![[Simplifing AI agents with powerful models and ontologies#^agents-inherit-confusion]] [[Tiago Forte]] taught a generation of knowledge workers to [[Build a Second Brain|build a second brain]]. The question every organisation now faces is the one Forte never answered: **how do you build a *shared* one** - for humans and agents alike? The recurring complaint is the same in every organisation that has tried to build one: ![[Knowledge problems#^scale]] The personal-knowledge playbook does not survive contact with a multi-consumer system. [[CODE method|Capture, organise, distill, express]] - the four moves that work for one person - collapse the moment [[Ownership|ownership]] is shared and nobody is on the hook. A platform issued without a method or an owner ends the same way: a graveyard of stale entries. > This is not a technology problem. It predates AI by decades. AI just makes the gap expensive. ## At a glance **Purpose.** Build the foundation an organisation needs before deploying a shared knowledge system or AI agents against a coherent knowledge base. **Objectives:** - name the pains - decompose what real questions depend on - find where each piece lives - and what is only in someone's head - judge the state of each piece, and match a method to each gap - name the owners accountable for keeping it true - sign the [[Decision record|decision record]] **Structure.** Three acts and a co-writing session at the end. Before anyone touches a tool. ## For whom ![[How agentic AI can reshape real estate’s operating model#^mckinsey-on-demos-not-scaling]] The workshop addresses organisations where: - The organisation knows more than it can find. [[Why does clarity in decision-making matter?|Decisions get made twice]]. - Standards live in people's heads as [[Tacit knowledge|tacit knowledge]] and leave with them. - [[Tribal knowledge]] dominates - and you cannot tell how much it costs you. - [[AI agent|AI agents]] get deployed without the context they need, then fail in ways nobody expected. - [[Decision log & Single source of truth|Knowledge is scattered across tools, channels, and teams]] - and nobody is on the hook for any of it. - [[Decision log & Tribal knowledge|Decisions made on tribal knowledge cannot be reconstructed]] when an audit or incident review needs them. The same investment serves both populations - the human one and the agent one: ![[Employees and agents as knowledge consumers#^claim]] **Audience:** Managers and leaders facing these problems. Technical staff with their own systems already in mind are not the intended audience. **Vocabulary:** Technical where the topic demands it - [[Knowledge management|knowledge management]] is a technical domain - but assumes no prior background in knowledge management theory. ## Preparation The workshop produces concrete outputs only if the room can answer concrete questions. Gather the following before the session: 1. **Specific incidents.** 3-5 recent moments when someone could not find knowledge they needed - what they were trying to do, what they could not find, and what it cost (a delay, a duplicated decision, a wrong call, an incident, time lost on onboarding). 2. **Informal experts by name.** The people who get the same question from different teams. 3. **Where knowledge lives now.** The tools and channels currently holding it: wikis, ADR repositories, Slack, code comments, project documents, anywhere else. Note which are active in name only. 4. **Recent AI or agent deployments.** What has been shipped or is about to ship, and what knowledge those systems need access to. 5. **Decision authority in the room.** Confirmation that someone with the authority to assign owners will attend. The workshop produces named stewards; those names only stick if the person who can resource them is present. A workshop where half the answers are "we do not actually know" wastes time - and the people in the room. For examples of what missing, stale, unlinked, or mis-permissioned knowledge has cost real software organisations, see [[Knowledge management failure cases]] - useful both as a reference while filling in the items above and as anchors when explaining the workshop to a sponsor. ## How it works ![[The Network Secrets of Great Change Agents#^battilana-casciaro-on-change-via-middle]] There are three ways to pin down what your organisation knows: you can **point** at it ("ask Sarah"), you can **write its definition**, or you can **list the cases**. Most knowledge lives only as the first - which is why it walks out the door when Sarah does, and why an [[AI agent|agent]], who can be handed a definition but cannot be pointed at Sarah, stalls without it. The workshop is one climb from pointing to a written definition, in three acts: 1. **What does a question depend on?** We take one concrete question a real knowledge consumer needed answered and decompose what answering it actually requires. 2. **Where does each piece live - and in what state?** We place each piece where it is really kept - a tool, or someone's head - and judge it: written with one clear source, written but scattered across places, only in a head, out of date, or missing. 3. **What do we do about each gap, and who keeps it true?** Each gap gets the move that closes it and a named owner accountable for it. A method without an owner is an aspiration; an owner without a method is overhead - so the last act binds the two, and neither resolves without the other. ## Acts The three acts of the technique - decompose meaning, place and colour, refactor to green - are defined on the [[Knowledge mapping|board]]: ![[Knowledge mapping#^dependencies]] ![[Knowledge mapping#^containers]] ![[Knowledge mapping#^methods]] ## Outputs A single artefact: a draft of a **[[Decision record|decision record]]** signed by the participants. ![[Impact-oriented decision making#^parts]] [[McKinsey]] raises the question this record answers: ![[How agentic AI can reshape real estate’s operating model#^mckinsey-on-owning-learning-loop]] It contains the outputs of all three acts: - The named pains the organisation is paying for every day - A dependency map - what real questions depend on, decomposed, and which pieces are only in someone's head - The state of each piece - what is written with a clear source, what is scattered with no single source, stale, or missing - and which questions an agent could not yet answer - A method per gap, and the implementation principles that survive any tool change - For knowledge scattered across many places, a decision on where it should live: kept local, centralised, or published for everyone - Named owners - the people walking out with a concrete to-do, connected upward to the leadership level that can resource it ^outcomes The record is readable by the team, by a new hire on their first day, and by an agent given the task of continuing the work. It is the written form of the shared understanding the participating organisation reached during the engagement - a record most organisations do not produce on their own. ## Examples Four end-to-end walkthroughs - one personal-scale, three organisational. They show how the room's own vocabulary survives end-to-end without the facilitator imposing a category. 1. [[Second brain example - personal notes|Second brain (personal notes)]]: ![[Second brain example - personal notes#^summary]] 2. [[Shared second brain example - ADR log|Shared second brain (ADR log)]] ![[Shared second brain example - ADR log#^summary]] 3. [[Shared second brain example - Expertise Radar|Shared second brain (Expertise Radar)]] ![[Shared second brain example - Expertise Radar#^summary]] 4. [[Shared second brain example - Experts Hub|Shared second brain (Experts Hub)]] ![[Shared second brain example - Experts Hub#^summary]] Different scales, different methods, different tools - same pipeline. The workshop never imposes either. ## Out of scope McKinsey reframes the underlying question: ![[How agentic AI can reshape real estate’s operating model#^mckinsey-on-workflows-vs-use-cases]] Tool adoption is the next step and is not part of this engagement. Tool selection requires knowledge of the organisation's stack, budget, constraints, and political reality - context this engagement does not develop. A tool recommendation from outside is a guess; a tool decision made against a requirements sheet and a method decision is a consequence of those inputs. The outputs of the engagement provide the inputs needed for that decision.