Each authorized tech vendor demo in 2026 ends with an AI module, and most of them are spectacular for in regards to the first quarter-hour. The issue is that feature-by-feature comparisons of vendor AI choices are unlikely to foretell which instruments will maintain up over the subsequent three years. The underlying fashions are altering too quick, the combination surfaces are nonetheless maturing, and the workflows that matter most to any given authorized division are particular sufficient that no vendor can anticipate all of them prematurely. What appears to be like like a aggressive benefit right now could also be desk stakes in 18 months, which implies the choice standards that really matter are people who decide how flexibly a software can adapt as situations shift.
The higher query for Authorized Operations consumers this 12 months is whether or not the software offers your staff the flexibility to construct the brokers you really need. Not the brokers the seller imagined after they designed their product roadmap, however the ones that remedy the issues you encounter on Tuesday afternoon when a reporting deadline accelerates, or each time a brand new regulatory requirement lands.
AI researcher, educator, and new addition to the Anthropic staff, Dr. Andrej Karpathy, not too long ago argued throughout a dialogue at Sequoia Capital’s AI Ascent event that infrastructure more and more must be designed with brokers, not people, as the first shoppers. That framing maps immediately onto authorized tech buying choices.
The instruments that expose open APIs, settle for structured inputs from exterior orchestrators, and allow you to outline your individual agent logic on high of their information are those positioned to develop with you. These that don’t will turn out to be closed home equipment operating another person’s product technique.
Most Authorized Ops consumers are nonetheless scoring distributors towards the identical RFP rubric they used 5 years in the past: person interface high quality, out-of-the-box characteristic protection, implementation timeline, and value. These standards stay related, however they miss the dimension that may matter most over the software’s helpful life. A elegant interface solves right now’s identified workflow. A robust improvement floor addresses issues that haven’t materialized but.
Distributors transport off-the-shelf AI options are, borrowing Karpathy’s distinction between “vibe coding” and engineering self-discipline, elevating the ground. They make fundamental automation out there to everybody, which is helpful, however it’s not what separates a Authorized Ops operate that merely operates from one which compounds its capabilities over time. The groups doing their very own builds on open platforms are doing the engineering work that preserves a top quality bar above what any generic module can ship. That’s the reason a software with a median interface and a powerful improvement floor is a greater long-term wager than a software with a cultured interface and a closed again finish.
Each vendor has extra requested options than they’ve the capability to ship, and the sincere ones will let you know that. The query isn’t whether or not a selected workflow will finally seem on their roadmap. The query is whether or not the software is open sufficient for you or a associate to construct it as soon as the necessity is concrete. When evaluating that openness, begin with the workflows the place agent outputs are verifiable with out human judgment. Bill evaluation towards billing tips, fee compliance checking, deadline calculation, and information extraction from structured paperwork are domains the place an agent’s output might be measured objectively and corrected programmatically. Much less verifiable work, like technique suggestions or judgment calls about litigation danger, will comply with later, however the early wins that justify the platform funding will come from the verifiable tier.
Selecting a software you may construct on is barely half the work. The opposite half is guaranteeing your staff retains possession of what will get constructed and why. Throughout the identical AI Ascent dialogue, Karpathy argued that whereas execution might be delegated, understanding can not.
Buildability lives beneath the interface. It exhibits up in whether or not the software has a documented API that covers the total information mannequin, whether or not occasions can set off exterior workflows in actual time, whether or not you may push structured information again into the software with out guide intervention, and whether or not the authentication mannequin helps service accounts that brokers can use with no human within the loop. Ask the seller whether or not you may construct an agent that reads from their system, decides, and writes again, all with no person clicking something. Ask whether or not their API fee limits will accommodate the quantity an agent produces versus a human operator. Ask whether or not their information mannequin is steady sufficient that an integration constructed right now will survive the subsequent two product releases. These are uncomfortable questions for some distributors, and that discomfort is helpful details about how the connection will evolve.
The buildability query lands in a different way throughout the Authorized Ops stack. Matter administration programs have to assist agent-driven matter consumption, automated area inhabitants, and programmatic standing updates. E-billing platforms want to permit agent-based bill evaluation that may approve, reject, or flag line gadgets with no reviewer opening a display screen. Contract administration instruments want to simply accept agent-generated metadata and floor paperwork to brokers by means of structured queries reasonably than solely by means of a search bar designed for folks. Doc administration programs want APIs that allow brokers file, tag, retrieve, and model paperwork the way in which a human would, however at machine pace. For every software within the stack, ask the seller the identical set of questions, rating the solutions, and deal with the ensuing image as a number one indicator of the place the software will sit in three years.
Selecting a software you may construct on is barely half the work. The opposite half is guaranteeing your staff retains possession of what will get constructed and why. In a latest Sequoia AI Ascent dialogue on agentic engineering, Karpathy noticed that you would be able to outsource your pondering, however you can’t outsource your understanding. In case your Authorized Ops staff delegates each the design and the execution of its agent technique to a vendor or a programs integrator with out retaining the conceptual mannequin of how these brokers serve the division’s aims, you’ll lose the flexibility to adapt as circumstances change. The design choices, the workflow logic, the standards for what constitutes a very good consequence: the Authorized Ops staff ought to personal them whereas the construct staff interprets them into implementation. That possession is what preserves institutional information and retains the agent portfolio aligned with the division’s precise priorities reasonably than drifting towards no matter is best to construct.
A software you can’t construct on turns into a ceiling in your operations the second situations change, and situations in Authorized Operations change continually. New laws seem, matter volumes shift, reporting necessities evolve, and the work itself transforms as AI capabilities develop month over month.
The primary half of the work is choosing the instruments with open, agent-ready surfaces that allow your staff reply to these shifts with out ready for a vendor launch cycle. The second half is constructing the inner muscle to resolve what to construct, verifying that agent outputs meet the standard bar, and sustaining the understanding that retains the entire system coherent.
Choosing platforms with open, agent-ready structure is barely the start line. The tougher operational query comes afterward: which workflows ought to be automated first, the place agent authority begins and ends, and who owns the logic governing how these programs function over time?
The following article on this sequence examines the implementation course of layer that sits on high of the expertise itself, together with the workflow guidelines, governance constructions, approval logic, and operational design choices that finally decide whether or not agent-driven Authorized Operations programs produce defensible outcomes or just sooner exercise.
Ken Callander is Managing Principal of Worth Methods LLC, a consulting follow that advises company authorized departments on outdoors counsel pricing technique. He beforehand served as Head of Authorized Operations at Uber Applied sciences. He’s a Licensed Pricing Skilled and holds a level in Physics from Stanford College.
The submit Why You Should Choose Legal Ops Tools You Can Build On (Part 1) appeared first on Above the Law.