Beyond the PDF: How Docusign's AI agents navigate the context problem in enterprise contracts

Asking an AI to summarize a single contract is straightforward. Asking it to calculate what one company owes a supplier—accounting for three amendments, a revised pricing schedule, and an internal procurement policy—is a different challenge entirely.

This distinction sits at the heart of how Docusign is reinventing itself beyond electronic signatures. The company's push into contract intelligence reveals a fundamental gap: generic language models lack the organizational and documentary context needed for real business decisions.

The Four Layers of Contract Context#

Docusign's chief technology officer, Sagnik Nandy, frames the problem around four categories of context. Within a single agreement, tracing how a term defined on page two gets referenced on page 20 is manageable. But when contracts span multiple amendments and revisions, the real work begins.

As Nandy notes, a typical LLM sees three interconnected documents as "three or four independent PDF files." In practice, they form a single contractual relationship. An agent must recognize that link and determine which version of each clause currently applies—something retrieval alone cannot accomplish.

Beyond individual contracts lies organizational context: the legal playbooks spelling out which terms to accept, the procurement rules about spending approvals, and the sales policies on commission structures. Without this embedded knowledge, an agent cannot assess whether a contract meets the company's requirements or needs escalation.

Finally, user context—language preferences, past questions, and feedback—shapes how the system responds to each person.

Building Iris: Combining Multiple AI Capabilities#

Docusign's response is Iris, a contract AI engine within its Intelligent Agreement Management (IAM) platform introduced in 2024. Rather than betting everything on a single language model, Iris combines intelligent extraction, semantic search, agentic reasoning, and domain-specific logic.

Consider a procurement manager asking which software contracts worth more than $1 million need renewal in the next three months. Answering that seemingly simple question typically involved multiple people gathering data across spreadsheets, reconciling supplier names that appeared in different forms across agreements, and manually checking renewal dates.

Iris handles this by finding relevant contracts, identifying upcoming obligations, checking agreements against organizational playbooks, and flagging issues that require attention. This requires orchestrating multiple tools simultaneously—recognizing the supplier as a distinct entity, pulling all related agreements, accounting for amendments, and calculating net obligations.