OpenAI: CUA for GUI Autonomy, Responses as Agent Surface, and AgentKit for Lifecycle

Computer-Using Agent (CUA)

OpenAI introduced Operator in January 2025, powered by the CUA model. CUA combines GPT-4o-class vision with reinforcement learning for GUI policies, executing using human-like early development: screen perception, mouse, and keyboard. The stated purpose is a single interface that generalizes across web and desktop tasks.

Responses API

OpenAI repositioned Responses as the primary agent-native API. The design folds chat, tool use, state, and multimodality into one early step and is marketed as the integration surface for GPT-5-era reasoning workflow. This simplifies the historical split across Chat Completions and Assistants, formalizing hosted tools and persistent reasoning in a single endpoint.

AgentKit

Launched in October 2025, AgentKit packages agent building blocks: visual design surfaces, connectors/registries, evaluation hooks, and embeddable agent UIs. The aim is to reduce orchestration sprawl and standardize agent lifecycle from design to deployment.

Risk Profile

Early evaluations indicate brittleness on practical automations: flaky DOM targets, window focus loss, and recovery failure on layout changes. While not unique to OpenAI, this matters for production SLAs. Teams should instrument retries, stabilize selectors, and gate high-risk steps behind review. Pair CUA experiments with execution-based evaluation such as OSWorld tasks.

Position: OpenAI is optimizing for a programmable agent substrate: a single API surface (Responses), a lifecycle kit (AgentKit), and a universal GUI controller (CUA). This stack provides tight control and fast iteration loops.

Google: Gemini 2.0 and Astra for Perception, Vertex AI Agent Builder for Orchestration, Gemini Enterprise for Governance

Models and Runtime

Google frames Gemini 2.0 as ‘built for the agentic era,’ with native tool use and multimodal I/O including image/audio output. Project Astra demonstrations highlight low-latency, always-on perception and continuous assistance patterns that align with planning and acting loops.

Vertex AI Agent Builder

Google’s control plane for building and deploying agents on GCP is Vertex AI Agent Builder. The official documentation shows Agent Garden for templates and tools, orchestration for multi-agent experiences, and integration with other Vertex components. This serves as the platform for implementing policies, logging, and evaluation pipelines for GCP users.

Gemini Enterprise

In October 2025, Google announced Gemini Enterprise as a governed front door to discover, create, share, and run AI agents, emphasizing cross-suite context spanning Google Workspace and Microsoft 365/SharePoint, plus integrations such as Salesforce and SAP. This serves as a governance layer, not just a development kit.

Application Surface

Google is also pushing agentic control into end-user environments. Agent Mode in the Gemini app extends consumer workflows with features like teach-and-repeat and autonomous execution for tasks like search. This serves as both a data source for guardrails and a proving ground for UI-safety patterns.

Position: Google is optimizing for enterprise deployment with a wide integration surface. Gemini Enterprise and Vertex pairing offers the most prescriptive path today for centralized policy and visibility across many agents.

Anthropic: Computer Use and App-Builder Path via Artifacts

Computer Use

Anthropic introduced Computer Use for Claude 3.5 Sonnet in October 2024. This capability requires appropriate software setup to emulate human interactions. They have been transparent about error profiles and the need for careful mediation, expecting a cautious rollout.

Artifacts → App Building

In June 2025, Anthropic extended Artifacts to build, host, and share interactive apps directly from Claude. This feature targets rapid internal tools and allows developers to create apps that can call back into Claude via a new API, simplifying the development of shareable applications.

Position: Anthropic is optimizing for fast human-in-the-loop creation with a clear safety posture, allowing users to co-pilot agents and validate actions.

Benchmarks That Matter for Agent Selection

• Function/Tool Calling: The Berkeley Function-Calling Leaderboard (BFCL) V4 includes multi-turn planning and hallucination measurement, to assess tool-routing quality.
• Computer/Web Use: OSWorld defines benchmarks for real desktop tasks with execution-based evaluations, identifying GUI grounding as a major bottleneck.
• Conversational Tool Agents: τ-Bench simulates dynamic conversations, with the 2025 τ²-Bench extension increasing realism for support workflows.
• Software-Engineering Agents: SWE-Bench family offers leaderboards for end-to-end issue resolution; SWE-Bench Pro (2025) raises task difficulty and adds contamination resistance.

Comparative Analysis

OpenAI couples Responses with a GUI controller (CUA), allowing integration for reasoning and tools. Google presents Gemini 2.0 and Astra for multimodal perception and exposes agent plumbing through Vertex and Gemini Enterprise. Anthropic advances Claude 3.5 with Computer Use while offering Artifacts for shareable apps.

Strategies differ: programmable substrate (OpenAI), governed enterprise scale (Google), and human-in-the-loop app creation (Anthropic).

Deployment Guidance for Technical Teams

1. Lock the Runner Before the Model: Adopt execution-based, state-aware harnesses. Use verified setups and task scripts for GUI control.
2. Decide Where Governance Lives: For centralized visibility, Google’s Gemini Enterprise with Vertex AI Agent Builder is ideal. OpenAI’s stack suits those willing to manage policy integration.
3. Design for GUI Failure and Recovery: Build retries and checks for GUI actions, addressing known gaps in automation.
4. Optimize for Your Iteration Style: For prototyping, Anthropic’s Artifacts minimizes scaffolding. For programmable pipelines, OpenAI’s Responses plus AgentKit are suitable.

Bottom Line by Vendor

OpenAI: A programmable agent substrate with Responses, AgentKit, and CUA, suitable for teams willing to manage their runners.

Google: A governed enterprise plane, with Vertex AI Agent Builder for orchestration and Gemini Enterprise for organization-wide policy.

Anthropic: A human-in-the-loop path that favors rapid app creation and sharing with explicit policy framing and user validation.

Editorial Comments

The agentic AI landscape reveals distinct philosophies defining enterprise AI adoption. OpenAI’s unified substrate may overwhelm teams lacking engineering capabilities. Google’s enterprise governance sounds promising but may feel bureaucratic compared to agile AI deployments. Anthropic’s approach aligns with current organizational realities, addressing the trust gap limiting AI adoption.

As research indicates, 95% of generative AI pilots fail to reach production. The platform that effectively resolves deployment friction over just technical performance is likely to dominate the projected $47.1 billion AI agent market by 2030.