OpenAI's Dots Agents Enter the Field: Early Impressions and Infrastructure Implications

OpenAI has begun rolling out Dots, a new class of always-on AI agents designed to execute multi-step web tasks such as shopping, booking, and form completion. Unlike standard chat sessions, these agents persist in the background, maintain context across days, and can proactively surface results.

In a hands-on evaluation, the agent—named "Toolie" during testing—demonstrated the concept's promise but also its current fragility. It misidentified the user, struggled with CAPTCHA challenges, and exhibited inconsistent voice synthesis. The agent did successfully filter furniture listings by door-frame dimensions and budget, showing that the underlying browser-control loop functions when not interrupted by anti-bot measures.

Why it matters for GPU / AI infrastructure

Persistent agents change the compute profile dramatically. Instead of bursty inference per chat turn, Dots require continuous availability, long-context retention, and repeated browser-environment interactions. That translates to sustained GPU occupancy, higher memory footprints for session state, and new scheduling priorities for cloud providers. At a $100 monthly price point, OpenAI is signaling that the unit economics of always-on agents depend on efficient, high-utilization GPU fleets—exactly the workload profile AiGpu's infrastructure targets.

Security architecture also shifts. Granting an agent access to email, payment, and identity stores demands isolated execution sandboxes, fine-grained permissioning, and audit trails—capabilities that must be baked into the hosting layer, not bolted on afterward.

Meta's competing Muse agent is free, suggesting a coming price war that will pressure inference costs downward. Providers who can deliver low-latency, high-throughput GPU capacity at scale will set the margin floor for the entire agent ecosystem.