Home

Blog

Building the Foundation for Agent-Ready SaaS Platforms

AI Agents & Orchestration

Customer Stories

Embedded Integration

Takeaways from a WOW 2026 Embedded session: What BDO, Upstart, and AMN Healthcare built first, and what they learned along the way.

Haillie Parker

Oct 6, 2026

AI capabilities alone don’t guarantee a competitive edge. In fact, they’re expected. The advantage goes to the team with the best orchestration layer underneath it.

That’s easy to nod along with but harder to act on, especially if you’re under pressure to “add AI.” We asked three Workato Embedded customers what this looks like in practice. Their answer: a strong foundation before agents entered the picture, and the speed it’s buying them now.

This was one of many hot topics that hit the Vegas stage at World of Workato (WOW) 2026, Workato’s annual user conference for enterprise orchestration and AI. In this session, Building the Foundation for an Agent-Ready Future, Josh Medina, VP of Embedded Sales at Workato, sat down with three product leaders from fintech, healthcare staffing, and professional services. They landed on the same idea: AI is only as useful as the systems, data, and workflows it connects to.

Nick Bessinger, Senior Manager, Identity Access Management & Governance, BDO: centralized user provisioning as the firm grew from 2,800 to 15,000 people.
Swetha Gajam, Principal Solutions Architect, Upstart: standardized integrations using Workato for an AI lending marketplace.
Justin Harmston, Senior MDM Developer, AMN Healthcare: uses Workato to connect with clients’ HR systems, and is now building toward MCP and agents.

Here are their takeaways.

Learn more about building a strong orchestration foundation with Workato ONE for SaaS platforms.

Takeaway 1: Start with the process, not the model

Before Upstart automates anything, the team gets specific about what needs to happen.

“Finalize your requirements. Delete everything that’s unnecessary, and only then automate.” — Swetha Gajam, Upstart

In regulated work, the people fixing a problem need to know what was decided and where the data went, especially when AI helped build the automation. This clarity is also what let Upstart standardize its lending partner integrations andcut onboarding from three to six months to four weeks.

Nick took the same approach at BDO, rebuilding onboarding, role changes, and offboarding on a whiteboard as if the old process never existed. The first call was the trigger, the event that starts the workflow. For offboarding, BDO’s biggest identity risk, HR often logs a departure weeks before the last day, which would cut off access too soon. Settling it took about a month and a half, before any connectors were built.

For any team shipping customer-facing integrations, the lesson carries over: every integration is a process before it’s a build.

Takeaway 2: Reliable agents run on deterministic tools and governed access

When Josh asked where AI is already live at AMN, Justin started with what leadership wants: AI that makes people’s jobs easier, not AI that replaces them. Then he got specific about what “useful” means.

“I’m not so much into the number of agents as the quality. If you have an engine that everybody’s using and can’t get enough of, that’s far more valuable than 100 niche ones that no one really cares about.” — Justin Harmston, AMN Healthcare

His first attempt, connecting an AI agent to AMN’s vendor management system through its API, exposed the gap: an agent can report that it placed an order when it didn’t.

The fix was Model Context Protocol (MCP). He packages a deterministic workflow as the tool, so the agent only handles what to do (“duplicate this order”) and the workflow dictates how, the same way every time. The result is far lower token usage, and as he put it, “I don’t have to worry about [the agent] making mistakes.”

Working with Claude and Workato’s connector SDK, he built a custom connector with about 100 actions in one day. Next on the roadmap, directing a solution like AIRO at the same challenge – Workato’s solution for building and shipping agents across systems from plain language.

Reliable actions are only half the job. You also have to consider what an agent can reach. At BDO, Nick’s team asks the same question of every tool: who is this, and what can it reach? Anything with AI built in goes to a review board of legal, identity and access management, security, and data leadership.

“If you’re rolling out agents and don’t know what they’re doing, you don’t know what access they have.” — Nick Bessinger, BDO

BDO won’t approve a SaaS app just because it says it has AI. For software teams, that’s the buyer’s view of your AI roadmap.

Takeaway 3: Build the team alongside the technology

The foundation these companies built wasn’t just technical. Each one invested in the people who would run it:

As one of BDO’s AI champions, Nick aims to train 25 colleagues in the next two months, so that everyone ends up with a personal agent.
Justin’s team takes a similar instructional approach. They teach non-developers what’s required to propose an AI idea, so good ones arrive prepared and don’t get shut down.
At Upstart, Swetha invested in expertise first, bringing in industry experts and the core banking providers to learn how those systems work before building anything.