Outsystems - An introduction to the Agent Workbench

From Clicks to Cognition: The New AI Frontier

Everyone in tech is scrambling to integrate AI into their business, but building with AI is the easy part. The real challenge is deploying it in a way that’s controlled, secure, and doesn't create a massive management headache. It's one thing to connect to a powerful language model, but it’s another thing entirely to weave it into critical applications with the right guardrails. A recent session led by experts from Providit offered a look at how this problem is being solved, shifting low-code platforms from simple app builders to sophisticated command centers for AI.

‍

Setting the Stage: An Evolved Platform

Kicking things off, Robin van Staveren from Providit, a firm specializing entirely in Outsystems, gave some background. Outsystems isn't new to the scene; the Portuguese company started in 2001 as a pure-play low-code platform. Now, nearly 25 years later, Robin explained that it's evolved into a platform built for the AI era. The goal is to enable companies to build their most critical applications with powerful AI working under the hood, all while keeping governance and security front and center.

Next, Tim Timperman jumped into the technical specifics. He outlined how the modern Outsystems platform, now called the Outsystems Developer Cloud (ODC), works as a complete ecosystem. It’s split into two main parts: the ODC Studio, the IDE where developers build their applications, and the ODC Portal, a centralized web interface for managing everything else. The portal is the key here. It handles deployments, versioning, monitoring, and, most importantly, the new AI Agent capabilities. It's a modern, container-based system built for the cloud. But what does that look like in practice?

‍

Practical Application: AI Agents at Work

Tim walked through a common business use case: a bank’s loan application process, now powered by a team of AI agents. It started on the customer-facing side with a simple form to apply for a personal loan and upload documents like a pay stub, bank statement, and tax form.

Once that data was submitted, the process moved to the back office. On a manager’s dashboard, the new loan request appeared, and a team of AI agents automatically got to work. Tim showcased four distinct agents working in concert:

1. Intake Agent: Its job was to verify that all the required documents were present and correctly matched the application. In the demo, it immediately flagged that the wrong documents had been uploaded.
2. Enrichment Agent: After a human (the "human-in-the-loop") approved the initial documents, this agent was triggered. It worked in the background to pull additional third-party data on the applicant.
3. Underwriter Agent: Using all the collected and enriched data, this agent analyzed the customer's financial profile. It noted a history of payment delays and a low credit score, then generated an initial recommendation to reject the loan.
4. Communicator Agent: This final agent took the underwriter’s decision and automatically drafted a clear, professional email to the customer explaining the outcome.

Tim pointed out that this whole workflow was built and managed visually. Every step, agent call, and decision was logged and monitored right inside the ODC Portal. This gives teams total transparency to track how the AI is performing, debug issues, and understand exactly why a decision was reached, without writing a single line of custom monitoring code.

‍

Building from Scratch: Your Own AI Agent

The presentation then pivoted to show just how quickly a new AI agent can be built. The process in Outsystems is template-driven and designed to be visual.

It starts with defining the agent’s core components. First is the Grounding Data, where you connect the agent to your internal databases or external APIs. This ensures its responses are rooted in factual, up-to-date information. Next, you Build the Prompt, which involves writing a system message to define the agent's role and personality, like “You are a helpful AI loan underwriter for our bank.”

From there, you connect your agent to an AI Model and give it Tools to perform specific actions, like calling an external function to check the weather. The platform is also rolling out Agent Guardrails, a new beta feature that helps prevent common issues like prompt injection attacks or getting stuck in loops. Once the agent is designed, you just hit publish. The platform compiles it, packages it into a container, and deploys it as a reusable function that can be called from anywhere. The entire process condenses a task that could take days or weeks into just minutes.

‍

The Future is Composable AI

The demonstration made it clear that the conversation around low-code is changing. It's no longer just about building simple apps faster. Platforms like Outsystems are aiming to provide a structured and transparent environment for building, deploying, and managing entire teams of AI agents. By offering visual workflows, built-in monitoring, and simple templates, the system handles the heavy lifting of orchestration and security. This allows development teams to stop worrying about the complex plumbing and focus instead on the business logic that delivers real value.

‍

Ready to empower your applications with AI? Contact us for more inspiration and to see how we can help