Models are replaceable. Governed legal execution is not.
Your vendor contract contains a confidentiality clause, and it is well drafted. It does not, however, answer the question professional secrecy raises.
MAX in its own voice: what we build, why it matters, and how we think
10 min
The same idea that structured enterprise data thirty years ago is returning, one layer above the models.
7 min
The whole market is optimizing generation. The real constraint has moved to the layer almost nobody is watching.
7 min
An assistant thinks in sessions; a firm lives in matters. That mismatch explains almost all the friction people feel.
9 min
The decisions that defined MAX are not what we added, but what we refused to build.
8 min
Where others fear the next model, a well-placed layer sees its value grow with every release.
7 min
Confidentiality is not added on top of a product. Three architectural choices are enough to guarantee it from the start.
Why legal AI needs a category of its own, and what makes it different
8 min
What separates a plausible analysis from an enforceable deliverable lies in everything that is not written.
7 min
When a senior reviews a junior’s work, they do not just check it: they perform legal work AI makes invisible.
6 min
Taking over a matter is not about retrieving documents. It is about retrieving decisions and the context that carried them.
6 min
Ten associates, one clause, five years: they converge. What makes them converge is written nowhere.
8 min
A matter never becomes sensitive on the day it is handled, but months later.
8 min
The cost of seven AI tools is not in their subscriptions. It is in everything they impose on you.
8 min
In a firm, access to information is never uniform. That is the first serious obstacle to AI adoption.
10 min
The best model changes every three months. For a firm deciding today, only one question matters.
6 min
Ask firms the question. Nobody answers, ‘one more assistant.’ What emerges has the shape of infrastructure.
7 min
Lawyers adopt AI to start, then drop it along the way. The reason is not distrust.
What AI deployments reveal in practice: costs, pilots, ceilings, adoption
7 min
There is a precise threshold where the question stops being which tool to add and becomes how to hold the whole.
6 min
“Which AI to buy” was a good question. To keep asking it is already buying the wrong one.
8 min
AI budget is read as subscription lines. What that reading misses often costs far more.
9 min
Pilots succeed and deployments fail. It is not the tool that changes, it is the environment that carried it.
6 min
What separates an organization from a group of practitioners is what it accumulates. And current AI accumulates nothing.
8 min
Over six months, which deliverables were produced with AI, how, under what control? Almost no one can answer.
7 min
Compliance was long a background subject. It is becoming a prerequisite for use, and not merely regulatory.
9 min
Every deployment follows the same curve: rapid rise, then stagnation. The ceiling is not technical.
Memory, coordination, semantics: legal AI as infrastructure, not software
7 min
AI memory is often treated as an option to add. That is a category error, and it has a cost.
8 min
A memory that ignores what it contains retrieves documents without knowing what they mean. That is not remembering.
6 min
With AI, logic moves down into the model and value moves up to the layer that coordinates. A change of level.
10 min
An architectural rule so constant it has become invisible: the layer that coordinates always sits above.
8 min
Every wave follows the same path: the lower layer commoditizes, value moves up. Models are entering that phase now.
8 min
There are two ways to use AI in the enterprise. Confusing them explains much of the disappointed investment.
7 min
Three years of debate on AI moats, one thesis displaced every six months. That instability is a clue.
8 min
A simple test separates infrastructure from demonstrator. It is not in the demo, but in what follows it.
The product choices that shape everything: integration, primitives, trust
7 min
The first obstacle to adoption is not model quality. It is the endlessly repeated gesture of changing tabs.
8 min
A rule anyone who has deployed software knows: an unintegrated tool is a forgotten tool.
7 min
An AI that starts over at every interaction is a brilliant junior who forgets everything every night. Exhausting.
8 min
Ask a lawyer to describe a six-month matter. They will not list documents; they will tell a story.
9 min
The unit around which a product is organized decides everything it will be able to do, and never be able to do.
8 min
The market offers tools that answer. As organizations mature, they ask for tools that execute.
8 min
The most reliable quality metric appears in no benchmark: the ability to become invisible.
9 min
A demo wins the decision, the contract is signed, and six months later the tool sleeps. Adoption is decided elsewhere.
10 min
A lawyer does not trust a model. They trust a chain. That difference changes everything.
6 min
Leaving verification to the user’s discretion silently transfers a responsibility the profession refuses.
How the legal AI sector may look by 2028, and how to read the signals
7 min
You do not guess the 2028 legal AI stack, you deduce it. Three levels, and only one will concentrate value.
6 min
A reliable signal announces that a layer is commoditizing: the debate about it becomes boring. Models are getting there.
7 min
The dominant category of enterprise AI changes every five to seven years. The assistant will not be the exception.
8 min
The pattern taking hold in law will reproduce in every profession of expertise. By necessity, not by chance.
6 min
We imagine consolidation through acquisitions until three survivors remain. In legal AI, it will happen between layers.
8 min
A discreet shift is changing the nature of firms: performance no longer depends on talent alone, but on the system.
7 min
All great technologies follow the same trajectory: they are everywhere, then they disappear. AI will follow.
9 min
A software cycle is never defined by its base technology, but by the layer above. The next one will not be either.
9 min
To read a market’s future, do not ask what it will do, but what its buyers will buy.
6 min
Recognizing today who will be ahead in 2028 is not a matter of intuition, but of concrete and observable markers.
What AI changes at the base of expert work: economics, training, responsibility
8 min
AI does not always turn saved time into revenue. Everything depends on what the firm has chosen to sell.
10 min
Automating junior work can preserve knowledge while breaking the mechanism that trained experts.
10 min
The stronger models become, the more decisive the frame that turns their power into usable legal work becomes.
9 min
When AI produces the reasoning, signing is no longer enough: responsibility needs a chain that can answer.
8 min
An error corrected in silence teaches nothing to the organization, because no shared method can absorb it.
9 min
When professional-looking content becomes free, provenance becomes the value that still creates real trust.
8 min
Models can master professional vocabulary without yet speaking the language that makes a profession think.
9 min
AI can compress production time, but it must not compress the time legal judgment needs to mature and be owned.
No posts match this selection.
