In an opinion piece published on SiliconANGLE, analyst Zeus Kerravala of ZK Research argues that the next phase of artificial intelligence in legal technology will not be defined by a better chatbot or a faster document-summarization tool, but by whether AI systems can take responsibility for real work that unfolds over days or weeks, spans multiple systems and communication channels, and returns control to an attorney at the moments when human judgment matters most. According to Kerravala, Supio is working toward this operational model using long-horizon AI agents. The company's vision extends beyond automating individual steps in personal-injury cases, aiming to create an intelligent operating layer for the law firm that understands the case, the firm's institutional knowledge, work-in-progress status, and the next actions required to move matters forward.
Supio describes its broader platform as a "Firm OS" — a system of action rather than merely a system of record. Kerravala notes that most legal AI products have focused to date on point solutions, such as summarizing medical records, preparing demand letters, searching discovery materials, or answering questions about a single case. While these capabilities are valuable, they leave attorneys and staff to coordinate the workflow around them: recognizing that an action is needed, identifying the right data, navigating communication channels, following up, and documenting the outcome. For this reason, legal AI has so far improved task-level efficiency but had minimal impact on law firms' bottom line. Long-horizon agents seek to take on this coordination and orchestration work, potentially changing firms' cost structures, their capacity to handle more matters, their ability to bring senior expertise to more decisions, and the client experience.
Defining Long-Horizon Agents and Workflow Complexity
A long-horizon agent is defined as an AI system designed to pursue an objective over an extended period, rather than generating a one-time answer or completing an isolated task. Such an agent can maintain context, recognize follow-up work, use multiple tools or channels, make bounded decisions, and escalate exceptions or judgment calls to a human. Kerravala illustrates the difference between asking a chatbot what one needs to know about a client's upcoming treatment, versus asking an agent to manage the entire treatment process: finding the provider, scheduling the appointment, communicating appointment details to the client, and retrieving the records after treatment. This process involves multiple steps with dependencies, shifting conditions, and a real-world outcome.
During a briefing with Supio, Head of Product Dan Zhang presented the example of medical-record retrieval. The simple description "get the records from the provider" obscures operational complexity: the agent may need to validate provider contact details, identify the provider's specific request process, complete forms and HIPAA-related paperwork, fax the request, follow up by phone or email, monitor for a response over days or weeks, ingest the records upon arrival, and alert the legal team if the process stalls. This is what makes the agent "long horizon": it understands the broader goal and can continue working toward it over time, across multiple channels and through intermediate decisions. This represents a more meaningful test of enterprise AI maturity than conversational interfaces alone. While a generative AI assistant can draft an email instantly, a long-horizon agent must determine when to send it, what information it requires, whether a reply has arrived, when escalation or approval is needed, and where the result must be recorded in the system of record.
From Legal Assistant to Firm Operating Layer
Supio's strategy is based on the recognition that plaintiff legal work is not a clean, fully digital workflow. Matters move across case-management platforms, email, voice calls, documents, provider offices, fax systems, and external entities including insurers, clients, and treatment providers. According to company data, roughly 66% (about two-thirds) of the work in a case involves some form of communication with an external party other than the client. This reality demands a connected operational environment rather than merely access to a language model.
Supio is building a platform that brings together case data, firm knowledge, authoritative case law from Thomson Reuters, work status, and communications. As work is performed, agents can document actions, identify follow-up tasks, and construct a picture of matter status. Future agents can then act on this evolving context rather than starting from scratch with each prompt. A traditional case-management system records activity after a person performs it, whereas an agentic system can both perform certain activities and record them as they occur. For the attorney, the value proposition includes spending less time coordinating mechanical work and more time applying legal strategy, exercising judgment, communicating with clients, and making case-management decisions. Supio's vision is for the agent to function like an experienced colleague who knows the organization and the lawyer's work, managing repeatable workflows in the background while lawyers focus on non-delegable decisions.
The Simon Law Group Use Case
Trial lawyer Bob Simon described using Supio to build a personalized agent tailored to his litigation approach. He connected the system to sources such as SharePoint, Outlook, his case management system (CMS), and OneDrive, and added past trial materials, depositions, litigation manuals, expert research, articles, and a book he authored on trying disc-injury cases. The goal was to codify a structured playbook: how Simon evaluates a case, prepares for an expert, identifies weaknesses in an opposing position, and plans to win at trial.
Simon used the system to review depositions against his prior work product, identify materials he might have missed, and refine the agent's analysis. He also applied it beyond traditional legal tasks, such as analyzing the firm's financial information in QuickBooks and reconciling meeting notes, agendas, and a conference website to surface gaps or inconsistencies. In one example shared by Simon, the system identified metadata in a discovery response indicating that the defense may not have produced all materials, and drafted a subpoena targeting the third party from which the information originated. Simon credited this process with helping resolve the case for a substantial sum of money.
At the same time, Simon emphasized the human role, noting that lawyers should verify the work because an agent's output is not a substitute for professional responsibility. In practice, Simon requests source links and verifies key exhibits, evidence, and medical records. This model relies on increasing autonomy for repeatable, low-risk workflow steps alongside human review at consequential decision points.
The Trust Question, Governance, and Domain-Specific Systems
Long-horizon agents face a higher bar than traditional AI assistants because they operate over extended periods and interact with external parties. The design challenge is making their autonomy observable, controllable, and appropriately constrained. Law firms will require clear permissions, audit trails, source attribution, escalation paths, and role-based access controls (RBAC). For example, after realizing that too many employees had access to a financial-analysis capability, Simon's firm restricted access to three authorized users.
According to Kerravala, the legal sector serves as a proving ground for long-horizon AI agents because it is document- and workflow-intensive, highly regulated, and reliant on judgment, trust, and accountability. Supio assesses that leading platforms will not be general-purpose systems attempting to serve every industry, but domain-specific intelligence systems that understand the language, workflows, authoritative knowledge, data, exceptions, and institutional memory of a specific field. In conclusion, Kerravala notes that the next winners in enterprise AI will be companies that transition from providing answers to advancing work execution—requiring deep domain knowledge, access to systems where work occurs, persistent memory, workflow awareness, disciplined governance, and the preservation of human accountability for consequential decisions.