How about law firms becoming software providers (instead of being replaced by them)? (Part II)

Convergence is inevitable. Bringing back the armour.

Part I is available here.

(You are reading a GenAI-free article, solely relying on auto-correct for some expressions and typos. The image above is AI-free as well, retrieved from Canva’s photo repository.)

Convergence is inevitable

As Amazon’s AGI lead, David Luan, explained in very convincing terms in a recent interview, all LLMs are likely to converge in a uniform view of the world. This is terrible news for the leading AI labs, having incurred the unfathomable losses mentioned above. Is it possible that a locally-run open weights model is more than capable enough to power every potential application of GPT5 or Claude to the delivery of legal services at a fraction of their cost?

But it is also bad news for every solution provider whose business model and static building blocks rely on the current cost structure, workflows, and capabilities of those labs. Are they in complete control of their offering when the underlying tokens are subject to wild fluctuations in price or availability? Will they be given priority when resources are scarce or an advertising model comes into play?

Bringing back the armour

With the AI-specific analysis out of the way, we can probably return to the bigger picture of LegalTech, or even PrivacyTech. After all, AI is just a last minute “sprinkle” of probabilistic magic for many, and even when it has become the core component of a vendor’s value proposition, there is a “boring” element of software development that remains essential, and yet we have tended to obviate it in recent months.

Most tasks need structure. Built-in know-how will often manifest itself through an underlying data model or input/output workflows. And tools will still require access controls, logs, or gradual improvements.

As a matter of fact, even in the realm of language and search, so very much affected by the advent of embeddings, deterministic building blocks have a role to play. Why would people keep on using LexisNexis or Westlaw despite all of the AI-powered alternatives popping up every other week to scan through the relevant case law? Quite simply, because you still need scaffolding and trust in the results. Even in the face of Natural Language Processing or semantic interpretation, boolean logic, keyword-based indexing and metadata have tangible value when the final results are mapped to very specific (very real) court cases or official guidelines.

But this is not a fight between deterministic and probabilistic worlds. It is a dance instead.

More to follow soon, on Part III (“How Agentic AI brokers peace”).