Bob has been with the company for 30+ years. He knows the SKU master better than the system does. He knows which suppliers are reliable in August and which ones quietly slip in December.
He knows that a certain customer always pulls orders forward in week 47, because of a regional promotion that officially stopped in 2018 but kept running informally. He knows the workarounds his younger colleagues use without knowing where they came from, because he invented half of them.
But what happens when Bob retires? In most cases, the company loses years of hard-won knowledge that lived in his head, was never documented, and will take years to rebuild. At a very high price.
When Bob retires, the company does not just lose a person. It loses 30 years of pattern recognition that was never written down.
This is not just a Bob problem. It is an industry problem, and the math is brutal. Accenture's 2026 supply chain workforce study projects that demand for core US supply chain roles will rise 19% by 2035. The labour pool is expected to grow by just 3.2%.
Translation: there will be 1.1 million supply chain jobs in 2035 that nobody is going to fill. The senior people we have today will retire. The juniors coming in are fewer in number, and they are walking into roles shaped by people they will never overlap with.
This is the real talent crisis. Not that we cannot find people. The people we have are leaving, and they are taking the institutional knowledge with them, one retirement party at a time.
When supply chain leaders talk about AI, the conversation is almost always about productivity. Automate the report. Speed up the cycle. Real and important. But the bigger problem is knowledge transfer.
Every organisation I have worked with runs on tacit knowledge that lives in people's heads. The senior planner who knows the supplier. The senior buyer who knows the workaround. The senior demand planner who knows the official forecast is wrong about a specific SKU every Q4, and how to override it without breaking the system.
When these people leave, the knowledge does not transfer. The new hire rebuilds it from scratch, badly, making the same mistakes the senior person learned to avoid 15 years ago.
The SharePoint folder is a fiction. The 90-day overlap is theatre.
Most AI strategies are not addressing this. It is also the problem AI is genuinely well-suited to solve today. We call the answer apprentice agents.
Pair
Pair a retiring expert with an apprentice agent that works alongside them for three to twelve months.
Capture
The agent observes how decisions get made: the workflows, the exceptions, the workarounds, and the rationale nobody wrote down.
Transfer
When the expert leaves, the knowledge stays. The next generation queries it and gets the answer in seconds.
The agent does not replace the senior person. It supports the next generation. New hires get onboarded faster, because the institutional knowledge is now accessible, queryable, and explainable.
A junior planner can ask the agent "what would Bob have done here?" and get a useful answer in seconds.
There is also a longer-term reason to build this now. The agents you will eventually run on routine work need context about how your business actually operates: the undocumented rules, the supplier quirks, the seasonal patterns, the customer-specific exceptions. Most companies try to deploy agents without that context, which is exactly why those deployments hallucinate or fail. Apprentice agents build the company brain that future automation needs. The investment compounds.
The technology to do this exists today. RAG-based agents on a properly scoped knowledge corpus can capture process knowledge and decision rationale at a fidelity that was impossible five years ago. The barrier is not technical. Few companies prioritise it, because the payoff does not show up on a productivity dashboard for two or three years.
And it connects directly to that 1.1 million role gap. If you take the gap seriously, you cannot hire your way out of it; the people simply do not exist. You can only make the people you have, and the people you bring in, dramatically more capable. Apprentice agents are the most direct path to that outcome. See how blueclip gets your data and processes AI-ready →
A working apprentice agent program has roughly these five elements.
- Find the irreplaceableIdentify the 10 to 20 people whose departure would create the biggest knowledge loss. Not necessarily the most senior. The ones whose colleagues say "ask Bob, he knows."
- Scope it narrowlyOne process. One product category. One supplier portfolio. Everything Bob knows is the wrong scope. Everything Bob knows about Q4 promotional planning for the dairy category is the right scope. Multiple narrow agents beat one broad one.
- Run it alongside the humanThree to six months minimum, not a one-week documentation sprint. Tacit knowledge surfaces during edge cases, and edge cases happen on the calendar of the business, not your project plan.
- Validate with the successorBefore the senior person leaves, test the agent with whoever is taking over. Can they actually use it? Are the answers right? The agent is done when the next person can do the job with its help, not when the senior retires.
- Treat it as a system, not a projectThe agent keeps learning as the business evolves, as new edge cases emerge, and as the next generation adds their own knowledge to it.
An agent that observes how someone works is monitoring software. In most jurisdictions, companies already have broad rights to monitor work performed on company systems, so the legal mechanism is mostly the same as the monitoring that already exists. EU works council consultation, California privacy rules, union agreements, and senior-executive contract clauses can add requirements. None are blockers. All are conversations to have early. Make the program consensual and transparent: let people see what the agent captures, correct it, and flag what should not be captured. Some companies formalise this with a contribution bonus tied to a successful knowledge transfer. The economics easily justify it.
