (Cell 08). The confidential-material decision is written as a task still to be defined rather than a decision made.
“What rule stops the agent from quietly leaving out a valid but inconvenient perspective, and when must it flag that it is out of its depth and hand back to a person?”The coach, on Rules & Boundaries (Cell 06)

Sources

What we can say publicly.

Only public-source claims are treated as facts. Anything about internal thresholds, exact prompt-routing rules, or approval workflows should be read as a design implication unless Kirin confirms it.

More worked cases