Browser infrastructure company Anchor has announced the launch of Action Index, a new system designed to mitigate the heavy computational expenses associated with computer-use AI agents. According to the company, the tool stores successfully completed web workflows in a shared execution memory, allowing agents to bypass real-time reasoning steps for repetitive tasks by replaying proven paths as deterministic code.
Shifting From Real-Time Reasoning to Shared Memory
Traditional AI agents navigate web environments by continuously reasoning through every click, form submission, and page transition from scratch. This process consumes extensive token budgets and introduces latency. Anchor addresses this inefficiency by maintaining a repository of proven web trajectories. When an agent encounters a recurring workflow, it retrieves the established execution path and runs it deterministically. The underlying language model is only invoked if a web page changes or unexpected UI elements appear, drastically lowering computational overhead.
Idan Raman, co-founder and chief executive officer of Anchor, noted that modern agents frequently waste resources re-solving problems that have already been handled previously. By treating execution memory as the primary bottleneck rather than raw model intelligence, subsequent iterations of a web task require a fraction of the initial computing power.
Benchmark Results and Cost Reductions
In evaluations conducted across 300 tasks from the Online-Mind2Web benchmark, Anchor reported significant performance gains. A smaller language model paired with Action Index achieved a 93.3% task completion rate at an average cost of $0.016 per task, matching the success rate of a costly frontier model while reducing total run expenditures from $472 down to $4.87. Without the system, the same smaller model managed a success rate of only 62.7%.
Additional testing across a paired study of 120 tasks demonstrated that Action Index decreased 90th-percentile token usage by 74%, cut execution steps by 75%, and shortened wall-clock time by 61%. Furthermore, the infrastructure lifted the performance of two leading open-weight models, raising their success rates from 67.7% to 88.7% and from 58.7% to 84.7%, respectively.
Availability and Developer Integration
Action Index is available to all Anchor accounts immediately, including users on the free tier. The tool can be accessed through tools on Anchor's Model Context Protocol server and via a REST API, making it compatible with any MCP-compliant agent framework or third-party model provider. Documentation is provided for developers looking to integrate the shared memory system into existing automation pipelines.
Anchor builds upon its core secure cloud browser infrastructure, which provides managed chromium instances capable of handling authentication, session management, and complex multi-page interactions. Developers building automated research agents and enterprise workflows leverage these isolated cloud environments to execute tasks reliably at scale without relying on traditional fragile brittle scripts.
Economic Implications for Browser Automation
The introduction of Action Index highlights a critical structural shift in how companies approach AI-driven web interaction. Historically, scaling computer-use agents has faced severe economic hurdles because token expenses scale linearly with every navigational step, quickly rendering complex browser automation economically unviable for high-volume enterprise applications. By converting successful agent runs into deterministic code patterns, Anchor effectively bridges the gap between expensive frontier intelligence and cost-effective operational execution.
This development carries significant implications for smaller startups and open-source practitioners. Because Action Index allows open-weight models to punch significantly above their weight class on standardized benchmarks like Online-Mind2Web, developers are no longer forced to rely exclusively on proprietary, high-cost foundation models for reliable browser workflows. If shared execution memory becomes an industry standard, the primary competitive moat for web automation will shift from raw inference power to repository quality and execution reliability, fundamentally altering the unit economics of autonomous software agents.
Cover: AI-generated illustration with headline text; not a documentary photograph.
Source: Anchor Launches Action Index, Cutting Computer-Use Agent Token Costs by 100X · Anchor – Secure Infrastructure for Computer Use Agents
SIGNAL TITANS
Leave a Reply