Major artificial intelligence firms Anthropic and OpenAI have unveiled a new generation of products designed to transform how people interact with AI. The new approach moves beyond simple conversational chatbots, introducing systems where users manage teams of specialized AI agents that work together on complex tasks.
This strategic shift positions the user not as a simple prompter, but as a supervisor overseeing a digital workforce. The simultaneous releases signal a significant industry trend, but also raise questions about the future of software and knowledge work, a concern reflected in recent market volatility.
Key Takeaways
- OpenAI and Anthropic have both launched products centered on managing multiple AI agents simultaneously.
- This model changes the user's role from a direct operator to a supervisor of AI teams.
- Anthropic released Claude Opus 4.6 with an "agent teams" feature, while OpenAI launched its enterprise platform, Frontier.
- The concept of AI performing complete workflows has contributed to significant financial market anxiety, with software stocks experiencing a major sell-off.
- Both companies also released more powerful AI models to support these new agent-based systems.
A New Paradigm: From Conversation to Delegation
The fundamental way we use AI is undergoing a major evolution. For the past few years, the dominant model has been a one-on-one conversation: a user types a prompt, and a single AI model provides a response. Now, leading AI labs are betting on a different future—one based on delegation and supervision.
The core idea is that instead of asking one AI to do everything, a user can assign a complex goal to a system that deploys multiple AI agents. These agents can then divide the labor, work in parallel, and coordinate with each other to complete the project. The human's role shifts to that of a project manager, monitoring progress, providing direction, and intervening to correct errors.
This approach is designed for complex tasks like reviewing an entire software codebase or analyzing large financial datasets. While the marketing suggests autonomous digital coworkers, current technology still requires significant human oversight to ensure accuracy and prevent mistakes.
What Are AI Agents?
In this context, an AI agent is more than just a chatbot. It is a system designed to pursue a goal autonomously. It can break down a task into smaller steps, use tools (like browsing the web or accessing a database), and make decisions to achieve its objective with minimal human input for each step. The new platforms allow multiple agents to work together.
OpenAI Introduces 'AI Co-Workers' with Frontier
OpenAI has made a significant move into this new territory with the launch of Frontier, an enterprise platform it describes as a way to "hire AI co-workers." This system is built to integrate directly into a company's existing infrastructure, connecting to customer relationship management (CRM) tools, ticketing systems, and data warehouses.
Each AI agent within Frontier is given its own identity, a set of permissions, and a memory to maintain context over time. This allows them to perform tasks that people typically do on a computer, from managing customer support tickets to analyzing sales data.
"What we’re fundamentally doing is basically transitioning agents into true AI co-workers," said Barret Zoph, OpenAI’s general manager of business-to-business.
Alongside Frontier, OpenAI also released a new macOS desktop application for its coding assistant, Codex. The app functions as a "command center for agents," allowing developers to run multiple AI threads that work on isolated copies of a codebase. This entire ecosystem is powered by a new model, GPT-5.3-Codex.
New Model Sets Performance Benchmark
OpenAI claims its new GPT-5.3-Codex model is a significant step forward for agentic capabilities. The company stated that early versions of the model were used internally to accelerate its own development, helping engineers debug training runs and diagnose test results.
On the agentic coding benchmark Terminal-Bench 2.0, GPT-5.3-Codex achieved a score of 77.3%, setting a new high mark for performance in this area and outperforming its direct competitors.
Despite the powerful capabilities, OpenAI's CEO of Applications, Fidji Simo, suggested the goal is not to replace existing software entirely. "Frontier is really a recognition that we’re not going to build everything ourselves," Simo told reporters, framing the platform as a layer that works with existing tools.
Anthropic's Answer: Claude Opus 4.6 and Agent Teams
On the same day as OpenAI's announcements, Anthropic launched its own powerful new model, Claude Opus 4.6, along with a new feature in its coding environment called "agent teams." This feature allows a developer to initiate multiple AI agents that autonomously split a task, coordinate their efforts, and run their work concurrently.
The interface presents a split-screen terminal environment where a developer can switch between agents, monitor their progress, or take direct control of any single agent while the others continue their work. Anthropic suggests this is ideal for tasks that can be easily parallelized, such as large-scale codebase reviews.
The new Opus 4.6 model brings substantial improvements. For the first time, it supports a context window of up to 1 million tokens in its beta phase. This massive capacity allows it to process and recall information from very large documents or complex codebases in a single session, a critical feature for effective multi-agent systems.
Leading in Reasoning and Long-Context Tasks
Anthropic released several benchmark results positioning Opus 4.6 ahead of competitors like OpenAI's earlier GPT-5.2 and Google's Gemini 3 Pro on various tests measuring reasoning and information retrieval.
- On ARC AGI 2, a test for solving novel problems, Opus 4.6 scored 68.8%, a dramatic improvement over the 37.6% scored by its predecessor.
- On a long-context retrieval benchmark, Opus 4.6 scored 76% on the 1 million-token version, demonstrating its ability to accurately find information within vast amounts of text.
This focus on long-context reliability is essential for agent teams, as they must track information across hundreds of thousands of lines of code or text without losing the thread of the original task.
Market Anxiety and the Dawn of 'Vibe Working'
The rapid advancement of these agentic AI systems has not gone unnoticed by financial markets. The releases followed a week of extreme volatility for software stocks, reportedly triggered by an earlier Anthropic announcement of open-source plugins for its agentic tool, Cowork.
The plugins extended Cowork's capabilities into specific professional domains like legal contract review, financial analysis, and compliance workflows. Investors reacted nervously to the idea that AI companies could package and offer complete workflows, directly competing with established software-as-a-service (SaaS) vendors.
In the wake of the news, an estimated $285 billion in market value was erased across software, financial services, and asset management stocks. A Goldman Sachs index of U.S. software stocks fell 6% in a single day, and Thomson Reuters saw its stock drop by 18%.
The concern is that platforms like OpenAI's Frontier could become the "operating system of the enterprise," mediating how work gets done across all other applications. Whether these tools can reliably achieve these tasks is still an open question, but the ambition is clear.
Scott White, Anthropic's head of product for enterprise, coined a new term for this emerging work style. He noted that developers have embraced "vibe coding," where they can turn ideas into code more fluidly with AI assistance. "I think that we are now transitioning almost into vibe working," he said, suggesting a future where knowledge workers can orchestrate complex tasks with similar ease.