OutSystems' Agent Workbench Makes Customising AI Agents Easy
OutSystems has officially launched Agent Workbench, a new platform designed to bring agentic AI to the masses.
In recent years, the company has become increasingly renowned for its low-code AI tools that allow businesses to customise and create AI models without the help of specialist engineers.
Agent Workbench, the general release of which was announced at OutSystems' annual conference in Lisbon this week, applies this same logic to agentic AI, which has been one of the breakout technologies of 2025.
With the help of AI agents, companies can effectively multiply their workforce through the miracle of automated digital labour.
“Organisations are excited by the promise of AI and agentic systems, but are struggling with endless pilots and ungoverned tool sprawl while the business impact stalls,” explains Woodson Martin, CEO of OutSystems.
“Legacy systems, siloed data, fragmented AI tools and complex AI development cycles are slowing progress.
“We built Agent Workbench to make it possible to unlock custom agents as a true business enabler, not an experiment,” he says.
Testing Agent Workbench with major companies
Axos Bank participated in the early access programme for Agent Workbench and is still using the platform today to automate log analysis and document processing tasks.
"We plan to expand the use of AI capabilities within OutSystems to realise immediate gains, without needing to invest in specialised AI roles," says Kevin Hearn, SVP and Head of Consumer Bank Development at Axos Bank.
Axos is also building agents for analysing error logs and automating data entry from documents.
Thermo Fisher Scientific was another major firm to test Agent Workbench in those early days.
With OutSystems' new programme, the organisation deployed a customer escalation agent to interpret unstructured data from customer interactions and eliminate manual triaging processes.
Elsewhere, Arch Company, the UK's largest small business landlord, rolled out a content classification agent that has helped it to route customer service enquiries without manual intervention.
These trial runs allowed OutSystems to show major clients what its technology could do, but it also gave the team early feedback from companies with massive operations which was likely an invaluable sort of road testing.
Addressing governance and integration challenges
For Woodson, the launch of Agent Workbench is a response to the widespread struggles that most enterprises have had with the deployment of agentic AI so far.
The company cited research indicating that 93% of organisations are prioritising AI agent development but face challenges around governance, security, integration and scalability.
"Agentic AI is one of the most powerful levers for enterprise innovation and transformation, empowering organisations to grow revenue, drive operational efficiency and deliver amazing customer experiences," he explains.
With this new platform, Woodson and co. hope that businesses will be able to streamline the whole process, unifying their data, workflows and intelligent agents into one trusted platform.
Technical capabilities and model support
The generally available version of the Workbench includes an agent marketplace and Model Context Protocol support, allowing agents to access enterprise systems and external tools directly.
Agent Workbench supports models from AWS Bedrock, Azure OpenAI and Anthropic, as well as newer LLMs including Gemini, Cohere, Mistral, Databricks, AI2 and IBM's watsonx.
The platform also accommodates custom-built models on VertexAI and open-source models on HuggingFace, making integration with any existing systems rather simple.
Enterprise use cases emerge
TeamWork, an international consulting group, built a multi-agent system that provides support teams with real-time guidance whilst automating routine ticket resolution.
The company anticipates that a notable share of tickets will be resolved automatically, with resolution times for complex cases expected to decrease considerably.
Grihum Housing Finance in India is deploying the platform to improve loan underwriting accuracy by analysing property evaluation reports and suggesting technical property deviations.
KPMG's Low-Code Centre of Excellence has been involved in testing the platform during its early access phase.
"Agent Workbench delivers the speed-to-value and governance guardrails that organisations have come to expect from OutSystems development," says Hélio Pimenta, Associate Partner at KPMG's Low-Code Centre of Excellence.