OpenAI has launched “Operator”, a new mode within ChatGPT that aims to navigate web-based workflows as a human would. This development is part of a broader trend in AI-assisted task completion and represents a significant step forward in the practical application of language models.
Key Points
- Availability: Currently limited to US-based ChatGPT Pro subscribers, with plans to extend to lower subscription tiers and API access in the future.
- Functionality: Operator can perform tasks such as finding the best sauna, paying bills received by postal mail, and booking flights with budget airlines like JSX.
- Research Preview: While impressive, it’s important to note that this is still labeled a “research preview” and may have limitations or errors.
Collaboration and Potential Impact
OpenAI is partnering with several companies to ensure Operator addresses real-world needs while respecting established norms. These collaborations include: DoorDash, Instacart, OpenTable, Priceline, StubHub, Uber.
Additionally, OpenAI sees potential in improving accessibility and efficiency in public sector applications. They’re working with organizations like the City of Stockholm to streamline enrollment in city services and programs.
Implications for UI Accessibility and SEO
The development of Operator and similar AI-assisted navigation tools may have significant implications for UI accessibility initiatives. As adoption increases, it may be prudent for organizations to consider AI capabilities when designing user interfaces and digital services.
From an SEO perspective, this trend introduces a new paradigm that goes beyond traditional search engine optimization. As Ethan Mollick, Associate Professor at The Wharton School, points out, different AI agents show preferences for specific websites when performing tasks. For instance:
- Claude (with Computer Use capability) tends to use Yahoo Finance for stock prices
- Operator often relies on Bing search results, favoring the top-ranked sites
These preferences aren’t always predictable or consistent across different AI agents. Moreover, the reasons behind these preferences aren’t always clear, and they’re subject to change.
This development suggests a potential shift in how we approach SEO. It’s no longer just about ranking high in search engine results, but also about understanding and optimizing for AI agent preferences. This could lead to the emergence of a new industry focused on “Agent Optimization” – a concept that Mollick humorously acknowledges might become the next big thing in digital marketing.
For businesses and content creators, this means:
- Monitoring which AI agents prefer their sites for specific tasks
- Understanding why certain sites are favored over others
- Adapting content and structure to be more “agent-friendly”
- Staying agile as agent preferences evolve
As Mollick notes, “It is going to keep getting weirder.” This unpredictability adds another layer of complexity to digital strategy, requiring businesses to stay vigilant and adaptive in their approach to both UI design and content optimization.
The Broader Context
Operator is part of the larger “model forward” trend in AI development that shifts focus from the underlying models to practical applications and productivity gains. (Jaana Dogan of Google DeepMind recently suggested (perhaps half-jokingly) on X that AI/ML professionals should consider transitioning to software engineering roles.)
Real-World Application in Germany
Tom Braegelmann shared an example of using Operator to search for German court decisions, demonstrating its potential in legal research. While the process was still somewhat clunky (and the video was sped up by a factor of three), it showcased impressive resilience. When access to the Federal Court of Justice website was blocked, Operator autonomously switched to searching for the full text on OpenJur, illustrating its ability to adapt and find alternative solutions.
For this, he used VPN to work-around the geo-block in place. It’s worth noting that there are rumors suggesting OpenAI may take a strict stance against VPN usage, potentially including account suspensions. However, it’s unclear whether such actions are targeted specifically at VPN use or are collateral effects from addressing misuse by some VPN users. Tom didn’t provide specifics about the type of VPN used (whether a private, law firm VPN or a commercial VPN service typically used for bypassing geo-restrictions).