Shield Forge and OpenAI GPT-6 Astra: Automating Legacy IT Portals

OpenAI’s customer story on Shield describes how Shield’s Forge platform uses GPT-6 Astra to let computer-use agents complete supported browser tasks in legacy IT portals that lack APIs. The practical takeaway is that AI agents can remove a common IT-support bottleneck—manual portal access—while technicians retain control over which actions are allowed and where review is required.

What Shield and Forge do

Shield built Forge, which OpenAI describes as an AI operating system for IT services providers. Forge brings together business data and customer context to help providers complete IT support tasks end to end. The platform is aimed at the everyday work of outsourced IT teams serving small and medium-sized businesses, from an employee locked out of a laptop to a misconfigured firewall that took an office offline. Forge’s agents also need to operate the older portals technicians use daily, which is where GPT-6 Astra enters the workflow.

The bottleneck was access to legacy systems

The constraint Shield identified was access, not just task knowledge. Before GPT-6 Astra, Forge’s agents could get stuck at systems they could not access, and customers had to wait for a technician to complete that step. Shield had tried computer use—where a model reads the screen, clicks, and types—with earlier models, but Jain says those models were not reliable or fast enough. Many of the affected systems lack APIs and are difficult to use even for humans.

“The only time you should have to bring in a human technician is if you need their judgment, not because you’re blocked on access.”

— Mayank Jain, Head of Technology, Shield

Jain also described why legacy portals are a hard target for computer-use models: “Usually, a system that doesn’t have an API is an old-school system that also doesn’t look very nice and is very hard to use, even by a human.”

How GPT-6 Astra works inside Forge

Forge’s AI Engineer assigns supported work to a computer-use agent, and GPT-6 Astra navigates the portal interface to complete it. The model logs into portals and legacy websites that may not offer API access, performing tasks that Shield says only a human could do six months earlier. Technicians define which actions the agent may take on its own and which require review or approval.

Shield says these browser workflows are live, with thousands of runs each month across its IT providers. The automation is deliberately scoped: GPT-6 Astra handles supported browser tasks, while phone calls, on-site work, and remote-desktop tasks still need human involvement.

“GPT-6 Astra allows our agents to embody what a technician does by logging into portals and legacy websites that might not have API access and performing tasks that only a human could six months ago.”

— Mayank Jain, Head of Technology, Shield

Reported results: faster runs and fewer failures

Shield reports that GPT-6 Astra was 35-40% faster on average than the legacy models the company had used for computer-use tasks. Jain also says customer satisfaction is higher on tickets where AI is used, with speed of resolution cited as a critical factor. Shield reports faster execution, fewer failures, fewer timeouts, and less time spent writing detailed instructions for an agent or helping it through a task.

These are customer-reported outcomes rather than an independent benchmark. The source does not provide a test methodology, sample size, or task-level breakdown for the 35-40% figure. What it does provide is a production-usage signal: Shield says the browser workflows are live and run thousands of times each month.

“What shocked me the most was how natural GPT-6 Astra was with computer use. We’d seen models perform computer use before, but you could clearly see them stutter as they try to figure out how to navigate the UI. With GPT-6 Astra, it feels like you’re just watching a human click around.”

— Mayank Jain, Head of Technology, Shield

Human oversight and the next workflow

Shield’s deployment keeps humans in the loop by policy and by task type. Technicians decide the agent’s allowed actions and review or approve work where needed. Phone calls, on-site work, and remote-desktop tasks remain human responsibilities.

Shield is testing remote-desktop workflows that would let agents perform more complex actions directly on a computer’s desktop. Jain frames the result as an extension of the technician rather than a replacement: “With GPT-6 Astra, Forge is now truly an extension of our human technicians.”

Why the Shield case matters

The Shield customer story shows a concrete use for computer-use agents in IT services: automating access to legacy portals that never exposed APIs, while preserving human judgment for exceptions and approval. For IT providers, that targets a specific source of delay—waiting for a technician to log in and finish a browser step—rather than the entire support process. The source is a vendor customer story, not a technical paper, so it does not detail model architecture, security controls, or evaluation design. Its central claim is operational: GPT-6 Astra makes supported browser workflows reliable and fast enough for Shield to hand them to agents in live production.

Sources