A field guide · by Melvin Yuan

The Knowledge Underneath

Extracting what your company already knows
so AI can actually use it.
The limiting factor in AI transformation is almost never the AI. It is the operational knowledge underneath — in documents that are incomplete, in the heads of experts who cannot articulate what they know, and in habits of work no one has written down.

Not a book about AI. A book about what AI needs from you.

The Knowledge Underneath — front and back covers of the paperback, a field guide by Melvin Yuan Coming December 2026
Why AI initiatives stall

The wrong starting place

1

It starts with tools

Which model, which platform, which framework — when the question that actually determines success is what the company knows about how it does its work, and whether that knowledge exists in a form that anyone — or anything — other than a seasoned employee can act on.

2

The knowledge lives in three places

In documents that look complete and aren't. In the heads of experienced people who cannot fully articulate what they know. In habits of work no one ever thought to write down. None of them are usable by an AI agent — or a new hire — on day one.

3

The failure is invisible

An AI agent executes exactly what it has been told. Where the instructions are incomplete, it does not fail visibly — it produces plausible-looking output that is wrong in ways only an expert would notice. The cost surfaces weeks later — as errors discovered downstream, customers who noticed first, and rework no one budgeted for.

Knowledge extraction is not preparation for AI transformation. It is the transformation. The AI is what gets built on top.
From the closing chapter
The method

MRI: three passes through the work

Pass one

Mapping

See how the work flows end to end before going near any detail: what triggers it, its major phases, its decision points, and the ways real runs differ from the clean diagram.

Pass two

Recording

Capture real work as it actually happens — screen-recorded, narrated. Performance reveals what description leaves out; the narration is where tacit knowledge leaks into view.

Pass three

Interrogating

Six questions at every significant step, engaging the expert's reaction rather than introspection — including the single most productive question in the method: what would a new hire get wrong on day one?

The output is not another SOP. It is a knowledge base of atomic, retrievable entries — one decision, one procedure, one failure mode, one escalation criterion each — structured for the way AI systems and humans under pressure actually consume knowledge. Why that structure matters as much as the content is the argument of Chapter 5.

Inside the book

Nine chapters, all operational

1
Why AI Transformation Often Starts in the Wrong PlaceThe three places knowledge lives, and the failure mode nobody sees.
2
Why Experts Cannot Document Their Own WorkThe curse of knowledge, and why "just write it down" always fails.
3
The Three Principles of Good ExtractionDefeat the curse. Observe the work. Structure for the reader.
4
The MRI MethodThe three passes in working detail, with done-when criteria.
5
Structuring Knowledge for AI and HumansAtomic entries, the fields that matter, organizing the base.
6
Where to Start and What to Do NextChoosing the first workflow, and how the work compounds.
7
The Human in the LoopThe four arrangements, and designing the handoffs between them.
8
Common Failure ModesSeven ways extraction projects fail, and the signals that catch them.
◆
What This Buys YouThe case that the knowledge base is the asset — and the AI is the leverage on it.
Who it's for

Three readers

The owner-operator

Founder, owner, or principal — whatever the title, you run a services business where the work follows patterns, and you suspect those patterns could be handled by something other than your most experienced people — if only what they know existed anywhere outside their heads.

The operations leader

You have been handed "AI transformation" and you already know the SOPs will not survive contact with it. You need a method, a sequence, and a defensible answer to "why isn't this deployed yet?"

The extractor

You are the one who will sit with the experts and pull the knowledge out — curious, patient, and outside the work. The book is the method; the Field Kit is your working copy.

Free companion materials

Don't just read it — run it

All companion materials →

MY
The author

Melvin Yuan

The method in this book comes from practice, not theory. Melvin Yuan is a co-founder and the CEO of Stellar, an AI-native company back-office operating system that helps founders handle company formation, compliance, bookkeeping, payroll, and related operational work through one coordinated team and one platform. The MRI method is how Stellar itself has moved from a mostly manual services business to one increasingly built around structured workflows, extracted operational knowledge, and AI-assisted execution.

His background spans several adjacent fields: six years as an infantry officer, developing early urban-warfare doctrine as the army shifted away from its traditional jungle-warfare focus — work that required extracting, codifying, and transferring operational knowledge for an organization undergoing structural change — then communications and digital strategy for companies including Microsoft, HP, and eBay during an earlier technological transition, and two prior technology ventures, including YFind Technologies, acquired by Ruckus Wireless.

He shuttles between San Francisco and Singapore. For occasional notes on the book and new materials, .

Get this right, and AI transformation becomes a matter of building on top of a solid foundation. Get it wrong, and no amount of model capability will compensate for the gaps underneath.