The legal AI market keeps acting like firms belong to one of two camps.
They do not.
Most firms already contain both.
One camp wants approachability.
These are the lawyers who are interested in AI, but cautious. They do not want to become engineers. They do not want to think about models, tokens, provider terms, prompt injection, or workflow design. They want something that helps them move faster without asking them to rebuild their practice around the tool.
The other camp wants control.
These are the heavy users and builders. They want access to the frontier, the ability to shape workflows around their practice, and fewer vendor abstractions between them and the work. They do not want a neutered chat wrapper sold as innovation.
Most firms have both groups at once.
That is why so much legal AI adoption feels mismatched. The products keep forcing firms to pick one camp when the firm itself is made of both.
The cautious camp is not unserious
It is easy to talk about cautious lawyers as if they are simply behind.
Sometimes they are. Often they are responding to real professional pressure.
Their concerns are not irrational:
- confidentiality
- supervision
- hallucinations
- sanctions risk
- discoverability
- client trust
- time cost
These lawyers do not want an AI ideology. They want a workflow they can defend.
When they ask whether the tool is safe, whether output is reviewable, and whether the system preserves professional control, they are not missing the future. They are asking the first serious buyer questions.
The builder camp is not unserious either
It is equally easy to react to the lawyer-builder movement as if it is just play.
That is wrong too.
The builder camp has seen something real before much of the market did: if the people closest to the work can build for the specific pain they feel, they can often produce a better surface than a one-size-fits-all vendor product.
That is the underlying truth in the LegalQuants moment, and it should be taken seriously.
The profession does need more lawyers who understand what is now possible, who can test the frontier, and who can specify better legal workflows than outside product teams could specify on their behalf.
The mistake is not in building.
The mistake is assuming the whole firm therefore wants raw access to the same surface.
Most legal AI products choose one camp
The current market map is clearer if you look at it through that lens.
One category sells approachability first:
- general enterprise legal AI assistants
- polished drafting tools
- chat-based research layers
- internal firm AI wrappers meant to feel safe
These products help the cautious camp because they reduce configuration. They often frustrate the builder camp for the same reason.
The other category sells capability and control first:
- raw model access
- coding tools
- API-first building blocks
- ad hoc internal tools
These products help the builder camp because they preserve flexibility. They often leave the cautious camp behind because they assume too much technical and workflow confidence.
Both categories are real.
Neither solves the whole firm.
The wrong response is trying to make one camp become the other
This is where adoption plans go off the rails.
Some firms buy the safe-looking product and then act surprised when the most capable internal builders route around it.
Some firms celebrate the builders and then act surprised when the broader attorney base does not want to live inside raw, half-structured AI workflows.
Both mistakes come from the same bad assumption: that the right product should turn the whole firm into one kind of user.
Real firms do not work that way.
Different people need different surfaces. What they need in common is not one identical interface. What they need in common is one underlying system that can preserve review, provenance, state, and organizational control across those surfaces.
That is an infrastructure problem.
One platform does not mean one user experience
When I say one platform, I do not mean one rigid front end.
I mean one system of record and control for work that becomes firm work.
That includes:
- shared matter state
- review and approval boundaries
- provenance
- auditability
- bounded context
- tenant-scoped data handling
- durable workflow records
On top of that, the surface can differ.
The cautious camp may need:
- guided specialists
- clear review states
- narrow task framing
- minimal configuration
The builder camp may need:
- configurable workflows
- extensible specialists
- matter-aware APIs
- the ability to shape task logic for specific practices
Those are different surfaces.
They do not require different underlying truth.
The useful distinction is tier, not ideology
Law firms do not need a single approved tool for every AI use case.
They need a tiered stack.
As I argued in the framework that is becoming more useful for buyers, the right questions are:
- what are you doing?
- how is the output being verified?
- why should you trust this source and workflow?
That yields different tiers.
Consumer AI may still have a place for brainstorming, rough synthesis, and internal experimentation where the stakes are low and no protected information is moving.
Enterprise drafting or research tools may still have a place for general firm productivity where terms and access controls are improved but the workflow is still fairly broad.
Firm-scoped workflow infrastructure belongs in the highest-stakes tier: the layer where work reaches clients, courts, regulators, adversaries, or operative internal records.
That top tier is where both camps need to meet.
The cautious camp needs the control. The builder camp needs the extensibility.
Why this is commercially important
This is not just a cultural observation. It is a go-to-market reality.
If a product only speaks to the cautious camp, the technical and workflow-savvy minority inside the firm becomes a source of resistance.
If a product only speaks to the builder camp, the supervising partners, firm operations leaders, and broader attorney base become the resistance.
A platform that can serve both camps changes the conversation.
It is no longer:
- adopt this AI product or resist it
It becomes:
- keep the tools that belong in lower-risk tiers
- add a shared layer where higher-stakes work becomes legible, reviewable, and governable
- let different users meet that layer through the surface that fits their role
That is a much more credible adoption story.
It is also a more honest one.
The profession needs both
The builder minority matters because it expands what the profession believes is possible.
The cautious majority matters because legal work does not stop being legal work just because the tools got better.
The future is not one camp defeating the other.
The future is infrastructure that respects both.
That means a platform where one attorney can use guided, bounded workflows with clear review boundaries, while another can build or extend specialist logic on top of the same shared state and approval architecture.
Neither has to become the other.
That is the point.
The market keeps selling camp identity. Firms need platform design.