Thank you so much to everyone who attended our recent Bengaluru developer meetup! It was great being able to talk about some of IAM’s capabilities, the direction we’re headed in the agentic era, and to give attendees hands-on experience and the ability to ask questions. The meetup was a huge success and we plan on doing more in the future.

Below are the key takeaways we learned from the meetup. If you have any questions, comments, feedback, or suggestions for future meetups, we would love to hear from you in the replies.

Key Takeaways

As Docusign developers, it’s likely that you have hundreds of agreements sitting in a repository somewhere. You might have renewal dates, payment terms, and risk clauses buried in PDFs that aren’t easy to query. The agentic-era promise of "just ask AI what's renewing in 90 days” sounds great, but we know that agents are only as good as their data.

The focus of this meetup was to address this gap between vision and implementation, and give attendees hands-on experience with some of Docusign’s IAM solutions.

The real agentic challenges (and how to solve them)

The data layer

Your agreements need a consistent model so that existing documents can be ingested and structured at scale. Through the IAM Toolkit and Docusign CLI, you can model your agreement data, build your ingestion pipeline, and validate that agents can query it reliably before you build anything conversational on top.

Once that foundation exists, you’ll need to define what actions an agent can take and what it shouldn't do without a person signing off. This is where the Docusign MCP Server can help you create an agent that queries across your entire agreement repository while operating within your set boundaries.

In the MCP workshop, we walked through modeling procurement agreement data with the CLI, ingesting and verifying extractions in Agreement Manager, then connecting the Docusign MCP Server to Microsoft Copilot Studio to build an agent that answers natural language queries about renewals and risk.

API design for agents

Agents don't read documentation or guess what an error means the same way that we do. When software is the consumer, APIs need to move beyond rigid developer-centric design toward clear semantic metadata, bounded execution environments, and scoped permissions.

The Docusign MCP Server is built for this: agents can discover available actions, understand what they return, and operate within defined boundaries without a human interpreting responses along the way.

The Building for the Agentic Era panel dugs into this with perspectives from Docusign, Google, and IBM on what API design looks like when agents are the primary consumer.
 

Workflow orchestration

Most agreement processes start in a CRM or ERP, not an agreement tool. The challenge is letting users stay in their system while agreement steps happen behind the scenes.

With Workflow Builder and integrations like Salesforce Agentforce, you can trigger the agreement process from the originating system, pass context forward, and let Docusign handle review, compliance checks, and signature prep.

The Agreement Automation workshop showed how to configure Claude via the MCP Server to kick off agreement workflows through natural language prompts, with Agreement Desk managing multi-stage approvals across legal and finance.

Human oversight

Balancing the "human in the loop" problem is more complicated than it initially seems. If you automate everything, you'll get burned on edge cases. If you require approval on every step, nobody will use it.

The solution is to classify your actions by consequence level. If you have a routine step, those can be automated through Workflow Builder. High-stakes decisions (sending a contract, escalating unusual terms, etc) get explicit checkpoints using AI-Assisted Review and scoped agent permissions.

The Agreement Automation workshop also showed this in practice with AI-Assisted Review flagging unusual terms for human decision during the approval flow.

Want to learn more?

Developers and architects worked through all of these patterns hands-on at the meetup. The session recordings walk through real-world examples of using IAM to solve complex agentic problems.

Watch the session recordings (keynote, panel, workshops)

Read my deep dive blog post

Visit the Docusign Developer Center

What's your experience?

Are you working through any of these challenges? What's been the hardest part? Would love to hear what's working and what's not.