Ten Legal Tech Predictions for 2026
The Year of the Great Shake-Out
Predictions are tricky, especially in a sector that moves as fast as legal tech. But I’ll take my chances. Not because I think I’ll get everything right, but because the exercise reveals where we stand and where things might be heading. 2025 was the year of experimentation and investment. I expect 2026 to be the year the market truly consolidates and comes of age.
Together with my legal tech colleague Pim Betist, I’ve mapped out ten predictions for 2026. Five from me, five from him.
Here are my five.
1. Half of all legal AI providers won’t survive
In 2025, we saw an explosion of legal AI vendors. New players popped up everywhere, each touting their own “AI solution for lawyers.” But look under the hood, and you’ll often find the same thing: ChatGPT with a legal veneer. The added value over the underlying model is minimal.
These so-called “wrappers” could still ride the wave of market curiosity in 2025. But in 2026, competition will sharpen. Law firms will ask tougher questions: what does this actually deliver? Why pay for this when I can use ChatGPT or Claude directly?
The thinnest wrappers (those that invested mainly in training and marketing rather than technical product differentiation) won’t make it. I expect a significant share, possibly half, of current providers to disappear or get acquired. What remains will be players that deliver real value: through better integrations, access to legal data, or distinctive workflows.
2. Major legal AI providers will start offering legal services themselves
Consider the funding rounds that companies like Harvey and Legora have raised. Those numbers demand growth, rapid growth. They can’t keep selling to the same sliver of elite lawyers. They’ll need to reach broader markets.
We’re already seeing these players move closer to the client. Legora Portal is a prime example: a shared environment where lawyer and client collaborate, with AI as the connective tissue. But why stop there? If you have the technology, can process the data, and understand the workflow, why not take the next step?
I expect 2026 to bring the first legal AI providers that offer legal services directly. The line between tool and service provider is blurring.
3. Legal publishers will need to shift from product thinking to platform thinking
SDU and Kluwer are sitting on a goldmine. Decades of legal data, meticulously annotated and structured. In an AI world where data is the new oil, they hold reserves that virtually every competitor envies.
But they’re guarding that data like crown jewels. They’re building their own AI tools (GenIA-L at SDU, AI Discovery at Kluwer), snapping up startups (as Kluwer did with Germany’s Libra), and digging ever-deeper moats. Understandable from a defensive standpoint, but strategically shortsighted in my view.
Meanwhile, every AI company is clamouring for access to their data. And those companies have the technology, the innovative drive, and the agility that publishers often lack.
I expect at least one major legal publisher to pivot in 2026. They’ll recognize that their biggest opportunity isn’t building the best AI tool, it’s becoming the legal data platform. They’ll open their data to third parties in a controlled, sensible way. Whoever gets this right first wins.
4. Small language models will take on a bigger role
Bigger is better, that was the mantra of 2024 and 2025. The larger the AI model, the better the output. But we’re now transitioning to “agentic” AI systems: AI that reasons and makes decisions on its own.
In that kind of workflow, you don’t need heavy artillery for every step. Some tasks are straightforward: classifying a document, extracting a date, generating boilerplate text. For those, you can deploy smaller, faster, cheaper models. Claude Haiku 4.5 demonstrated this year just how effective that can be: blazingly fast, remarkably cheap, and surprisingly capable.
In 2026, we’ll see small language models (SLMs) deployed far more widely in production. Not as replacements for large models, but as complements. The orchestrator stays big; the execution increasingly shifts to specialized, leaner models. The result: AI applications that are not just cheaper, but faster and more efficient.
5. Legal embeddings will pull vector search out of the mud
Embeddings are the silent engine behind every AI tool that searches legal sources. They translate text into vectors (strings of numbers representing a text’s meaning) so a system can calculate which documents are relevant to a query. Poor embeddings mean your AI retrieves the wrong sources and delivers the wrong answers.
I’ve previously written about knowledge graphs as a solution to traditional vector search limitations. Standard embedding models from OpenAI and Google aren’t trained to capture legal context in vector space. They match on linguistic similarity, not legal meaning. The system doesn’t inherently know whether “liability” refers to civil or criminal law. A knowledge graph sidesteps this by modeling explicit legal relationships: which rulings cite which statutes, which case law connects to which judgment, when provisions were amended.
But the landscape is shifting. Useful specialized legal embedding models (like Kanon 2) are developing rapidly. That makes the vector database relevant again for legal data. Not as a replacement for the knowledge graph, but as a complement.
In 2026, I expect this combination to emerge as the standard for serious legal AI: knowledge graphs for legal structure and hierarchy, specialized embeddings for semantic relevance. Anyone still building on generic vector search alone is building on shaky ground.
Those are my five predictions for 2026.
Curious about the rest? Pim has made five predictions of his own, at least as interesting as mine. Possibly better, but I’ll let you be the judge. You’ll find them here:
What’s your take? Do these trends ring true? See things differently? I’d like to hear it.
We’ll revisit these predictions at the end of 2026 to see where we got it right and where we missed the mark.