Chapter 02 · The method
AI at the core. Nothing taken on trust.
This is a firm built around AI, not a firm that bought some. The systems are engineered by foundational.works on a single premise: the AI is a fast, capable, and unreliable narrator. Everything else follows from that. Nothing it produces advances until it has been grounded in the source, challenged by other systems, and stress-tested for what it missed.
The knowledge comes first. Everything else flows from it.
Every engagement starts by building the body of law and fact underneath your questions: gathered from official sources, structured so it can be queried rather than re-read, and cross-checked so contradictions surface instead of hiding. This is the production layer, the part a traditional firm bills as weeks of associate hours. We automate it, and everything after draws on it.
The law, from the source
Most of a base is built from publicly available material: the statutes, regulations, and controlling authorities that govern your questions, gathered from official sources, so the base rests on what the law says rather than on what the AI recalls.
How it operates in practice
Whatever material carries weight in that market (regulator speeches, agency guidance, practitioner analysis, and industry material are common examples) is curated and captured like any other source. The practical view is derived from them, cited to them, and labeled as a derived view rather than a statement of law.
A separate base, if you want one
If your own documents should inform the work, they live in a knowledge base of their own, kept separate from the publicly sourced one and drawn on only at query time.
Get the knowledge right once, and the work that flows from it is faster, cheaper, and already grounded. That is the whole economic argument of this firm, and it is why the base is built before anything is written.
The two kinds of source stay distinct in the work. What the law says is stated from the authority. How it operates is a practical view derived from published material and labeled as such. Blurring those two is how confident-sounding legal work goes wrong, so we don't.
Stress-tested, reviewed, and challenged.
AI systems invent things, get things confidently wrong, and contradict themselves. That is not a reason to avoid them; it is a reason to build around the failure mode. Four checks run before anything reaches you, and every conclusion that survives them traces back to an authority you can read and verify yourself.
Nothing starts from memory
Every claim begins with the authority itself, the statute or the regulation or the decision, captured from the official source. What the AI remembers a rule saying is not evidence of what it says.
It has to survive disagreement
Conclusions are reviewed and challenged by systems independent of the one that produced them. Models from different vendors have to agree before anything advances, and where they don't, the disagreement is surfaced rather than averaged away.
For the authority that isn't there
The hardest error to catch is the omission. The work is probed for controlling authority it hasn't seen, so a conclusion that the law does not require something has to survive that probe before it stands.
It says so when it can't stand behind an answer
What the checks cannot settle is flagged and surfaced, with the gap explained. An open question you know about is worth more than a confident answer you can't rely on.
The point of all of this is not that the machine is clever. It is that you should not have to take our word for a conclusion. You should be able to follow it back to the authority and see for yourself.
Your information, and where it goes.
This is the first question most in-house counsel ask about a firm built on AI, and it deserves a direct answer rather than a link.
- Most of the knowledge base is built from public authority. Your confidential material is not needed to build it, and usually isn't used.
- Where your own documents do inform the work, they live in a knowledge base of their own, kept separate from the publicly sourced one and drawn on only at query time. Your material is never merged into a shared store.
- Nothing you provide is used to train a model.
- The engagement is with the law firm. Foundational Law, PLLC is a law firm; Foundational Works, LLC is an independent technology company that builds the systems the firm practices with.
- Confidentiality and privilege attach as they would with any counsel. The firm is responsible for your matter and for the confidentiality of everything in it, including how the infrastructure handles it, and the technology operates under that obligation rather than beside it.
A memo ages the day it's delivered. The base doesn't have to.
A traditional work product is accurate at one moment and stale soon after, and every refresh is a fresh engagement. Because the work here rests on a maintained base, an update happens where the change happened, and the deliverables built on it can follow.
The one-shot memo
- Weeks of work, accurate as of a single moment
- Stale within weeks, never updated
- Every refresh is a re-engagement, re-billed
- The knowledge stays at the firm, not with you
A maintained base
- One source across every jurisdiction you operate in
- Updated where the law or your business moved
- Deliverables refreshed from it rather than rewritten
- The base is yours: you own it and can take it with you
Maintenance is optional and ongoing; how it is priced lives in How we bill. The point here is simpler: the asset does not have to age.
Law is by foundational.law. Technology is by foundational.works.
The two are independent companies: the firm practices with the technology, and the technology has to earn its place in the practice. If you are evaluating the systems themselves rather than looking for counsel, that conversation belongs at foundational.works.
Questions about how it works?
We’re glad to walk through the method, the checking, and how your information is handled.