How an AI advisory board stress-tests a decision

An AI advisory board for startups is a set of specialized AI advisors you consult before you lock in a plan that spends cash, hires people, or changes the product in ways that are hard to undo. In Nautis the board lives under Company as the AI Advisory Board. Built-in advisors sit next to custom ones you create with a role, instructions, tone, module permissions, and their own knowledge files. Advisors can consult each other, and a board room puts several of them in one thread. This post follows the Custom AI Advisors and AI Co-Pilot help. It does not claim advisors replace a human board, lawyer, or accountant.

What an AI advisory board for startups is built to do

An AI advisory board for startups exists to surface pushback you would rather hear before money moves. You open a card, ask a concrete question with your real constraints attached, and read a reply that is limited by that advisor's instructions and by your own role permissions. When the question spans specialties, the advisor can consult up to three others and credit them by name. That is different from a single chat that tries to be finance, legal, and GTM at once.

Use the board for reversible or partly reversible calls where a second view is cheap relative to the cost of being wrong. Pricing tests, channel bets, hiring sequence, and fundraising narrative checks fit. Do not treat the board as final legal, tax, or medical authority. Put that limit in each advisor's Never do field so the model has a hard stop.

Match the decision to the right stress-test

The table is a sorting tool, not a score. If the row says weak, skip the board and open the right module instead of forcing a chat.

Build one custom advisor the way you brief a hire

Owners and Admins create advisors from the Create advisor card at the end of the board grid. Plan limits decide whether the card is available and how many custom advisors you can hold. When the card is greyed out, you are at the plan limit or the plan does not include custom advisors.

On Profile, set Name, Role, Short description, Icon, and up to four Suggested prompts. The description is what people see on the card, so write it as a job title a teammate would understand. Names must be unique and cannot contain @, <, or >.

On Instructions, write What should this advisor do in plain words: the checks it runs, the format of a good answer, and when it should escalate to a human. Add Expertise and company context for market, policies, or frameworks it should always know. Pick a Tone (Balanced, Direct, Warm and supportive, Analytical, or Concise). Fill Never do with hard rules such as never give final legal advice or never share salary data.

On Access, choose Workspace (roles you tick) or Only me. Module permissions are View, Create, Edit, or Delete per module. Start with View only. Create, Edit, or Delete imply View. Modules left unticked are closed; the advisor says it is not allowed if someone asks there. An advisor can never do more than the person chatting with it. Changes it proposes still go through the usual approval cards.

On Knowledge, upload PDF, Word, Excel, CSV, text, Markdown, or JSON files up to 15 MB each (up to 50 files per advisor). Wait until each file shows Ready. Link data rooms the advisor should search first. Only that advisor uses those files. People still only see room files they can open themselves. Scanned PDFs without selectable text cannot be read; upload a text-based version.

How consultations and board rooms change the answer

Every advisor on the board knows who else is there and what each one covers. Ask one advisor as usual. If part of the question belongs elsewhere, it says so, asks that advisor, and shows a Consulted card. You can also direct the handoff: ask Keel whether a SAFE clause is standard, or get Anchor and Keel's view on a term sheet. Up to three advisors can take a spanning question. A consulted advisor only gives advice during that handoff. It cannot take actions or create records, and it only uses data you are allowed to see. Consultations also work in voice (Talk): the advisor names who it is checking with, then answers.

A board room is a different mode. On the AI Advisory Board page, open the Board room card, name it if you want, pick 2 to 6 advisors (built-in or custom), and start. Mention who should answer with @ chips or @all, or let the room pick up to three relevant advisors. Each advisor replies as its own message and can see the others, so it adds or disagrees instead of repeating. Board rooms are for discussion. Advisors can look things up in your workspace; they do not create, change, or send anything there. To act on a decision, open a one-on-one chat with the advisor that should drive the write.

Every message and every consultation uses AI credits. When credits run out, sending stops until the monthly cycle resets or an admin tops up.

Walk one decision through the board (illustrative)

The walkthrough below is an illustrative example only. It is not a real company, real advisors' chat, or a measured result from Nautis customers.

You are choosing whether to raise list price 12 percent for new logos next month while grandfathering current customers for six months. Cash runway is eight months. Sales says churn risk is low; product says support load will rise.

You open a finance-oriented advisor with View on Financial Planner and Startup Health. You attach a one-page memo with the price table and the kill rule (revert if new ARR growth slows more than X percent in 45 days). You ask for the cash case, the churn case, and what would make the raise a bad idea.

The advisor flags that runway math needs your actual burn, not a round number from memory, and asks a GTM advisor about packaging objections. The consulted reply appears as its own card. The combined answer lists three failure modes: enterprise deals stalling on list vs discount, support tickets rising faster than the memo assumes, and grandfathering creating two SKUs that sales cannot explain.

You do not auto-accept. You tighten the kill rule, update the memo, and open a board room with finance, GTM, and a custom Pricing Skeptic you created with Never do: never recommend a price change without a dated kill metric. After the room, you log the decision in Decision Loop with category Sales or Strategy, confidence Medium, reversible on, review in 60 days, and an expected outcome you can measure. That closes the loop the chat alone cannot close.

Customize built-ins without erasing the defaults

You can tailor built-in advisors (Anchor, Helm, Gale, and the rest) with the same editor. Changes apply only to your workspace. Under Instructions, Built-in instructions stay; Additional instructions, Expertise, Tone, and Never do sit on top and win where they conflict. You cannot replace built-in instructions completely. If you need a wholly different role, create a custom advisor instead. Customized cards show a Customized label. Reset to default restores name, role, description, icon, prompts, instructions, and module access, and deletes that advisor's knowledge files. Who can use it stays as you set it. To customize the Chief of Staff, open Chief of Staff, Settings, Agent Autonomy, then Customize on the card at the top.

Keep permissions, credits, and governance honest

Private advisors (Only me) do not appear for anyone else, and other advisors cannot consult them on someone else's behalf. Deleting a custom advisor removes it for everyone and deletes its knowledge files; linked data-room files stay in their rooms. Past chats remain in history. Renames update the conversation list; board-room replies already given keep the name used at the time, and @oldname still works in a room.

Writes from advisors follow the same Agent Autonomy and Approvals path as other agents. AI Governance (AI Activity) records actions Owners and Admins can review: status, agent, model when recorded, and details such as prompt and outcome. Chat lookups that only read data may not appear. Set risky capabilities to Ask every time if you want a human gate before anything leaves the workspace. See AI Governance for the log and autonomy controls, and Custom AI Advisors for create, access, knowledge, and consultations.

Where the board ends and the operating system begins

The AI Advisory Board stress-tests judgment. The Chief of Staff runs daily operations with Approvals and Autonomy. Decision Loop stores the call you actually made: title, category, confidence, reversible flag, review timing, reasoning, assumptions, and expected outcome, then asks you to record Better, As expected, or Worse later so patterns show up. Use the board to argue the case. Use Decision Loop to own the choice. Use Autonomy so routine writes do not pretend to be strategy debates.

If your plan includes custom advisors, create one narrow role this week with View-only modules and two Never do rules. Attach one real document. Ask one reversible question with a kill metric. Export or pin the chat if you need it later. Then log the decision you took so next quarter's you can see whether the stress-test helped.