Packaging your professional judgment: how to turn your tacit knowledge into AI agents
After years of practice, you solve cases in minutes that would take someone else days. You spot a problem before you can explain why. That “trained eye” is the most valuable part of your work. It is also the hardest to delegate. If you try to write it down in a manual, it becomes rigid. If you ask AI to do it with a generic prompt, you get answers that are correct but lack depth.
In the previous chapters, you saw how to write system instructions, prepare Brains, design specialized agents, and choose a model for each one. Now we bring all those pieces together with a specific goal: turning your experience into a system that works using your judgment. You will see why expert knowledge is difficult to explain and how to extract it in four phases. You will also see where to place each part within the system and how it comes together in a practical example.
Polanyi's paradox: we know more than we can tell
In 1966, the philosopher and scientist Michael Polanyi wrote a now-famous line in The Tacit Dimension: “We can know more than we can tell.” Years later, economist David Autor called this idea Polanyi's paradox to explain why certain jobs are so difficult to automate.
An experienced professional rarely works by following a step-by-step manual. They might be a neuropsychologist interpreting an attention pattern, an expert assessor evaluating a risk, or a consultant reorganizing a company. In every case, they are guided by expert intuition: their mind very quickly compares the case in front of them with thousands of situations they have encountered before.
Psychologists Daniel Kahneman and Gary Klein analyzed this kind of intuition in a 2009 paper. Their conclusion includes an important qualification: expert intuition is reliable when it has developed in an environment with regular patterns and clear feedback on whether a judgment was right or wrong. Outside those conditions, it may simply be overconfidence.
Why neither a manual nor a prompt is enough
When experts try to digitize or delegate their work, they usually encounter two dead ends:
- A traditional procedures manual is flat and rigid. It works for standard cases, but not for the ambiguity of real-world cases.
- A generic prompt gives answers at the level of a first-year student: formally correct, but without the insight that experience brings.
The challenge is not to ask the machine to be smarter. It is to bring your tacit knowledge to light and turn it into instructions, knowledge bases, and decision criteria that a system can apply.
The ECE protocol: extracting expert judgment in four phases
To formalize your experience without losing its depth, apply this four-phase protocol.
Phase 1: look through your most difficult cases
Expert knowledge does not show up in textbook cases that anyone can solve. It shows up in the exceptions.
- Choose between 5 and 10 real cases from your past work that were particularly difficult, ambiguous, or unusual.
- For each one, ask yourself: what subtle signal led me to choose solution X when the theory pointed to Y?
- Note those signals in a pattern log. Those less obvious factors are what you will later give the agent through its knowledge base.
Example: a consultant remembers rejecting a seemingly solid business plan because all revenue depended on a single client who had not yet signed. That signal does not appear in any manual, but it defines his judgment.
A practical tip: if you find it hard to explain your decisions, ask another person or an agent to interview you about each case. Have them ask “And why?” over and over. Many rules only come to light when someone makes you justify them.
Phase 2: list what you would never do
What distinguishes an expert from a beginner is not just what they do. It is also what they know not to do.
- Write down the ten diagnostic errors, hasty assumptions, or misleading solutions that someone with little experience would fall into when facing that type of problem.
- Turn each error into an explicit constraint in the agent's system instructions.
Example constraint:
“If sales rise but the repeat customer rate is below 20 %, do not conclude that there is organic growth. First assess whether it is due to seasonality or a one-off campaign.”
This is the brakes block of the R-C-M-F-E framework we covered when discussing system instructions, but now with your own known errors.
Phase 3: make your decision criteria explicit
Intuition is often a fast, unconscious way of assigning more or less weight to certain factors. The goal of this phase is to put that weighting system in writing. Answer three questions:
Which conditions are essential?
Requirements without which a conclusion is not valid
“No valuation is issued without financial statements for the last two years”
Which factors aggravate or mitigate risk?
Factors that raise or lower the risk level
“Dependence on a single supplier raises the risk by one level”
In what order are options ruled out?
The sequence you follow in your analysis
“First, data consistency. Then, market assumptions. Finally, strategy”
The result is a rubric: a table of criteria that the agent can consistently apply in the same way.
Before moving on, validate your rubric. Remember Kahneman and Klein's qualification: not all intuition is reliable. Test it against past cases whose outcomes you already know. If a rule would have failed often, revise it before encoding it. A system does not just amplify your judgment; it also amplifies your biases.
Phase 4: distribute your judgment across the system's three components
Once your judgment is written down, place it in the right component. This is the anatomy we covered in chapter 3: engine, fuel, and mechanisms.
A simple rule for deciding: if it is knowledge, it goes in the Brain. If it is a way of thinking, it goes in the instructions. If it is an exact check, it goes in a tool.
Practical example: auditing viability plans
Let's see how all this applies in an illustrative scenario. The figures help explain the change; they are not a guaranteed outcome.
Before: hands-on work
A senior consultant charges 2.500 € to review economic viability plans. He spends around six hours reading each document for inconsistencies, guided by his experience. He writes a tailored report and takes around three days to deliver it. Taking everything into account, he can handle around four reports a month.
After: judgment turned into a system
- Brain: contains the industry's reference cost manual, a history of budget variances, and five excellent reports that serve as examples of the expected output. The reports are anonymized.
- Extraction agent: extracts the cost tables, revenue forecasts, and contingency items from the plan and organizes them into a JSON record.
- Calculation tool: checks that the figures reconcile and calculates margins and ratios. The model does not do the arithmetic.
- Auditing agent, using the consultant's rules. For example: “If customer acquisition cost does not include sales staff, flag a HIGH alert.” “If the gross margin exceeds 80 % in a physical products business, ask for justification of the supply chain or flag it as a possible inconsistency.”
- Writing agent: prepares the draft report using the consultant's tone, structure, and format.
- Expert review: the consultant reviews the alert matrix, adjusts the strategic nuances, and signs the report.
The result
In this scenario, the consultant's time per case can drop from around six hours to around 25 minutes. His capacity could rise from 4 to around 40 audits a month, because his work focuses on what only he can provide: the final judgment.
Maintaining rigor requires two conditions:
- Human review must be genuine. If the consultant signs without reading, the system loses its main safeguard. This is the automation bias we covered in module II.
- The system must be reviewed frequently. Carry out sample audits: from time to time, thoroughly review a complete report as if you were doing it from scratch, and compare.
From hands-on professional to asset owner
Packaging your judgment changes the nature of your work in two ways.
Your method no longer depends on your presence
Your methodology can be applied to several clients at once. You step in at decision points, not at every step. This is the structural leverage we covered in chapter 7, as opposed to simple productivity gains on microtasks.
Your knowledge becomes a transferable asset
As long as your knowledge exists only in your head, your business depends entirely on you: if you are not there, the service disappears. A well-documented set of agents, Brains, and workflows can become a company asset. You can pass it on to your team, sell it, or license it.
To have real value, that asset needs:
- Clear documentation of how it works and its limitations.
- Maintenance: someone to update the rules and documents when the industry changes.
- Clearly defined rights in your contracts covering the methodology, documents, and agents.
Precautions before packaging your judgment
- Protect confidentiality. Your past cases may contain client data. Anonymize them before uploading them to a Brain, and comply with the GDPR and any professional confidentiality obligations that apply to your work.
- Review the regulatory framework. Auditing business plans is usually a minimal-risk use. If your judgment is used to evaluate people in areas such as employment, education, or credit, it may fall within the high-risk category under the AI Act. In healthcare, such as diagnosis, medical device regulations may also apply. Review the chapters in module III and consult a specialist.
- Start with one part of the service. Do not try to package all your work at once. Choose the most repetitive phase with the clearest rules.
Conclusion
Your experience is your greatest advantage, but as long as it exists only in your head, it is also your greatest limitation. By analyzing your difficult cases, writing down the mistakes you know to avoid, making your criteria explicit, and distributing them across Brains, instructions, and tools, you turn your intuition into a system that applies your judgment consistently. You still make the decisions at key points, but you no longer have to be there at every step.
Want to start turning your professional experience into agents and workflows? Visit libertadia.com and sign up in the app.