OpenAI released ChatGPT Wallet this week, a payment feature allowing agents within ChatGPT Work to complete financial transactions directly. The wallet integrates with users' existing credit cards and fraud-detection systems, enabling autonomous agents to place orders, book services, and execute purchases without intermediation. Available across ChatGPT Work, Codex, and dots—OpenAI's suite of agentic tools—Wallet operates as a thin layer routing payments through existing infrastructure rather than creating new financial products. The feature addresses a growing friction point: as agents become more capable and reliable, explicitly preventing them from completing money-requiring tasks feels increasingly artificial.

The approval mechanism operates through category pre-approval rather than per-transaction prompts. Organizations set spending parameters—'auto-approve supplier invoices under $500' or 'IT hardware replacements up to $1,000'—allowing agents to operate autonomously within boundaries while maintaining human oversight for outliers. Consider IT automation firm Shield, which uses OpenAI to help technicians hand off work in legacy portals. With Wallet, an agent could autonomously order replacement hardware from approved suppliers, compare pricing, and route parts to inventory—completing real workflows without manual negotiation of deprecated interfaces. The stakes cut both ways: agents that cannot spend money struggle to complete operational tasks, but unsupervised spending poses obvious risks. The bounded-autonomy model seeks middle ground.

The launch arrives amid accelerating adoption of OpenAI models for autonomous business work. Amplitude built Wave, a proactive product agent identifying which features to ship next and enabling 1–2 day deployment cycles. Gondola reported 50% productivity gains using OpenAI's models. Holiday Extras is rethinking product development by turning customer expertise into better applications and improved booking conversions. These successes signal a market eager for agents with deeper operational autonomy. OpenAI's bet is pragmatic: the bottleneck isn't agent reasoning—it's financial constraint. Rather than waiting for marginally smarter models, OpenAI is betting that financial autonomy is what agents need now. Whether that's prescient or premature remains unclear, but the velocity at which the industry is adopting agentic workflows suggests OpenAI isn't taking chances.