Yesterday, Anthropic launched Claude Cowork’s legal plugin.

Within hours, public legal and information services companies - Thomson Reuters, RELX, LegalZoom, and Wolters Kluwer - saw double-digit percentage drops in their share prices.

The market read the signal clearly: frontier model providers are coming for legal tech.

But we believe this reaction fundamentally misunderstands where the real value in legal - and white-collar work more broadly - actually sits.

Frontier Models Have Hit a Learning Ceiling

Frontier models were trained by scraping the open internet. That era is over.

The modern internet is increasingly locked behind paywalls. High-value legal analysis, commentary, precedents, regulatory interpretation, and practical guidance are not freely accessible - and never will be again.

As a result, frontier models do not continue to learn from the most important sources. Their knowledge becomes increasingly static, even as the world - and the law - continues to evolve.

A model trained on yesterday’s internet cannot reason accurately about tomorrow’s problems without being given new context.

Why Context Beats Training

At Jylo, we believe the next phase of AI capability will not come from retraining base models on proprietary data. It will come from context based expertise provided to the model at runtime.

There is a crucial distinction here:

• Training or fine-tuning hands your intellectual property to the model provider

• Context engineering lets you apply that expertise without giving it away

This is not just a technical choice. It is a strategic one.

White-collar industries should not centralise their IP inside frontier model providers who may later compete with them.

Legal Is the Perfect Example

A base model may scan legislation.gov and “know” the law - but it will often be confused about:

• Whether an act has commenced

• How regulations interact

• How courts are actually interpreting provisions

• What practitioners do in practice

If instead you feed the model legislation, regulation, and relevant judgments in context, you get far more precise reasoning.

That value does not come from the model. It comes from the curation and ownership of the context.

Why the Market Reaction Is Misplaced

The sharp sell-off in Thomson Reuters, RELX, Wolters Kluwer, and LegalZoom assumes that Anthropic - or any frontier model provider - can simply “eat” the legal value chain.

They can’t.

These companies are not valuable because they host documents. They are valuable because they have:

• Deep, proprietary work product

• Editorial judgment and domain expertise

• Trusted relationships with professionals

• Embedded workflows in mission-critical environments

That IP sits behind paywalls that frontier models will never legally or practically access.

Hiring a few lawyers does not replicate decades of institutional knowledge.

The Enterprise Incentive Problem

Frontier model providers have spent billions training these models. They face extreme inference costs and pressure to give models away cheaply.

To make the economics work, they must move upmarket into enterprise.

But here’s the problem:

Why would enterprises centralise their expertise, IP, and competitive advantage into a duopoly of frontier model providers?

Especially when those providers are incentivized - economically and strategically - to climb the value chain.

We very much expect enterprises won’t.

In-House vs Law Firms: A Subtle but Critical Difference

Claude’s legal offering appears primarily targeted at in-house legal teams, whose core incentive is cost reduction on legal fees.

Those teams may not care if some external law firm IP leaks into a $20-a-month AI tool.

Law firms, however, will care deeply.

Firms that are thinking clearly will not allow their captured work product, precedents, and expertise to be absorbed by platforms that may later compete with them.

Instead, they will offer AI-enabled services to clients through trusted partners who are not incentivized to replace them.

The Real Moat Is Trust

At Jylo, we believe yesterday’s news reinforces a simple truth:

Trust is the only durable moat in AI.

White-collar industries are desperate to modernize. They want AI-first capabilities. But they do not want vendors who are quietly positioning themselves to take their place.

The future belongs to organisations that:

• Help incumbents transform without disintermediating them

• Apply AI without absorbing proprietary IP

• Align incentives with long-term professional services models

Three Types of AI Companies Will Exist

We see three clear categories emerging:

1. Trusted transformation partners for incumbents

2. Data-hungry platforms planning to compete later

3. AI-first vertical operators replacing incumbents

All three exist today.

At Jylo, we are unequivocally in the first category.

We believe it is the largest opportunity - and the most honest position.

We also believe some players are trying to be all three at once. Yesterday’s announcement suggests Anthropic may be one of them.

AI Will Proliferate, Not Consolidate

The world cannot run on a handful of frontier models.

Knowledge work will fragment, not converge. See our recent article on this.

Firms will:

• Build their own tools

• Run models on their own infrastructure

• Partner with vendors they trust

• Retain control of their expertise

AI companies will proliferate. Consolidation will fade as a relic of the SaaS era - when software was a tool, not how the work itself got done.

The Bigger Economic Tension

Enterprises are being asked to pay more and more for AI at the exact moment AI is eroding the value of what they sell.

That is not sustainable.

Markets are beginning to notice.

Frontier models are approaching parity. Breakthroughs are incremental. Meanwhile, value destruction is accelerating.

The response will be decentralisation.

The Signal We Take From Yesterday

Yesterday was not the death knell of legal incumbents.

It was a warning shot for anyone who believes that frontier models will own professional services end-to-end.

They won’t.

People, projects, trust, and captured work product remain the true defensibility.

Knowledge work is being commoditised - but trust is not.

That is the future we are building towards at Jylo.