A warning, never a pitch.
Founder-bylined essays and explainers. Every piece attaches to one of the four flagships or it is not published here.
Founder-bylined, weekly.
Two kinds of uncertainty
Legal AI conflates two different kinds of uncertainty: shallow ambiguity (you have not found the case yet) and genuine ambiguity (courts are actually divided). Presenting contested doctrine as settled is worse than hallucination, because the citations are real. The fix is architectural: build systems that detect and flag indeterminacy rather than smooth it over. Computational methods cannot eliminate doctrinal indeterminacy. What they can do is systematically identify it.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.
Are you still looking for the clutch?
Citators were a manual verification step bolted onto a manual research process. They flag documents, not holdings, carry no directional stance, and model no binding-authority chains. When authority health is a pre-computed, continuously updated property of every proposition, the separate validation step dissolves into the architecture. The attorney never holds an unverified case.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.
Which bridge are you standing on?
Legal reasoning is constrained, not creative. A “therefore” in a brief must cross one of twelve recognized doctrinal bridges (direct holding, multi-case doctrine, analogical extension, a fortiori reasoning, statutory codification, overruling-chain completion and the rest) or it is argument, not law. The eight common failure modes are exactly the leaps no bridge covers. The craft of advocacy is knowing which bridge you walk across, and never writing the brief as if you were flying.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.
Burden-shifting: the hidden engine of American law
Law is risk allocation. The burden of production shifts; the burden of persuasion never does. Dozens of named frameworks descend from one nineteenth-century distinction, and the twelve most common confusions (which burden shifts, multi-factor tests mistaken for frameworks, the wrong pointing-out standard, overruled deference doctrine) are where motions are denied. A series, with the federal-versus-California matrix and the six deadly errors.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.
We are long the model
A smarter model upgrades drafting, which was never the moat, and cannot touch determinism, provenance you can check, or negative-space completeness. You cannot bake next month’s reversal into last month’s weights. The law is a mutable graph; weights are a photograph. A fluent answer is an opinion. A verified record is evidence. Courts want evidence.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.
Ready for review is where sanctions live
Execution is commoditizing. The deciding step, whether this proposition is true and still good law, is the unsolved problem, and it grows with every agent shipped. One surface, litigation motion practice, a mile deep, in the single place where being wrong ends a case. Verified is a word until someone builds the test.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.
Is your AI conversation privileged? Three variables courts now weigh
Recent decisions do not hold that using AI destroys privilege. They apply settled doctrine to bad facts: a client using a consumer chatbot under terms that permit disclosure, then trying to launder the output through counsel. Attorney direction, enterprise-grade tools and contractual confidentiality are the three pillars. The risk is not AI per se; the risk is careless disclosure and weak vendor terms.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.
Does firm knowledge win cases? What the evidence actually says
The claim that knowledge management wins cases is partially supported, at low-to-moderate confidence. Representation quality moves outcomes; no public dataset isolates a firm’s knowledge system as the cause. Generative AI commoditizes public-corpus knowledge and revalues governed, permission-aware, proprietary work product upward. Claim consistency, continuity and error avoidance. Never claim outcomes.
Full essay forthcoming here; the canonical home is this site. Syndicated copies follow with a canonical link.