How to capture tacit knowledge with AI
What is tacit knowledge, and why do AI agents need it?
Tacit knowledge is what your experienced people know but never wrote down. Which client needs a call before an invoice goes out. Which source is current and which one is stale. Who to check with before something ships, and who actually reads it after.
New hires pick this up over months, mostly by getting things wrong. AI agents never pick it up at all unless someone gives it to them. An agent with your documents and none of this judgment produces work that looks right and still needs fixing by someone who knows better.
That is the gap between an agent that drafts and an agent you can trust to act.
Why aren't documents enough?
Documents record the what: the report that was sent, the SOP, the ticket, the decision in the thread. They rarely record how someone got there, and almost never why.
Logan Reilly, COO of Dactic, puts it simply: "What we're looking at is the how and the why to the what."
Here is where it shows up. Someone searches, filters, and ends up with three results that are all fair game. An experienced person picks one based on what they have seen before. An agent does something different. In Logan's words: "An AI harness is just going to go top result. Boom. That's the one."
The top result is often fine. When it isn't, the agent has no way to know, because the reason to choose differently lives in someone's head. More documents will not fix that. Neither will a longer prompt.
How does an AI interview capture the how and the why?
Interviews are how good consultants have always captured tacit knowledge. The trouble is that they are slow, they depend on everyone's calendar, and someone is always on vacation.
An AI interview changes the effort involved. With Dactic, the whole flow fits in one sitting. Connect your sources, answer a 5 to 7 minute AI interview, and deploy your first agent with one click.
Because Dactic reads your connected sources first, the interview does not ask you to describe what is already written down. It asks about the gaps: the steps that never show up in a document, the exceptions, and the judgment calls behind them.
Which questions surface the why?
Good interview questions ask for decisions, not descriptions. Use this checklist whether a person or an AI runs the interview.
- "When three sources all look right, how do you pick one?" Surfaces source judgment.
- "When do you skip or change the standard process, and why?" Surfaces the real exceptions.
- "Who do you check with before this goes out?" Surfaces approvals that no policy lists.
- "Who reads the result, and what do they look at first?" Surfaces the real audience.
- "What would a smart new hire get wrong in their first month?" Surfaces lessons learned the hard way.
If your AI agent cannot answer these five for a given task, it is working from the what alone.
What are the three circles of context?
Dactic builds context in three circles: personal, company and industry, and flags where you work differently from your industry.
| Circle | What it holds | Example |
|---|---|---|
| Personal | How one person works: their sources, steps, check-ins and judgment calls | An account lead reads a client's last three emails before drafting any proposal |
| Company | How your firm works: approval rules, client commitments, house style, who owns what | Invoices above a set amount need a partner's sign-off before they go out |
| Industry | How firms like yours usually work, so your differences stand out | Most firms in your space bill monthly. You bill per milestone, so Dactic flags it |
The industry circle matters because an agent can only see your side of the story. Knowing where you differ from the norm tells it which defaults to ignore.
How does captured knowledge reach your AI agents?
Captured knowledge only helps if an agent can act on it. Every agent ships with its instructions, context files, skills, tools and model choice. Dactic recommends where agents would help most, and you choose where they run. One click, and about 20 seconds later your agent is in your AI platform.
Deploy to Claude, OpenAI or monday.com, or to local models through MCP. Nothing deploys to your AI platform until you approve it.
You can read and edit every agent's instructions, give it feedback, and see your own context graph. Every dot in your context graph is a piece of evidence, a person or a workflow signal.
What happens to your data during capture?
Dactic doesn't store your raw data. It's processed ephemerally. What carries forward is the context an agent needs: the steps, signals and decisions behind the work.
Your agents run in your own AI platform. No compute happens inside Dactic.
How do you keep tacit knowledge current?
Tacit knowledge changes faster than documents do. A new client rule, a price change, a new approver. A snapshot taken once goes stale.
Context refreshes every 30 days on Launch, every 14 days on Growth, every 7 days on Pro, and daily or continuously on Enterprise. When your business changes in a big way, Dactic runs a short follow-up interview.
It also covers the change every firm dreads. When someone gives notice, their context stays with your agents.
Where should you start?
Pick one task your best person does better than anyone else. Run the five questions above on it. Then check whether your AI agent has the answers. If it doesn't, you have found the gap, and you know what to capture first.