Founders ask Collective 54 this once in our records, and that was in 2026. The knowledge capture and data infrastructure answers on this site cover getting knowledge out of senior people and structuring firm data; this page covers building the assistant people query.
The operations essay in the newer book describes the founder as the default operations manager, answering questions no one else should be answering and resolving issues that should never reach them. It lists founder load protection among the capabilities the role must have: intercepting issues before they escalate and escalating only when judgment is required. The knowledge capture answer on this site recommends using the questions senior people get asked as the list of what to capture.
As an inference, collect the questions first. For two weeks, have senior people note every question they answer that the firm should already know: how we price this kind of work, which template to use, what we told this client last quarter, how we run a kickoff. That list tells you what the assistant must answer, and it keeps the project from becoming an attempt to index everything the firm has ever written.
The lead generation essay lists the material most firms already generate and leave idle: meeting notes, proposals, case studies, interviews, project plans, pricing rationale, analysis of lost deals, onboarding materials and delivery logs. It says the advantage is not the AI but what the firm feeds it, and describes diagnostic frameworks, methodologies, interview scripts and delivery philosophy as what makes AI behave like your firm rather than a commodity.
As an inference, begin with the sources that answer the questions on your list and that someone would stand behind today: current methods and playbooks, standard templates, pricing guidelines, policies and approved case studies. Leave out old drafts and abandoned versions at first. An assistant that confidently quotes a superseded price list does more harm than having no assistant. The data infrastructure answer on this site covers structuring material so AI can reason over it.
The IT essay says boutique firms lack a reference architecture and so select tools in isolation, fracture workflows across systems and bolt AI on rather than embedding it. It describes five layers: intelligence, workflow, data, integration and governance, and says the firm must own the design even when outside providers execute it. As an inference, decide where the assistant sits in that picture before choosing a tool: which data it reads, which workflows it serves, how it connects to the systems where work happens, and who governs it. The build or buy answer on this site covers the choice of tool; Collective 54 names none.
The IT essay says the governance layer enforces security, access control, compliance and decision rights, and the legal essay lists confidentiality among the client terms a firm must control. As an inference, the assistant should show each person only what they could already see without it. Keep client-specific material separated by client, so an answer for one engagement never draws on material from another client. Record who asked what, so the firm can review how the assistant is being used and spot answers that drew on the wrong material. Keep personnel, compensation and ownership documents out entirely unless the access rules are tested. The data protection answer on this site covers what may go into which tools.
The delivery professional essay says AI produces content but cannot produce accountability, and that clients pay for outcomes owned by someone responsible for what is true and safe to act on. As an inference, require every answer to cite the document it came from, with a link, and to say plainly when it does not know. People should be able to check an answer in seconds, and a missing answer should become a question routed to a person rather than a guess.
The operations essay describes AI holding decision memory, retaining what was decided, why, and what was deprioritized, and preserving institutional memory beyond individuals. The service design essay says documentation overhead falls when systems generate and update it automatically. As an inference, name an owner for each source area, such as delivery methods, pricing or policies, who reviews it on a schedule and approves changes. When a senior person answers a question the assistant could not, the answer goes back into the right source, which is the rule the knowledge capture answer describes, so the same question never reaches them twice.
As an inference, launch with one area where questions are frequent and answers are stable, often onboarding, delivery methods or internal policies, and a small group of users. Track three things: the share of questions answered with a correct citation, the questions it could not answer, and whether questions to senior people in that area fall. Add the next area when the first is trusted. The onboarding answer on this site describes how new hires use this kind of resource in their first weeks.
As an inference, adoption depends on habit more than on the tool. Ask senior people to reply to routine questions with a pointer to the assistant rather than the answer, once the answer is there. Show new hires how to phrase questions and how to check the cited source. The staff AI adoption answer on this site covers helping a team use AI without fearing it.
The operations essay lists leadership transferability among the capabilities of its role: institutionalizing how the firm runs, preserving knowledge beyond individuals and reducing founder dependency, which it says determines whether the firm can exit. The IT essay says buyers treat fragmented systems and unclear data lineage as risk. As an inference, a governed knowledge base that answers the questions once routed to the founder is evidence that the firm runs without them.
Collective 54 names no AI tools or platforms and publishes no architecture diagram, source list or access policy for an internal assistant. The published positions are the founder answering questions no one else should, founder load protection, decision and institutional memory, the five-layer reference architecture owned by the firm, the governance layer, client confidentiality, idle firm data and proprietary context as the advantage, AI producing content but not accountability, documentation updated automatically, and leadership transferability.
If your firm is small enough that everyone sits together, as an inference, a short, well-kept set of documents may serve better than an assistant for now.
If your knowledge is mostly undocumented, capture comes first; the knowledge capture answer is the place to start.
And if client contracts restrict where their information may be processed, those terms decide what the assistant may hold.
Collect the questions senior people keep answering, then feed the assistant a small set of current, trusted sources that answer them. Keep the design inside the firm, as the IT essay says, and make permissions match what each person could already see, with client material kept separate. Require every answer to cite its source and to admit when it does not know, because AI cannot be accountable. Give each source an owner, put unanswered questions back into the sources, start with one area and measure whether questions to senior people fall.
As an inference, start with current methods, playbooks, templates, pricing guidelines, policies and approved case studies, chosen to answer the questions senior people get most. The lead generation essay lists the idle material most firms already hold.
The IT essay calls for a governance layer covering security and access control. As an inference, show each person only what they could already see and keep client material separated by client.
As an inference, require citations, let it say when it does not know, name an owner for each source area and put every unanswered question back into the right source.
Collective 54 names no tools. The IT essay says the firm should own the design even when providers execute it; the build or buy answer on this site covers the choice.
Sources: Greg Alexander, The AI-Native Boutique Firm (Advantage Books, January 2027), specifically The AI Operations Manager for founders answering questions no one else should, founder load protection, decision memory and institutional memory, and leadership transferability; The AI IT Manager for tools selected in isolation and AI bolted on, the five-layer reference architecture, the firm owning the design while providers execute, the governance layer for security, access control, compliance and decision rights, and buyers reading unclear data lineage as risk; The AI Lead Generator for meeting notes, proposals, case studies, interviews, pricing rationale, analysis of lost deals, onboarding materials and delivery logs as idle data, and proprietary context as the advantage; The AI Legal Manager for client confidentiality terms; The AI Delivery Professional for AI producing content but not accountability; The AI Service Design Manager for documentation generated and updated automatically. Related Collective 54 answers on this site: how do we get knowledge out of senior people so junior staff are not stuck asking; what data and infrastructure do we need to build to support our AI use cases; should we build AI tools ourselves or find and buy existing software; how do we protect client and sensitive data when using AI tools; how do I design a great onboarding experience for new hires; how do I get my team to adopt AI without fearing it will take their jobs. Note on scope: Collective 54 names no tools and publishes no assistant design or access policy. Collecting questions first, the small set of trusted sources, permissions matching existing access, separating client material, required citations, source owners, starting with one area, the three measures, the adoption habits, and the flips are inferences used here to organize the source material rather than published Collective 54 positions.
Collective 54 is the private community for founders and executives of boutique professional services firms between $5M and $50M in revenue. Members work these answers against their own numbers.