Manufacturing is a knowledge business. The popular picture is machines and assembly lines. The reality is that much of the value is created before anything is made: in the evaluation, estimation and decisions that settle what gets built, at what cost, and whether the margin is worth it.
That work sits between an enquiry and a quote. It means reading partial information, drawing on past jobs and committing the business to a price, a timeline and a specification. In most manufacturers, experienced engineers do it, and most of what they know has never been written down.
The people who hold the knowledge are leaving
Deloitte and The Manufacturing Institute estimate that US manufacturing will need up to 3.8 million new workers between 2024 and 2033, and that 1.9 million of those roles could go unfilled. Similar pressures run through Europe and India. As senior engineers retire, their judgement goes with them unless the business has a way to keep it.
What evaluation and costing involve
For custom and semi-custom products, every enquiry is a reasoning exercise. It arrives as some mix of specifications, drawings, part lists, requirements and hints about volumes and timelines. Rarely is any of it complete.
A senior engineer reads it and sees what others miss. Which process routing suits this material at this tolerance. Where the costs will concentrate. Which assumptions are safe and which need checking. What similar jobs taught the business last time. These calls are fast, contextual and almost never recorded. They produce a cost sheet, not the reasoning behind it.
Costing then layers on material rates, process costs, tooling, finishing, packaging and margin. Much of it depends on individual knowledge: which suppliers are competitive for which materials, what yields are realistic, where past estimates ran short.
From enquiry to quote can take a day or several weeks. Part of that is the work itself. Part is the queue: one person's output is another's input, and every absence adds a wait.
Why documentation keeps failing
The standard answer to this dependency is documentation. Write it down. Build a knowledge base. Train the juniors. Manufacturers have tried this for decades, and it keeps underdelivering.
The reason is structural, not a lack of goodwill. Documentation is a separate task from the work. An engineer who has just finished a hard evaluation faces the next enquiry straight away. Writing up the reasoning or starting the overdue job is not a real choice. The next enquiry wins, every time.
The documents that do get written describe principles rather than decisions. They capture what should happen, not what did. And they go out of date faster than anyone can update them. Any fix that adds effort on top of the work will hit the same wall.
Make the work itself the record
The approach that works captures knowledge as a by-product of the work. The engineer works through the same decisions as before, through a system that organises the incoming documents, asks for what is missing, and assembles the evaluation or cost sheet from the answers. The intellectual work does not change. The difference is that the medium keeps what passes through it.
Nobody fills in a form. Nobody maintains a wiki. Over time, the business builds something it never had: a searchable record of how it actually evaluates and costs its products. Which routings it chooses for which materials and tolerances. Where its estimates hold and where they need correcting. Which enquiries are routine and which need senior attention.
What changes for people
A junior engineer working through the same system sees how similar products were evaluated before. They get specific prompts for the decisions in front of them. The scaffolding that normally takes years of mentoring is there from the start. Expertise still has to be earned, but newer engineers contribute real work much sooner.
Senior engineers spend less time assembling documents, and more time on the judgement only they can give. Mitochondria's agentic AI systems prepare and structure; the engineer decides. When someone is on leave, changes role or leaves, the reasoning behind hundreds of evaluations stays with the business.
Operational memory compounds
After a thousand evaluations, the business can study its own patterns. Which product lines carry the healthiest margins. Where estimates are most and least accurate. Which customer requirements generate the most exceptions. How cycle times vary by complexity and by who is involved.
Today, most manufacturers answer these questions from instinct. The sales director knows which products pay. The senior engineer knows which specifications cause trouble. Structured and searchable, the same knowledge supports pricing grounded in real cost accuracy, engineering time spent where it matters, and onboarding built on how the business actually works.
The same pattern beyond manufacturing
Manufacturing is where this approach is most developed, but the pattern is everywhere experienced people make judgement calls that nobody captures. Relationship managers in financial services carry their clients' needs in their heads. Field teams in agriculture hold local knowledge of soil, seasons and farmers. Merchandisers know which combinations sell and why. In each case, the answer is the same: design the tools people already use so that the work becomes the record.
A matter of timing
The question is when, not whether. Demand cycles are shorter, products more complex, margins tighter, and experienced engineers are retiring faster than new ones are trained.
A manufacturer that starts building operational memory now gains an advantage that widens with every enquiry. After two years of structured evaluation data, it understands itself in a way a newcomer cannot quickly copy. The question for leadership is plain: will what your best people know still be available to the business when they are no longer in the room?
Signs that engineering know-how is at risk
Evaluations for non-standard jobs depend on one or two people.
Newer engineers take far longer on the same job, and their estimates are revised more often.
Estimates for similar parts vary depending on who prepared them.
Nobody can quickly find how a comparable job was costed, or why.
Key engineers are due to retire or move within the next few years.
How Mitochondria's agentic AI systems keep engineering know-how
The approach. The technical offers and costing evaluation units (units 62 and 63) work inside the engineer's existing process. They organise incoming specifications, drawings and part lists, ask the specific questions an evaluation needs, and assemble the offer or cost sheet for the engineer to review. Decision memory, part of the shared foundation, keeps the reasoning in the manufacturer's own environment.
What it enables. Engineers spend their time on judgement rather than assembly. Every evaluation adds to a searchable record of how the business routes, costs and prices its work. Drawing change control (unit 65) keeps revisions and their consequences traceable.
The impact. Faster, more consistent offers and costings. Know-how that stays with the business when people move on. And a record leadership can use to price with confidence and plan engineering capacity.
Read more about manufacturing.
Sources
—
Mitochondria builds agentic AI for evidence-led core operations. Research-led and model-agnostic. ISO/IEC 27001:2022 certified, with controls aligned to the GDPR, UK GDPR, India's DPDP Act and Australia's Privacy Act. Amsterdam and Pune. Request a working session →