From Automations to Agents: The Mumbai Pathfinder Recap

We had a great time at the Mumbai meetup, co-hosted by our Community Captain, Alpesh Sexana. If you were there — thanks for making it worth it. If you missed it, here's a quick recap of what we covered. The session had two parts: a product deep-dive into where agentic automation is heading, and a hands-on where we actually built something together.

The Product Session — What Is Agentic Process Automation?

Before diving in, our Principal Product Manager, Vineet Pujari got the basics straight — because this part matters more than most people realise: what's the difference between a bot, a process, an AI agent, and agentic process automation?

Here's the breakdown we walked through:

• Automations are your classic task-level bots. They do one thing — extract data, fill a form, move a file. Deterministic, linear, reliable. No surprises.
• Processes are when you string those automations together into an end-to-end workflow. Think invoice processing or employee onboarding — multiple steps, maybe a human approval in the middle, defined paths for when things go wrong.
• AI Agents are a step change. Instead of following a fixed script, they reason about what to do based on context. They can call tools (other automations, APIs, even other agents), handle variability, and adapt on the fly.
• Agentic Process Automation (APA) combines all three. It's the intelligent orchestration of agents, automations, and humans to run complex, cross-functional business processes — mixing the reliability of deterministic automation with the flexibility of AI.

Automations are like machines on an assembly line. Agents are like the foreman who knows when to stop the line, reroute work, or call in help.

How Our Platform Supports This

We looked at how Automation Anywhere is building toward this vision. A few things stood out:
 

• Mozart Orchestrator is the layer that coordinates work across systems, users, and agents. It's already got 700+ customers using it, with 4,000+ processes created last month alone.
• AI Agent Studio lets you build custom process agents. Combined with tools like Prompt to Automate, that lets you build automations from natural language, the AI Agent Studio dramatically lowers the barrier to building agents.
• Automation Co-Pilot brings agents into the flow of work — inside browsers, desktops, or as a conversational interface for end users.

On the governance side (which often gets skipped in demos but matters a lot in production), we walked through:

• Guardrails — real-time content policies, PII/PHI masking, regulatory compliance
• Observability — audit trails for every tool call an agent makes, dashboards for runtime performance
• AI Evaluation — automated scoring of agents pre- and post-deployment, no ML expertise required. You can even generate synthetic datasets for testing.

Trust & Evals — Why It Matters in Production

Deploying an agent is one thing. Knowing it's doing the right thing is another. This is where Trust & Evals comes in. The platform has a built-in evaluation framework that lets you assess both agents and the models they run on — before deployment and while they're running in production. No ML expertise needed; it's designed for developers who want answers, not another system to learn.

• Automated scoring — agents are evaluated against pre-defined metrics, with results delivered in hours rather than days
• Activity traces — every tool call, input, and output is logged so you can see exactly what the agent did and why
• Synthetic dataset generation (on the roadmap) — the platform can generate its own golden datasets for testing, which removes a huge manual overhead
• Scheduled evaluations — you can set ongoing checks so agents are continuously monitored, not just tested once at launch

The roadmap for FY27 is focused on pushing task completion rates up, cutting workflow failures, and expanding multi-agent execution.

Community Spotlight — Akbar's Lab Report Severity Check Agent

One of the highlights of the meetup was a demo by Akbar Shaikh, our Community Captain and RPA Principal Architect at Novatio Solutions — who also took home the Community Favourite title at the Agentic Bounty challenge. Akbar built a Lab Report Severity Check AI Agent that tackles a very real problem in healthcare: doctors starting their day with 200+ lab reports, most routine, a few critical, and no fast way to tell which is which.

The agent automates the triage — retrieving reports via API, classifying severity using AI, auto-generating clinical notes, and routing only the severe cases to doctors for a Human-in-the-Loop review. The result: critical cases flagged in minutes instead of hours, with a 60–70% reduction in manual effort. What made it land in the room was the closing line: we're not replacing doctors — we're giving them their time back.

The Hands-On — Building a GTM Agent

Sales reps spend a lot more time on admin than anyone likes to admit — logging leads, updating CRM records, chasing follow-ups. In this hands-on session, attendees stepped into that world and built an AI agent to fix it. In this session led by Sejal Dhondkar, Sales Engineer and Pratyush Priyadarshi, Solutions Architect, using Automation Anywhere's AI Agent Studio, participants automated a real sales workflow end-to-end:

• Lead Capture — Extract lead data from emails or web forms and auto-populate the CRM. No duplicate entry.
• Deduplication — Built-in checks identify and merge duplicate records, keeping data clean.
• Follow-Up Reminders — The agent schedules timely nudges so no prospect slips through the cracks.
• Process Standardization — Consistent workflows mean every lead is handled the same way, every time.

What made the session click was the developer experience itself — low-code tools, drag-and-drop components, and prebuilt templates meant participants could build, test, and deploy a working agent without getting lost in setup. Real-time dashboards tied it all together.

The takeaway: with features like AI Agent Studio and Mozart orchestrator, the agent development life cycle has dramatically reduced. Developers can now prototype agentic automation and ship quickly, with immediate impact on sales operations.

What stood out wasn't any single feature or demo — it was how many different starting points were in the room, and how quickly the conversation got real. People weren't asking if agents are useful. They were asking how to govern them, how to test them, how to know when to trust them. Those are the right questions. We'll be back with more — watch this space!

Once again, a big thanks to Alpesh Saxena, our Community Captain, for co-organising this meetup and making it happen! If you would like to be a part of and contribute to our upcoming meetups, reach out to us.