Shield's Forge agents use GPT-6 Astra to work in IT portals that have no API
Shield says agents in its Forge platform now log into legacy portals and finish browser tasks for IT technicians. Phone calls, on-site work and remote-desktop tasks still need people.
Shield, which builds an AI platform for outsourced IT services providers, says its agents can now complete tasks inside older web portals that technicians would otherwise handle by hand. The agents use OpenAI's GPT-6 Astra to read the screen, click and type. According to an OpenAI customer story, these browser workflows are live and run thousands of times each month across Shield's IT providers.
The story is narrow, but it shows a specific boundary. Much of IT support runs through systems that were never built for software to use. When an agent cannot get into one of them, the work stops and waits for a person. Shield says that wait is now shorter for a set of supported tasks.
The access problem
For many small and medium-sized businesses, an outsourced IT services provider handles everything from a locked laptop to a misconfigured firewall. Shield's product, Forge, is described as an AI operating system for these providers. It brings together business data and customer context so that support tasks can be completed end to end.
The obstacle was access. Technicians spend much of their day in portals that manage tickets, devices and software licenses. Many of those systems lack APIs. Before GPT-6 Astra, OpenAI writes, Forge's agents could get stuck at systems they could not reach, and customers then waited for a technician to finish that step.
Shield had tried computer use with earlier models. Mayank Jain, Head of Technology at Shield, says they were not reliable or fast enough. "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," he says.
Jain frames the case in terms of speed. "If a customer is locked out of their computer or there's a phishing attack, speed is paramount," he says. "Whether you're going to solve their issue or you're going to escalate to a human, you want to do either as fast as possible."
How the work is divided
Inside Forge, a component called the AI Engineer assigns work to a computer-use agent. GPT-6 Astra then navigates the portal interface to complete the task, if it is one of the supported tasks. Technicians set the rules: they decide which actions the agent may take on its own and where it must stop for review or approval.
Jain describes the result as agents that "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."
The limits are stated plainly in the same account. Phone calls, on-site work and remote-desktop tasks still need human involvement. Shield says it is testing remote-desktop workflows that would let agents act directly on a computer's desktop, which would move the boundary further. That work is not yet live.
What Shield reports
The figures come from Shield and are published by OpenAI. They have not been independently measured.
- Jain says GPT-6 Astra was 35 to 40 percent faster on average than the models Shield had previously used for computer-use tasks.
- Shield reports faster execution, fewer failures and fewer timeouts.
- Its teams spend less time writing detailed instructions for the agent or guiding it through a task.
- Jain says customer satisfaction is higher on tickets where AI is used, and that speed of resolution is a critical factor.
The story does not give absolute numbers for failure rates, resolution times or satisfaction scores, and it does not say which specific portals or task types are supported. The volume figure, thousands of runs per month, is spread across all of Shield's IT providers.
What it means for work
The task being handed off is the click-through step: logging into a ticketing, device or license portal and carrying out a routine change that a technician already knows how to make. For IT support technicians at managed service providers, that removes waiting time and repetitive navigation, and leaves escalations, judgment calls, phone conversations and physical work with people. Technicians also gain a new task: deciding which actions an agent may take alone and reviewing the ones it may not. Shield also reports less time spent writing instructions for agents, which suggests that supervising the agent is itself becoming a smaller part of the job. What remains open is how far this extends once remote-desktop work is automated, and how providers handle errors made inside systems that were never designed to be operated by software.
Sources
This article was researched, written and illustrated by AI, and published without human editing. Spotted an error? Let us know.
More from māchine
All storiesGoogle has added computer use to the Gemini app for macOS. Paying US subscribers can hand it multi-step tasks, decide which apps it may touch and approve its key actions.
Yale researcher Julián Posada argues that platforms recruiting doctors, lawyers and PhDs to train AI pay well at first, then cut rates and drop workers, as they did in Venezuela and Kenya.
Microsoft offered voluntary exits to about a quarter of its workforce in October. Its CEO says the reason is AI, as a reorganisation of work around managing agents, not as a replacement for people.