Richard Tromans recently shared something that caught my attention: "Anthropic Moves Into Legal Tech. For years people have wondered if the big AI players would ever target legal tech. Well, now they have." When I read that, I immediately thought: this is not just another product update. It is a signal that the boundary between general-purpose AI and legal-specialist tooling is getting thinner by the month.

In his post, Richard points to Anthropic launching new capabilities for its agentic "Cowork" facility across business functions, including "legal," delivered via plugins. He also highlighted Anthropic's own framing: plugins that "extend Claude’s agentic capabilities beyond general tasks into specialized business functions," bundling "skills, connectors, slash commands, and sub-agents" so Claude can work like a domain expert for teams spanning sales, finance, legal, marketing, and more.

Richard's punchline is the part worth sitting with: commoditized legal AI offerings now face a battle on two fronts - DIY vibe-coders and Big AI encroaching into their market space. I agree, and I think the implications will show up in pricing, product design, and buyer expectations faster than many legal tech teams are ready for.

The moment Big AI stops skirting legal

For years, the biggest AI labs and platform providers tended to either:

Avoid legal workflows beyond generic drafting and summarization
Partner quietly with legal tech vendors rather than competing head-on
Talk about compliance and safety as guardrails, not go-to-market

Richard's observation suggests we are moving into a new phase: mainstream AI providers are productizing role-based agents and making "legal" one tile on the same dashboard as sales ops or customer support.

If "legal" becomes just another plugin category in an enterprise AI suite, the differentiation bar for legal tech products rises overnight.

The shift is not simply that Claude (or any LLM) can draft a clause. It is that the provider is building an ecosystem where agents can take actions across tools, use connectors, follow commands, and coordinate sub-agents. That is much closer to an operational assistant than a chat box.

What a legal plugin actually changes

A legal plugin, as described in the Anthropic statement Richard shared, is not only a prompt template. The important words are "connectors" and "sub-agents." In practice, that can mean:

Pulling and pushing data to systems of record (DMS, contract repositories, ticketing systems)
Executing repeatable workflows (intake, triage, escalation, playbook application)
Acting within defined permissions (who can approve, who can send, who can file)
Following org-specific instructions as packaged capabilities, not ad hoc prompts

In other words, the AI is being packaged to behave more like a domain teammate. That has two direct consequences.

1) Buyers will expect end-to-end workflow, not point features

Legal leaders are increasingly impatient with tools that only help with one step. If the general AI platform can support intake, first-pass review, risk spotting, and routing to humans, then a niche tool that only does "summarize" or "redline suggestions" will feel thin unless it is dramatically better in a measurable way.

2) The integration tax becomes the product

One reason legal tech vendors existed in the first place is that connecting to enterprise systems is hard. If Big AI vendors standardize connectors and make them easy to deploy, they remove a chunk of the integration advantage.

That does not mean integrations stop mattering. It means the market starts rewarding whoever can operationalize them fastest, with the least friction, and with the clearest security posture.

Richard's "two-front battle": DIY builders and Big AI

I like how Richard framed the competitive pressure because it captures what many teams feel but do not articulate.

Front one: vibe-coders and DIY legal ops

Thanks to low-code tools, lightweight agent frameworks, and internal innovation budgets, teams can now build "good enough" legal automations without buying a full product. A legal ops manager can prototype:

An intake form that triggers a Claude workflow
A triage agent that classifies requests and assigns owners
A contract Q&A bot grounded in an internal playbook

Even if these DIY tools are imperfect, they can be tailored to the company's own policies and data. That is often what stakeholders value most.

Front two: Big AI platform bundles

When a large vendor sells an enterprise AI subscription that includes legal plugins, procurement starts asking hard questions:

Why pay again for a standalone legal AI tool?
Can we get 80 percent of the value from the platform we already have?
Is this problem really unique to legal, or is it a workflow problem across functions?

Platform bundling does not need to be best-in-class to win. It only needs to be "good enough" plus easy to buy, easy to deploy, and easy to govern.

This is where commoditization accelerates: when the baseline capability becomes bundled, differentiation must move up the stack.

What this means for legal AI vendors

Richard's hint about commoditized offerings facing pressure is the key strategic takeaway. Here are a few practical implications.

Stop competing on generic drafting

Drafting, rewriting, summarizing, and basic clause extraction are becoming table stakes. Vendors that sell these as the core value proposition will feel the squeeze.

Instead, defensible value tends to come from:

Proprietary datasets or deeply curated legal knowledge
Workflow ownership (intake-to-resolution, not one screen)
Vertical specificity (regulated industries, jurisdictional nuance)
Performance guarantees and evaluation methodology

Make governance a feature, not a footnote

If Big AI is entering legal workflows, buyers will scrutinize security, audit trails, retention, and access controls. Legal teams need answers to:

What data is stored, and where?
Can we restrict tools to approved playbooks and clause libraries?
Do we have logs that stand up in an audit or dispute?

Vendors that can provide strong governance without killing usability will stand out.

Embrace the ecosystem instead of fighting it

Some legal tech products will compete with Big AI. Others should plug into it. If the market is moving toward plugin ecosystems, legal tech companies can:

Offer specialized connectors that are harder to build
Package domain playbooks as agent-ready components
Provide evaluation and monitoring layers across agents

In that world, "legal tech" might look less like a single monolithic application and more like a set of composable modules.

What in-house legal teams should do next

If you are in-house and reading Richard's post thinking "interesting, but what do I do Monday," here is a simple approach.

1) Map where agents could safely take first-pass work

Start with low-risk, high-volume tasks:

Intake classification and routing
First-pass document summarization
Policy Q&A with strict sourcing
Contract metadata extraction for reporting

2) Decide your build-vs-buy threshold

Richard's DIY point matters because internal builds can be fast. Define what must be true to buy a product:

Must integrate with X systems
Must provide auditable logs
Must achieve Y accuracy on our data
Must reduce cycle time by Z percent

3) Prepare for vendor consolidation and repricing

As Big AI expands, expect:

More bundled features and downward price pressure on generic tools
More M&A as smaller vendors seek distribution
A premium on truly specialized legal intelligence

That is not a reason to pause innovation. It is a reason to buy deliberately and insist on measurable outcomes.

The bigger picture: legal becomes a standard workflow category

Richard Tromans is right to call this "interesting times." Legal is being treated less like a mysterious exception and more like an operational function that can be supported by the same agentic infrastructure used across the enterprise.

The winners will be the teams and vendors that recognize the shift early: generic capability is becoming cheap, while trust, integration, governance, and domain depth are becoming the real moat.

This blog post expands on a viral LinkedIn post by Richard Tromans, Founder, Artificial Lawyer. View the original LinkedIn post →