OpenAI released ChatGPT agent, expanding its flagship product from a conversational AI into an autonomous task execution platform. The system can browse websites, write and run code, complete forms, and perform multi-step actions on a virtual computer with limited human involvement, according to OpenAI’s announcement.
The rollout represents a direct competitive response to Anthropic’s Claude Managed Agents and Perplexity’s agent-first pivot, which recently pushed that company to $450 million ARR.
What It Does
ChatGPT agent combines OpenAI’s previously separate tools into a single reasoning layer. Operator (web browsing), Tasks (scheduling), code execution, and file handling now work together under one interface, as MindStudio detailed. The system can receive a prompt like “analyze three competitors and create a slide deck,” then autonomously research, filter, and format the output without step-by-step human guidance.
Other documented capabilities include calendar navigation with contextual briefings, data analysis across CSV, JSON, and PDF formats, shopping with product comparison, and document review with anomaly detection, according to Incremys.
The agent scored 41.6% on Humanity’s Last Exam and 27.4% on Frontier Math, per OpenAI’s benchmarks. Those results suggest strong multi-step reasoning but also clear limitations.
Architecture and Approach
“At the simplest level, what OpenAI has released is an agent framework that has many more degrees of freedom than previous agentic capabilities,” Gabe Goodhart, Chief Architect for AI Open Innovation at IBM, told IBM Think. The system appears to combine ReAct patterns (iterative action-observation loops) with ReWOO (pre-planned execution), according to IBM’s analysis.
Goodhart flagged the tradeoff directly: “With degrees of freedom comes also degrees of risk.” The system operates on what he called “probabilistic computing rather than deterministic computing,” meaning it works on a best-effort basis rather than executing identical instructions every time.
Consumer Focus, Enterprise Gaps
The design targets individual users and personal tasks. ChatGPT agent connects to consumer services like Gmail and GitHub, while enterprise-focused competitors like IBM’s watsonx Orchestrate integrate with Salesforce, SAP, Workday, and ServiceNow, according to IBM Think.
That consumer focus is both a strength and a constraint. OpenAI’s distribution advantage (500 million+ ChatGPT users) gives it an unmatched consumer install base for agentic AI. But the system currently assumes a single user operating across tools and data without the governance guardrails enterprise buyers require: role-based access, audit logging, and on-premises deployment.
The Competitive Landscape
The launch intensifies a three-way race in consumer and SMB agentic AI. Perplexity’s agent pivot drove 50% month-over-month revenue growth and $450 million ARR. Anthropic’s Managed Agents reached general availability on April 8 with enterprise controls. OpenAI is betting that native integration within ChatGPT, rather than a separate agent interface, will drive adoption.
For developers and businesses evaluating agent platforms, the key distinction is scope: OpenAI is building a unified consumer surface where agents are a feature of the chat experience. Anthropic and enterprise vendors are building agent infrastructure that organizations manage and govern independently.