Insider Brief

• Emergence AI unveiled the Emergence Orchestrator, a meta-agent platform enabling seamless coordination between AI agents for automating complex enterprise workflows, integrating modern and legacy systems with enhanced scalability and modularity.
• The platform features specialized agents, including the Web Agent and API Agent, to streamline tasks like web automation, data retrieval, and compliance monitoring, while maintaining robust data privacy and security through options like Virtual Private Cloud deployment.
• With successful beta testing across industries and use cases such as QA testing, supply chain management, and research analysis, the Orchestrator is set to redefine enterprise productivity by offering adaptable, efficient, and secure multi-agent solutions.

PRESS RELEASE — Emergence AI has announced the Emergence Orchestrator, an autonomous meta-agent that coordinates and manages interactions between AI agents across enterprise systems. Through dynamic orchestration, it enables multiple autonomous agents to work together seamlessly, handling sophisticated workflows that span modern and legacy software platforms.

‍

‍

In this first release, the Orchestrator showcases its capabilities for web automation through two foundational agents:

· The Web Agent: Optimized for interacting with over the top web interfaces.

· The API Agent: Tailored for managing enterprise APIs.

Together, these agents provide a powerful synthesis, orchestrated by the Emergence Orchestrator, to deliver holistic, enterprise-wide automation.

Modern enterprises rely on hundreds of applications across diverse domains, with AI agents increasingly front-ending these systems. As enterprises deploy more autonomous agents to manage their software ecosystem, orchestration becomes essential for coordinating these agents at scale.

‍Why Multi-Agent Orchestration Matters

Organizations use a wide variety of applications, from decades-old mainframes to highly modern cloud services. Many of these systems are siloed, leading to inefficiencies, gaps in data, and significant integration challenges. Traditional automation tools struggle to effectively navigate this landscape because they tend to be narrowly focused, treating each task as an isolated process.

The Emergence Orchestrator tackles these challenges by enabling cooperation among specialized agents. Unlike traditional means of integrating disparate systems, which often require rigid integration or bespoke solutions, our orchestrator coordinates agents to perform complex workflows as a team.

Our roadmap extends these capabilities significantly. We’re advancing web automation by integrating Vision-Language Models with DOM processing (See our NeurIPS OpenWorld Agents paper “Multimodal Auto Validation For Self-Refinement in Web Agents”) and developing live-streamed demonstrations of web agents in action. We are developing additional system agents including data science and compliance agents that adapt to enterprise systems and constraints. Over the next year, we’ll introduce a multi-turn conversational chat interface and release our Agent Software Development Kit (SDK), enabling enterprises to create and register their own agents in the Orchestrator’s Registry. Additionally, our upcoming “Build Your Own Orchestrator” platform will allow enterprises to configure custom orchestrator instances, further expanding the possibilities for enterprise automation.

‍Our Approach: Intelligent, Loosely Coupled Systems

The Emergence Orchestrator takes inspiration from best practices in distributed systems and systems theory, creating a loosely coupled network of agents that collaborate to solve diverse enterprise problems. The Orchestrator maintains a registry of all agents — whether first-party, customer-built, or third-party — and dynamically assigns tasks based on each agent’s capabilities.

This architecture makes the Emergence Orchestrator fault-tolerant, scalable, and extensible. It allows us to reduce integration complexity by leveraging natural language for communication between agents, making them more capable of handling non-standardized systems. This is particularly beneficial in legacy environments where many tools lack modern APIs.

Emergence’s agents are self-improving: they continuously refine their capabilities by learning nuances of systems as they evolve, such as adjusting to layout changes or new features. We are advancing fundamental research in this area with techniques such as skill harvesting, which improves both task accuracy and efficiency (requiring fewer LLM calls).

‍System Agents

We’ve started by orchestrating two core agents:

The Web Agent interacts with web interfaces, automating processes such as data entry, extraction, and web navigation. As outlined in our NeurIPS paper, “Agent-E: From Autonomous Web Navigation to Foundational Design Principles in Agentic Systems”, our web automation framework employs skills modeled after human-like browser interactions. This managed service version navigates dynamic menus, fills forms with inferred data, adjusts settings, processes embedded files, and extracts context-relevant data from PDFs or HTMLs. It handles errors like page failures or pop-ups with vigilant self-improving, adaptive navigation.

The API Agent seamlessly integrates enterprise APIs and databases, ensuring data flows smoothly between different systems. Designed to work seamlessly with other agents, like the Orchestrator and Web Agent, the API Agent enables natural language interactions with enterprise SaaS applications. With features like secure credential management, robust error handling, and an expandable API registry, the API Agent eliminates the complexity of manual integrations while maintaining enterprise-grade security and resilience.

This dual strategy allows us to support both modern and legacy systems a swell as handle multi-system integrations, adapting to API changes and upgrades without disruption. While these two agents demonstrate the power of our orchestration approach, they represent just the beginning of what’s possible with multi-agent orchestration.

‍Benchmarks for Real-World Enterprise Challenges

Web Voyager Benchmark Performance

‍

‍

Our technology demonstrates leading performance on public benchmarks. In independent evaluations conducted by H Company, which recently introduced their Runner H agent, our Web Agent (Agent-E) achieved a 61.1% task completion rate, surpassing both the WebVoyager benchmark and Claude Computer Use. However, in our evaluations (also published in an original research paper) against the Web Voyager benchmark suite of 643 tasks across 15 dynamic websites, our text-only Web Agent achieves a 73.2% overall task completion rate, significantly outperforming both WILBUR (52.6%) and the multimodal Web Voyager Agent (57.1%) [1]. This highlights the difficulty of evaluating these systems consistently in real-world conditions. When we enhance our approach with multimodal validation, accuracy increases substantially — reaching over 90% on platforms like GitHub (94.7%), flight booking systems (95.1%), and ESPN (95.3%), with false positive rates below 5% [2]. These results validate our path towards deterministic web automation.

‍Setting New Standards for Enterprise-Specific Benchmarks

One of the major obstacles in evaluating enterprise automation tools is the lack of benchmarks that accurately reflect realistic and challenging enterprise tasks. At Emergence, we’re addressing this by releasing benchmarks specifically designed for real-world enterprise scenarios. These benchmarks will help set the standard for evaluating multi-agent systems in enterprise environments, providing a reliable measure for effectiveness and scalability. Read more about our benchmarks in our latest whitepaper.

‍Benefits of Multi-Agent Orchestration for Enterprises

Increased Efficiency and Productivity: While individual agents continuously improve at achieving their respective goals, the orchestrator ensures that they work together to automate diverse enterprise workflows. This interplay between specialized agents and the orchestrator enhances scalability and modularity, creating a cohesive system that enables sustainable, long-term maintainability.

Adaptability to Complex Environments: Enterprises often face changing requirements, from regulatory updates to evolving business processes. The Orchestrator uses a plan-verify-execute loop to adjust workflow plans dynamically, ensuring reliable and consistent outcomes even if the agents’ environments shift between uses.

Data Privacy and Security: One of the biggest hurdles for enterprise adoption of AI is data privacy. In addition to its hosted, API-accessible solution, the Orchestrator also offers the ability to run entirely within the customer’s Virtual Private Cloud (VPC) for maximum security. This meets compliance standards without sacrificing any benefits of automation.

Scalable Collaboration Across Agents: The Orchestrator isn’t limited to Emergence’s in-house agents. Using our Agent SDK (releasing in2025), enterprises can register custom or third-party agents, meaning there’s virtually no upper-limit to the customizability of this solution.

‍Example Use-Cases

Emergence’s Orchestrator demonstrates its value across many applications and industries. Below are example workflows a business might deploy using multiple AI agents.

Supply Chain: The Web Agent gathers data from supplier portals. The API Agent retrieves information from systems like SAP. The data is synthesized to dynamically identify risks and generate comprehensive supplier ports.

QA Testing: Agents simulate web navigation, form validation, and error handling, to accelerate QA testing in SaaS or e-commerce applications.

Research Analysis: Structured data retrieval via APIs is combined with unstructured data scraped from research papers and public records, offering deeper insights for analysts.

Compliance: Agents monitor transactions, flag policy violations, and generate detailed audit reports for regulatory adherence.

‍Highlights from field testing

1. Broad Market Appeal Across Company Sizes: Beta users range from startups with fewer than 25 employees to enterprises with thousands of employees.

2. Cross-Industry Versatility: The API is being actively tested in sectors ranging from technology and health care to government, education, finance, and manufacturing.

3. Strategic and Technical Engagement: Adoption spans organizational hierarchies, including C-suite executives like CTOs and product leaders, as well as technical users like software engineers, product managers, QA managers, ML engineers, and data scientists.

4. Diverse and High-Impact Use Cases: Key use cases emerging from beta testing include QA testing, research analysis, product vetting, data collection, and workflow automation.

‍Unlocking Enterprise Automation

The Emergence Orchestrator advances the capabilities of AI-driven enterprise automation. By enabling a flexible, scalable, and secure multi-agent orchestration platform, we’re helping enterprises break down silos, improve productivity, and accelerate their transformation journey.

Contact us today to discuss your unique needs and begin building with multi-agent orchestration.

‍Resources

1. Web Automation API and associated documentation

‍2. How-to: Emergence Web Automation: Getting Started with the Online Playground

‍3. How-to: Web Automation API from Python

‍4. Example use-case in supply chain

‍5. Playlist with more use cases

SOURCE

‍