The Emergence of Accountability in Agentic Contract Management
The introduction of agentic artificial intelligence in contract management marks a significant shift in how agreements are negotiated and executed. Software is evolving beyond merely drafting clauses or identifying risks to actively negotiating and making decisions within the workflows where commitments are established.
This transition raises important questions regarding accountability when agents act on agreements. For instance, in a scenario where two agents transact under a master agreement, if one agent places an order and the other accepts it but the goods do not arrive or the pricing is incorrect, how can one prove that a contract was formed? Currently, this requires log files and an understanding of the system's configuration, which does not align with the expectations of commercial parties.
Audits rely heavily on internal controls surrounding contracting processes. These controls ensure that the individual committing the company had the authority to do so, that the commitment adhered to established rules, and that the details were accurately recorded in the relevant systems. Each of these controls presumes that a human signed off on the agreement. The challenge arises when an agent acts autonomously; it becomes unclear where the accountability lies.
A partial solution to this issue has been implemented through Docusign’s eSignature solution, which provides a certificate of completion alongside executed agreements. This certificate documents the signing process, detailing who acted, when, and in what order, while ensuring that any later alterations are detectable. Although this does not encapsulate the substance of the agreement, it does offer proof of how the agreement was executed.
The trust in an agreement is fundamentally linked to the presence of a signature, which signifies a commitment from a party that can be held accountable. The signature serves as an evidentiary element, confirming who acted, their authority to bind the party, the integrity of the agreement, and the timing of the commitment. The moment of signing delineates the transition from negotiation to obligation.
In contrast, agentic actions lack distinct markers. An agent's operations consist of a continuous flow of actions, with no clear separation between steps such as filling in a field or accepting liability. The critical question is determining when an agent's action constitutes a decision rather than just a step.
Once decisions can be identified, additional requirements emerge. Each decision must be traceable to the authority under which it was made and the information that informed it. Furthermore, these decisions must compile into a durable, attributable record that can be relied upon by a person. This process involves recognition, traceability, and record keeping, each building on the previous stage.
It is essential to embed accountability into the workflow from the outset. Conducting audits post action is akin to archaeology, piecing together a narrative from whatever evidence remains. In contrast, designing accountability into the workflow allows for proactive management of actions and decisions.
As organisations increasingly integrate agents into their agreement processes, it is crucial to establish accountability from the beginning. This includes defining the scope of an agent's decision making authority, ensuring traceability of those decisions, establishing clear points for human oversight, and maintaining a comprehensive record of actions and their justifications.
At Docusign, the belief is that trust must be integral to every aspect of an agentic workflow. This involves combining human oversight with transparent audit trails that elucidate how agents acted and where human intervention occurred. As agentic AI assumes a more prominent role in contract management, these safeguards will be vital in fostering confidence among businesses in leveraging this technology.