Leah 聯合創辦人兼 CEO Sarvarth Misra 專訪
Sarvarth Misra, Co-Founder and CEO of Leah – Interview Series
Leah 聯合創辦人兼 CEO Sarvarth Misra 在訪談中闡述公司如何由合約生命週期管理拓展至企業 AI 智能體平台,並將重點由協助用戶處理工作轉為跨系統執行工作流程。Leah 與 IDC 的研究指,66% 受訪機構已有 AI 智能體投入生產,並預期部署量到 2027 年初增至六倍。訪談亦談及智能體規模化的治理、互通性及人類監督。
Sarvarth Misra is Co-Founder and CEO of Leah, formerly ContractPodAi. He began his career as a corporate and commercial lawyer, where the time spent locating information, coordinating documents, and tracking obligations inspired him to build technology for legal teams. His work has since expanded from contract lifecycle management to agentic systems that coordinate enterprise workflows, with a focus on combining domain expertise, controlled execution, and human judgment.
Leah develops an enterprise agentic AI platform spanning legal, contracting, procurement, finance, and shared services. Its platform connects specialized agents with enterprise systems, business context, permissions, and approval workflows to help organizations execute work across departments. In this interview, Misra discusses the shift from AI assistance to execution, the governance and interoperability challenges of scaling agents, and how enterprises can measure reliable business outcomes while retaining human oversight.
You began your career as a corporate and commercial lawyer before co-founding Leah (formerly ContractPodAi). What problems did you encounter in legal work that originally inspired you to build the company, and how has that founding vision evolved into what Leah is today?
When I was practising corporate law, I saw firsthand how much of a lawyer’s time was consumed by managing processes rather than exercising legal judgment. Highly skilled professionals were spending countless hours locating information, routing documents, tracking obligations and coordinating work across disconnected systems. While functions like sales and finance had undergone significant technological transformation, legal had largely been left behind.
That recognition led us to build ContractPodAi around a simple premise: legal teams needed a system of record and intelligence for contracts. But as the technology evolved, so did our understanding of the opportunity. We realised that managing work was only part of the challenge. The much greater opportunity was to transform how that work was actually executed.
That conviction shaped Leah. Contracts were our starting point, but the ambition was always bigger: to change how enterprises execute work, not simply how they manage it. Today, we’re building systems that can plan, coordinate and execute complex enterprise workflows, allowing people to focus on judgment, strategy and the decisions where human expertise matters most.
ContractPodAi rebranded as Leah in January 2026 as the company expanded beyond contract lifecycle management into a broader agentic AI platform. What convinced you that the future was not simply adding AI capabilities to existing enterprise software, but fundamentally redesigning how work gets done around autonomous agents?
We came to that conclusion because we’d spent years building the CLM category and were among the first to introduce AI into it. That experience gave us a unique perspective on both the potential of AI and the limitations of the underlying software model.
By 2023, we had reached a fundamental conclusion: adding AI to software designed for people to operate would make individual tasks faster, but it wouldn’t fundamentally change how work gets done.
People would still be responsible for gathering information, coordinating handoffs, moving between applications and ensuring that each stage of a process was completed. AI could accelerate individual activities, but the underlying operating model, and its dependence on human intervention, remained largely unchanged.
We believed the next generation of enterprise software needed to be designed around execution rather than assistance. Systems should be capable of understanding an objective, orchestrating the necessary activities and progressing work across functions and applications, with human involvement where judgment or approval is required.
That required a fundamental architectural shift, not just another AI feature. It was a significant strategic decision, but one we believed was essential. The future of enterprise software would be defined not by how effectively it helps people manage work, but by how reliably it can execute work and deliver outcomes.
The term “agentic AI” is increasingly used across the industry. From your perspective, what separates a genuinely agentic enterprise system from a traditional AI assistant, copilot, or workflow automation tool?
The fundamental distinction is execution.
AI assistants and copilots help people perform tasks, while traditional workflow automation follows predefined sequences. Both can deliver meaningful productivity improvements, but people generally remain responsible for progressing the work and coordinating the steps between systems.
A genuinely agentic system goes further. It can understand an objective, determine the actions required, coordinate specialised capabilities and execute work across enterprise systems, bringing in human judgment when necessary.
But there is an important distinction between an intelligent agent and an enterprise-ready agentic system. Intelligence alone isn’t enough. Reliable execution requires an underlying harness that provides the right context, tools, orchestration, deterministic workflows, permissions and governance.
It also requires domain depth. An agent executing a contracting workflow, for example, needs more than general reasoning capabilities. It must understand the organisation’s contractual positions, negotiation playbooks, approval requirements and the commercial consequences of its actions.
Ultimately, the model provides intelligence, the harness provides the architecture for controlled execution, and domain expertise makes that execution meaningful in a business context.
Leah’s research with IDC found that 66% of surveyed organizations already have AI agents in production and expect deployments to increase sixfold by early 2027. What do you think enterprises are underestimating as they move from deploying a handful of agents to potentially managing hundreds or thousands?
Many organisations are focused on getting their first agents into production. But the challenge changes fundamentally when they begin deploying agents at scale.
The question is no longer whether an individual agent can perform a task. It’s whether the organisation can coordinate and govern hundreds or potentially thousands of agents operating across different teams, systems and business functions.
Our research found that nearly eight in ten companies lack effective tools to identify unsanctioned agents, while approximately 47% identified centralised governance as a critical area for improvement. That points to a growing gap between the pace of adoption and organisations’ ability to manage what they’re deploying.
Without the right architecture, every additional agent risks introducing another layer of complexity. Different teams may deploy agents using different models, tools and platforms, each operating with its own context, permissions and workflows.
Enterprises therefore need more than visibility into individual agents. They need a coordinated environment where agents can share relevant context, operate within clearly defined boundaries and execute together across business processes.
Governance and orchestration cannot be afterthoughts. They need to be foundational capabilities. Otherwise, organisations risk replacing fragmented software applications with an equally fragmented landscape of intelligent agents.
The same research identified governance, orchestration, and cross-functional collaboration as three critical pillars for agentic AI. Why does governance become more difficult when agents can independently reason, make recommendations, and initiate actions across enterprise systems?
Governance becomes more complex because agents are no longer simply producing information for people to assess. They can interpret context, make decisions and initiate actions, often across multiple enterprise systems.
That fundamentally changes the nature of the risk. Organisations need to understand not only what an agent is authorised to do, but also what information informed its actions, which controls were applied and what happened as a result.
The complexity increases when agents operate across functions with different policies, permissions and risk tolerances. An action that is acceptable within one workflow may require additional validation or approval in another. Governance therefore needs to extend across the entire execution process, rather than being applied to individual agents in isolation.
This is why we believe governance must be embedded into the underlying architecture. Enterprises need deterministic controls around critical actions, clear escalation pathways, configurable approval requirements and an auditable record of execution.
The objective isn’t to constrain autonomy unnecessarily. It’s to create the conditions in which organisations can confidently expand it. Effective governance is what makes enterprise-scale autonomy possible.
One of the challenges with enterprise agents is interoperability—agents may be built by different vendors, use different models, and operate across disconnected systems. How should companies architect an environment where these agents can securely share context and coordinate with one another?
Enterprises need to move beyond thinking about agents as standalone products and start treating them as participants in a connected system of work.
The real value emerges when agents can share context, coordinate actions and contribute to a common business outcome, even when they operate across different functions, platforms and underlying models.
Achieving that requires an orchestration layer that sits above individual agents and enterprise applications. This layer provides the architecture for managing context, permissions, integrations, workflows and governance, allowing different capabilities to operate as one coordinated system.
Equally important is preserving enterprise sovereignty. Organisations should retain control of their data, institutional knowledge, business rules and workflows, regardless of which models or technology providers they choose to use.
We believe enterprises should be free to adopt new models and technologies without having to rebuild the processes and intelligence that underpin their operations.
Ultimately, interoperability isn’t only about connecting agents technically. It’s about enabling them to execute work coherently, securely and accountably across the enterprise.
Legal and contracting are particularly sensitive environments because AI-generated decisions can carry financial, regulatory, and legal consequences. Where should organizations draw the line between actions an AI agent can execute autonomously and decisions that should continue to require human approval?
The boundary should be determined by risk, not by the type of task being performed.
Many activities within legal and contracting are structured, repeatable and governed by established policies. These are well suited to autonomous execution, provided the appropriate controls are in place.
For example, an agent reviewing a routine NDA may be able to assess it against an approved playbook, identify deviations, apply agreed fallback positions and progress the agreement without requiring a lawyer to intervene at every stage.
But an agreement involving significant financial exposure, an unusual liability position, regulatory uncertainty or a strategic commercial concession may require human judgment or approval.
Importantly, I don’t believe the objective should be to remove people from the process. It’s to change where and how they contribute. Rather than requiring human intervention at every step, organisations should be able to configure the level of autonomy appropriate to each workflow, including the points at which human involvement is mandatory.
That allows skilled professionals to concentrate on exceptions, negotiation strategy, complex risk assessments and decisions that genuinely require their expertise.
The opportunity isn’t simply to make legal teams faster. It’s to fundamentally rethink how legal work is delivered, with humans and AI each contributing where they create the greatest value.
Leah is expanding agentic AI across legal, contracting, procurement, finance, and shared services. What becomes possible when agents can coordinate across these functions rather than automating each department independently?
Most enterprise work doesn’t begin and end within a single department. A contract negotiated by legal has implications for procurement, finance, operations and compliance. Yet enterprise software has historically been built around individual functions, leaving people responsible for connecting the information, decisions and processes between them.
Agentic AI creates an opportunity to change that operating model.
When agents can coordinate across functions, they can bring together the relevant information, policies and expertise needed to progress work from beginning to end, rather than completing an isolated task and leaving someone to manage the next handoff.
Consider a supplier agreement. The contractual terms influence procurement decisions, payment obligations, financial exposure and ongoing supplier performance. In a connected agentic environment, that contractual intelligence can inform and trigger actions across those functions, rather than remaining locked within a contract repository.
This is where we see the emergence of commercial intelligence: the ability to connect contractual, legal, procurement and financial context so that decisions and actions reflect the full commercial picture.
The value extends well beyond productivity within individual departments. It creates the opportunity to optimise the processes that connect them, reducing fragmentation, improving decision-making and enabling work to move across the enterprise with far less manual coordination.
As enterprises deploy more agents, how should they measure whether agentic AI is actually creating value? Do traditional metrics such as productivity and cost savings remain sufficient, or do companies need a new framework based on outcomes and the amount of work agents can reliably execute?
One of the lessons from the first wave of AI adoption is that activity and value are not the same thing.
Organisations can measure prompts, tokens, interactions and time saved, but those metrics tell us relatively little about whether meaningful business outcomes are being achieved.
Productivity and cost savings will remain important, but agentic AI requires a broader framework focused on execution. How much work is being completed? How reliably is it being executed? Are processes moving faster? Are backlogs reducing? Are exceptions being identified and resolved more effectively?
We should also be measuring the degree of autonomy achieved within appropriate controls. For example, it’s one thing to reduce the time a lawyer spends reviewing a contract. It’s another to enable the entire contracting process to progress, from intake through review, negotiation and approval, with human intervention concentrated where it is genuinely needed.
That distinction matters because the economic value of agentic AI isn’t limited to hours saved. It also comes from increased capacity, greater consistency, reduced operational friction and better business outcomes.
Over time, I expect organisations to place much greater emphasis on the volume and complexity of work that AI can reliably execute, the quality of that execution and the commercial impact it delivers.
The question will shift from how much AI are we using to what can our organisation now accomplish that it couldn’t before?
Looking ahead over the next three to five years, how do you expect the relationship between employees, traditional enterprise software, and autonomous AI agents to change, and what will distinguish organizations that successfully make the transition to an agentic operating model?
I believe we’re moving from an era defined by systems of record to one increasingly defined by systems of execution.
Traditional enterprise software will remain important, providing the underlying data, business records and infrastructure that organisations depend on. But increasingly, agentic systems will sit across those applications, coordinating and executing work that previously required people to move between systems, manage handoffs and progress individual tasks.
For employees, this represents a fundamental change in how they engage with technology. Instead of operating every stage of a workflow, people will increasingly define objectives, establish boundaries, exercise judgment and intervene where their expertise adds the most value.
But achieving that transition requires more than deploying agents or adding AI capabilities to existing applications. Organisations need to rethink how work is structured, how decisions are made and how intelligence moves across functions.
Our research with IDC reinforces the importance of governance, orchestration and cross-functional collaboration. I would add that deep domain understanding is equally critical. A system needs both the engineering capability to execute reliably and the business context to understand what good execution looks like.
The organisations that succeed won’t necessarily be those deploying the greatest number of agents. They’ll be those that recognise agentic AI as an opportunity to fundamentally redesign their operating model.
The real opportunity isn’t to make the way we work today more efficient. It’s to create an entirely new model for how enterprise work gets done.
Thank you for the great interview, readers who wish to learn more about enterprise agentic AI and connected legal, contracting, procurement, and finance workflows should visit Leah.
來源:Unite.AI · unite.ai