Platform vendor Forge now delegates navigation tasks to GPT-6 Astra-assisted agents on systems without an API. Shield reports large-scale executions, acceleration over its previous models, and continued human intervention for out-of-browser operations.
Speed gains and fewer failures, but humans keep whole areas
Mayank Jain, chief technology officer at Shield, says GPT-6 Astra performed on average 35 to 40 percent faster than the legacy models Shield uses for IT usage tasks. It also reports higher customer satisfaction when AI intervenes, with speed of resolution being a determining factor. Shield reports faster execution, fewer failures and reduced wait times, and says its teams spend less time writing detailed instructions to guide an agent. According to the company, these browser-based workflows are in production with thousands of executions each month across vendor partners. However, telephone calls, on-site interventions and remote desktop operations remain the responsibility of technicians.
What agents do in portals without API and how they are supervised
Many tools used by technicians do not have an API, requiring on-screen navigation similar to that of a person. Technicians spend a large part of their day in these portals, including managing tickets, devices and software licenses, and Forge agents must interact with these same legacy systems. In Forge, the AI Engineer assigns work to an IT user agent, and then GPT‑6 Astra takes care of navigating the interface to perform the supported tasks. Technicians define which actions the agent can perform autonomously and which actions require review or approval.
Each output (GPT, Claude, Gemini, Mistral, etc.) deciphered the same evening, in 5 minutes. Free.
Included upon registration: our selection of the best AI guides & comparisons.
Choose your pace
Free · No spam · Unsubscribe in 1 click
Astra gets around the access bottleneck that held back previous models
Prior to the adoption of GPT-6 Astra, Forge agents encountered inaccessible systems, forcing customers to wait for a technician to respond. Shield had tested approaches based on screen reading, clicking and typing with earlier models, which Mayank Jain found to be insufficiently reliable and too slow. He points out that systems without APIs are often old and difficult to use, even for a human. Now, supported browser tasks are delivered to agents through GPT‑6 Astra, with the goal that the call to a human technician arises from a need for judgment, not an access issue. Jain points out that, in situations such as account blocking or a phishing attempt, speed of action remains essential.
Continuous deployment and upcoming remote desktop tested capabilities
Shield describes browser-based workflows already in use with thousands of monthly executions. Mayank Jain believes that GPT-6 Astra allows agents to take on technician tasks by connecting to legacy portals and sites without an API, something that was only possible for a human six months ago. He also affirms that the use of the computer is more natural than with old models, which showed hesitations in the interface. At the same time, Shield is testing remote desktop scenarios so that agents can perform more complex actions directly on a workstation. According to Jain, Forge thus becomes a true extension of human technicians. This approach targets the needs of SMEs usually supported by external providers, and is based on Forge, presented as an AI operating system bringing together business data and customer context to manage end-to-end IT support tasks.

