How to capture tacit knowledge from everyday work
The order of a task, the sources someone checks, and the questions they ask AI agents reveal how they work. Use work history to turn those habits into knowledge a team can reuse.
Read in Japanese →The know-how that never makes it into the deliverable
Two people can research the same topic and use the same AI tools, yet approach the work very differently.
Which source do they check first? What do they tell an agent, and which answers prompt a follow-up question? When do they stop discussing and try something? Those small decisions reflect experience.
Because these choices often feel natural, someone may struggle to explain them when asked for tips later. The finished deliverable rarely shows the decisions along the way.
To share that know-how, start by tracing what happened during the work.
An example: a setting changes, then reverts after a restart
Consider someone investigating an app setting that reverts after the app restarts. The screenshots below show sample data created for this article in ContextBerg for Mac. The same task runs through both screens.
Instead of telling an agent only that settings are not being saved, they explain that the new value takes effect immediately but disappears after quitting and reopening the app. They then ask the agent to compare the code that saves the setting with the code that reads it at startup.
Review the sources they consulted and the conditions they gave the agents, too. The useful material is not just the final fix, but what they checked first and how they made their questions more precise.
Turn the work history into approaches others can try
The aim is to capture more than a fix for this particular settings file. Use the task sequence, reference material, and requests to agents to identify approaches another person could reuse.
In the Chat example, the same task report is attached. The request asks for possible approaches to checking the problem, separately from questions that still need the person's input.
| What the history shows | What to make explicit |
|---|---|
| They explained that the setting reverts after a restart | State the actions and conditions that reproduce the symptom at the start of the request |
| They compared the save location with the startup read location | Check for mismatches between paired operations, rather than investigating only one side |
| They separated tests from checks in the actual app | Say what has been verified and what still needs checking |
These exchanges may feel routine to the person doing the work. Describing their sequence gives someone less familiar with agent-assisted work an approach they can try.
Make the reasoning explicit before sharing it
A recorded action does not explain why someone took it. Opening a reference page, for example, does not establish that they adopted its advice.
The person reviews the AI's suggestions and adds why they followed that sequence, and under what conditions the approach would work again.
What to check first, how to brief an agent, when to test something, and what to reconsider if it fails: once those choices can be explained, others have explicit knowledge they can refer to.
From personal experience to knowledge a team can use
You do not need to document an entire role at once. Start with one investigation, one decision, or one task completed with an agent.
Use ContextBerg's records to identify an approach, have the person check it, and share it. Let someone else try it and fill in what was unclear. Repeating that process turns habits that once felt instinctive into knowledge a team can use.