Shield Hands Legacy Portal Work to OpenAI-Powered Forge Agents

OpenAI's November 13 customer story describes how Shield has moved work inside legacy portals onto Forge agents, freeing IT technicians from screens that demand clicks, typing and navigation but little judgment.

As the story frames it, the agents operate inside existing portal systems and carry out steps that would otherwise be handled by hand, while the technicians move to the issues that actually require their judgment. That division of labour is the central claim of the piece.

Legacy portals are a familiar problem in enterprise IT. The systems are old, APIs are missing or incomplete, and the only route to automation runs through the interface itself. Having an agent operate that interface the way a person would is one practical way to bring those existing systems into an automated workflow.

The case study does not name the specific models used, the agents' accuracy, or how widely they have been deployed, and it offers no quantified time savings. What it presents is a way of splitting work rather than a reproducible benchmark.

The interesting question is where that split sits. The story's answer is that judgment stays with people, but exactly where the line falls depends on each organisation's process auditing and on who is accountable when an agent misclicks.

For buyers, the value of a case like this is a concrete shape to compare against. What still determines success is the engineering around it: permissions, audit trails, and how exceptions get routed back to a human.