vcrouter
Start a battlePlay
Leaderboards / Market map / Battle
Market map

Browser agents and computer-use automation

September 13, 2026 · $0.10 for both answers and the judge

Grok 4.6 wins

On September 13, 2026, Gemini 3.1 Pro and Grok 4.6 each wrote a market map for the same brief, without seeing the other's answer. The judge, Claude Fable 5, read both twice, once in each order, and picked Grok 4.6 both times.

Why, in the judge's words“More accurate, current placements (Operator, Mariner, Nova Act, Browser Use) and sharper value-flow logic; A misplaces Skyvern and includes Mistral wrongly.”

The two answers

Gemini 3.1 Pro
$0.021 · 19s · 1,745 tokens

Here is an end-to-end market map for the Browser Agents and Computer-Use Automation space.

1. Market Thesis

The market is transitioning from brittle, rules-based RPA (Robotic Process Automation) reliant on static DOM selectors to dynamic, multimodal AI agents capable of generalized reasoning and autonomous action across unstructured web and desktop environments.

2. Market Segments

Segment 1: Foundational Models & Computer-Use APIs

  • Companies: Anthropic, OpenAI, Google, Adept (Note: Adept's core team was recently absorbed by Amazon, but their foundational work remains highly relevant), Mistral.
  • Dynamics: Competing fiercely on raw spatial reasoning, multimodal capabilities, and reducing pixel-to-action latency at the OS and browser level.

Segment 2: Web-Native Autonomous Agents (Prosumer/SMB)

  • Companies: MultiOn, Skyvern, HyperWrite, Induced AI, Lindy.
  • Dynamics: Racing to own the long-tail workflow layer via browser extensions and cloud-hosted virtual browsers that execute tasks on behalf of users.

Segment 3: Enterprise RPA 2.0 (Cognitive Automation)

  • Companies: UiPath, Automation Anywhere, Sema4.ai, Kognitos, Orby AI.
  • Dynamics: Leveraging massive existing enterprise distribution to upgrade legacy, high-maintenance bots into resilient, intent-driven agents that can self-heal when UIs change.

Segment 4: Agentic Infrastructure & Sandboxing

  • Companies: Browserbase, Steel.dev (Note: early-stage but highly relevant), E2B, Apify, LangChain.
  • Dynamics: Providing the "picks and shovels"—secure, scalable headless browsers, session management, and orchestration frameworks required to run autonomous agents in production without getting blocked by anti-bot software.

3. Value Capture Winner

Enterprise RPA 2.0 (Cognitive Automation) will capture the most value. Why: While foundational models provide the intelligence and web-native agents offer slick UX, the Enterprise RPA incumbents and challengers already possess the hardest things to acquire: enterprise distribution, SOC2/compliance approvals, and deep integrations into legacy systems of record (SAP, Oracle). Upgrading existing enterprise workflows with agentic resilience commands massive ACVs and extreme lock-in, whereas prosumer web agents face high churn and low willingness-to-pay.

4. White-Space Opportunities

  1. Agentic IAM (Identity and Access Management): Currently, agents use human credentials to log into software, creating massive security and audit risks. There is a whitespace for an "Okta for Agents"—a protocol and platform that provisions temporary, least-privilege, dynamically scoped credentials specifically for autonomous computer-use agents, complete with real-time blast-radius containment.
  2. Cross-Device/Cross-OS State Handoff: Current agents are siloed to a single environment (e.g., a cloud browser or a specific desktop). There is no infrastructure layer that allows an agent to begin a complex task on a user's desktop, seamlessly hand off the context to a mobile OS to bypass a mobile-only 2FA or app requirement, and return the result to the desktop.
Grok 4.6Winner
$0.012 · 38s · 1,882 tokens

Thesis: Generalist vision-LLM agents that operate any GUI are collapsing brittle RPA/scripts and custom integrations into a model-native computer-use layer, shifting value toward frontier models plus the execution infrastructure that makes them reliable at production scale.

Segments

  1. Frontier computer-use models
    2. OpenAI (Operator), Anthropic (Claude Computer Use), Google (Project Mariner), Amazon (Nova Act — less sure on exact product branding/availability). Dynamics: Capability race on reliability, latency, and long-horizon tasks; still mostly demo/early-access vs. production-grade.
  2. Horizontal agent platforms / copilots
    3. MultiOn, Induced, Convergence (Proxy), Lindy, HyperWrite. Dynamics: Wrapping current models into consumer/prosumer “do this for me” products; high product-market-fit risk as models improve and platforms can go direct.
  3. Browser execution / session infrastructure
    4. Browserbase, Steel.dev, Skyvern, Browser Use (library + ecosystem). Dynamics: Picks-and-shovels for stealthy, scalable, observable headless/GUI sessions; selling to everyone building agents.
  4. Legacy RPA + enterprise automation
    5. UiPath, Automation Anywhere, Microsoft (Power Automate / Copilot Studio), SS&C Blue Prism. Dynamics: Installed base, compliance, and sales motion; grafting LLMs onto existing bots rather than starting from generalist computer use.
  5. Developer/open tooling layer
    6. Playwright (plus emerging agent wrappers), Stagehand (Browserbase-adjacent), Apify, plus open computer-use loops. Dynamics: Commoditizing the agent loop and evaluation; talent and standards layer more than a large standalone market.

Who captures the most value: Frontier models (OpenAI/Anthropic/Google, and Amazon if they execute). They own the scarce capability, can ship first-party products (Operator-style) and APIs, and will keep improving the core loop faster than wrappers. Infra is a solid but smaller picks-and-shovels market; horizontal apps get squeezed as models get better and platforms go direct; RPA keeps enterprise budget in the near term but is structurally defensive.

White space (no one fully owns today)

  • Cryptographically auditable, permissioned computer-use with SSO/delegation, action logs, and policy enforcement suitable for regulated enterprises (finance, healthcare, government).
  • Persistent, user-specific computer-use agents with long-term memory, skill libraries learned from a person’s actual workflows, and hybrid local/cloud execution for privacy — not another generic “Operator” session.

Gemini 3.1 Pro vs Grok 4.6, overall

Across 30 judged battles, Gemini 3.1 Pro won 10, Grok 4.6 won 15, and 5 were ties. On market map alone: 1 to 9.

Compare Gemini 3.1 Pro and Grok 4.6 Judge one yourself, blind

Costs are the provider's list prices at the time of the run.