What Anthropic Means For The Legal Sector

The Operating Model Nobody Designed

When legal teams implement technology—whether it's a CLM, a contract review tool, or now, vibecoded prototypes—they're almost always solving for task completion. Faster redlines. Quicker NDA triage. Automated clause extraction. These are real improvements, and they really matter – but fundamentally, they're improvements to individual transactions within an operating model that, in most organisations, was never actually designed at all.

Ask a legal operations team to map their end-to-end contracting process and what you'll typically get is a combination of official workflow (the idealised version in the CLM) and actual practice (the Teams messages, email chains, phone calls that route around the system).

The gap between these two is where institutional knowledge lives and dies – and where 9 times out of 10, the real work gets done.

• “Why did we accept that liability cap in the last three deals but not this one?” - Not documented.
• “How did our risk appetite on data processing shift after that regulatory change?” - Might be in someone's head.
• “Which business units routinely ignore playbook guidance, and should we update the playbook or enforce it more strictly?” - Nobody's tracking that systematically.

This is the context problem. Legal teams are running sophisticated workflows on top of organisational infrastructure that's essentially informal — held together by individual relationships, institutional memory, and behind?the?scenes effort from people who know how things really work. Technology keeps getting layered on top of this informal infrastructure — and then we're surprised when it doesn't transform anything fundamental!

Why Vibecoding Feels Like Liberation

The vibecoding trend is fascinating not because lawyers are building tools—it's because they're experiencing, many for the first time, what it feels like to have technology that actually maps to their mental model of the work.

When a senior associate at Clifford Chance builds a custom tool for a specific workflow, they're encoding their understanding of that workflow with absolute precision. The tool does exactly what they need because they understand the requirement completely. There's no translation layer, no vendor telling them this is "best practice," no compromise between what they need and what the system allows.

This is useful, feels amazing in practice and is in many ways empowering for individual lawyers… It's also revealing a truth that legal tech vendors should find uncomfortable. For many use cases, the gap between "I understand this workflow" and "I have a working tool" has narrowed so much that the value of a vendor relationship is increasingly unclear. But here's where I think the vibecoding narrative needs pushback—not because it's wrong, but because it's incomplete.

A vibecoded tool reflects one person's understanding of one workflow at one point in time. It doesn't capture how that workflow should evolve as the business changes. It doesn't integrate with adjacent workflows that touch different parts of the organisation. It doesn't encode the institutional learning that should happen when something goes wrong. And it doesn't help when that senior associate leaves and takes their mental model with them.

Individual tools (vibecoded or not) can optimise tasks. They cannot, on their own, create organisational coherence.

The Infrastructure Layer We're Not Building

In every other corporate function, the maturity curve has moved from task automation to process optimisation to strategic infrastructure. Finance didn't stop at faster invoicing—it built FP&A systems that turn transaction data into forward-looking insights. Sales didn't stop at CRM for contact management—it built revenue operations that connect pipeline data to resource allocation to market intelligence.

Legal is still largely at the task automation stage. We've built tools that do legal work faster, but we haven't built infrastructure that makes legal judgment scalable, visible, and improvable.

What would that infrastructure look like?

It would capture not just the output of legal decisions (the final contract, the approval, the sign-off) but the reasoning that led to them. Why we took that position. What constraints we were operating under. What trade-offs we considered. What similar situations looked like and how this differs.

It would surface that context when it's relevant—not as a reference library that requires someone to know what to search for, but as active decision support that says “the last four times you saw this clause pattern, here's what you did and why.”

It would create feedback loops between individual decisions and organisational standards, so playbooks evolve based on actual practice rather than aspirational policy, and deviations from standards are visible not as violations but as data about where standards might need to adapt.

It would distinguish between consistency (doing things the same way) and coherence (making decisions that fit together even when they differ), because legal teams need both and most systems collapse them into one.

The answers to these questions aren't in the text of any single contract. They're in the accumulated pattern of decisions across contracts, over time, within a specific organisational context.

And it's why neither a Claude plugin nor vibecoding nor most current legal tech actually addresses the core problem.

What This Means for Legal Leaders

If you're running a legal team, the question isn't whether to embrace vibecoding or whether to worry about the Anthropic plugin. It's whether you have decision infrastructure—or just better point solutions.

Can you answer these questions about your legal function?

• How has your risk appetite on key contract terms evolved over the last two years, and is that evolution intentional or drift?
• When your team deviates from playbook positions, is that variance captured anywhere, or does it just disappear into individual deals?
• If your most experienced lawyer left tomorrow, how much of their judgment about "how we do things" could be recovered from your systems versus their memory?
• When you implement new technology, how do you know whether it's creating consistency or just moving inconsistency into a new tool?

If the answer to most of these is "not really," then you don't have an infrastructure problem—you have an operating model problem that technology is covering up rather than solving.

The way forward isn't to panic about better models or to dismiss vibecoding as a fad. It's to get serious about designing the operating model first—understanding how legal decisions should be made, how judgment should scale, how institutional learning should accumulate—and then building or buying the infrastructure that makes that model work.

This is unsexy work. It requires confronting how things actually happen versus how they're supposed to happen. It means making implicit judgment explicit, which is uncomfortable — it often reveals that the problem isn't the technology—it's that nobody has actually designed how the legal function should operate at scale.

But it's the work that matters. Because the models will keep getting better, the tools will keep getting easier to build, and the only sustainable advantage will be organisational: whether you've built infrastructure for legal judgment that compounds over time, or whether you're just doing the same work faster.

The Real Threat Isn't Anthropic

The companies that will matter in legal tech won't be the ones that react fastest to the Anthropic announcement. They'll be the ones building for the problem that persists after the noise settles. And the legal teams that will matter are the ones who recognise that technology abundance creates a design problem, not a procurement problem.

When you can spin up a working tool in a weekend, when foundation models can handle sophisticated legal analysis, when the constraint is no longer "can we build this" but "should we build this, and how should it fit with everything else"—that's when operating model design becomes the bottleneck.

The threat to legal tech isn't that Anthropic shipped a plugin. It's that after a decade of investment, most of the market still hasn't built infrastructure for the problem that actually matters: making legal judgment visible, scalable, and improvable across an organisation.

The teams and vendors that figure that out won't be reacting to whatever gets announced next week. They'll be building the infrastructure that makes the announcement irrelevant.