On September 15, 2026, London-based startup Telon received Select Partner status in OpenAI’s partner network. The news spread quickly through trade outlets, but the headlines missed the real point.
OpenAI has effectively acknowledged something: the model itself isn’t the main problem. The real difficulty sits elsewhere — in everything that happens between buying a license and having a working system in production.
And that part gets solved by lawyers who know how to translate law into algorithms.
What Happened
Telon was founded in June 2026 by Lewis Bretts, a former PwC partner who previously built the firm’s LegalTech practice, and Tom Mellor, former COO of SYKE.
In three months the team grew from 15 to 30 people, and the startup raised funding from The LegalTech Fund.
The company now officially helps OpenAI’s clients deploy its models inside law firms and legal departments.
Telon’s working model looks nothing like selling another legal chatbot. The company hires and trains “legal engineers” — most of them former practicing lawyers — and embeds them inside the client’s team to:
- configure models and write system instructions;
- build AI agents and working scenarios;
- connect the AI to internal documents and software;
- train lawyers and oversee how the technology gets used.
The pricing works differently too: Telon charges for results, not billable hours. That said, results-based pricing is no guarantee the client wins every case.
Payment depends on whether the digitized pipeline actually works:
- requests get triaged correctly;
- no critical logic errors slip through;
- it genuinely saves in-house lawyers’ time.
A Correction on the “300,000” Figure
Select Partner status is only the first of three tiers in the OpenAI Partner Network (Advanced and Elite come after it).
OpenAI has invested $150 million in building out this network and expects to certify 300,000 consultants by the end of 2026.
Here’s the catch: that figure covers the entire partner network, across every industry at once (founding partners include Accenture, Bain, BCG, McKinsey, and PwC). Telon is one member of that network, focused specifically on law.
None of that makes the news any smaller. If anything, the opposite is true: OpenAI is building a deployment channel exactly where its own resources fall short.
A specialized team of former lawyers filled that gap in law — not developers, and not a generalist consulting giant.
How This Differs From “Just Another AI for Lawyers”
Lewis Bretts frames the market’s problem this way: law firms buy the technology, run successful pilots, and a few months later staff are using only a fraction of what the system can do.
Models keep getting updated on top of that, so AI adoption can’t be a one-time project: it has to be maintained continuously.
I see exactly the same pattern working with LegalTech and FinTech startups trying to build legal logic into their products.
A developer knows how to write code, but doesn’t know how to spot the moment an algorithm’s decision turns dangerous for the user.
Take an example: the AI needs to stop and hand a case to a human specialist whenever a request shows signs that call for manual review.
What decides that isn’t the code — it’s the law, translated into precise “IF… THEN… ELSE” logic, well before a developer ever touches the task. The code comes second.
In practice, this logic isn’t an abstraction. It’s a concrete branching diagram you can map out in advance.
Here’s what that looks like in a hypothetical example: simplified logic for automating insurance payouts.
[Insurance claim submitted]
│
├──> Scenario A: damage confirmed by documents, first-time claim
│ └──> [Automatic payout calculation]
│
└──> Scenario B (Red Flag): signs of fraud OR repeat claim on the same case
└──> [Automation blocked] ──> [Escalated to a human expert]Legal architecture isn’t an attempt to limit what the neural network can do.
The smarter a model gets, the more sophisticated and convincing its mistakes become in edge-case legal situations — exactly where reality departs from the typical scenario.
Triage Logic and a Red Flags matrix serve a different purpose: putting hard safeguards exactly where the cost of a single model error is real money or a person’s rights, not just lost time.
The market has now given this work its own name — and it’s what I do: I translate legal expertise into LegalTech and FinTech product architecture.
That means condition matrices, decision trees, critical-risk registers, and data-verification specifications for developers — the raw material a working digital solution actually gets built on.
Legal Architecture for LegalTech and FinTech Founders in the CIS
Telon’s model is built for large law firms and corporate legal departments abroad. That’s a different scale than most startups in Russia and the CIS operate at early on.
But the principle this news confirms is universal: it doesn’t depend on company size or jurisdiction.
Without legal architecture designed in advance, AI in a legal product will eventually either produce a dangerous recommendation or stall on an edge case nobody anticipated.
If you’re designing or scaling a product where an algorithm makes the decisions, and the cost of a mistake is real money or a user’s rights, build this part in from the start. Patching it in later costs more.
Frequently Asked Questions
What is legal architecture? Translating legal expertise into a format ready for building a digital product: condition matrices, decision trees, critical-risk registers, and technical specifications for data verification. It’s work at the intersection of law and systems design, done before developers write a line of code.
What is a legal engineer? A specialist — usually a former practicing lawyer — who translates legal logic into algorithms and technical specifications developers can work with, then configures and maintains how AI models operate inside legal processes.
How is a legal engineer different from an AI-product developer? A developer writes code and wires up models. A legal engineer determines which rules, constraints, and red flags need to be built into the system before the developer starts building it. Without that, the AI can’t tell a safe scenario from a dangerous one. A detailed breakdown of that line → How a legal engineer differs from a prompt engineer is its own piece: Prompt Engineer vs Legal Architect.
Do small startups need this, or is it only for large law firms? Yes, if the product makes decisions — even partially — that affect a user’s rights or money.
Company size shapes how much work this is. Building the right logic in before launch is necessary either way.
Designing or scaling a LegalTech or FinTech product with legal logic inside? See how legal architecture works — the formats, what I do, and how it fits into your development process.