Unlocking exponential value with AI agent orchestration
Autonomous AI agents may be transformational, but orchestration can be key for intelligent automation. Open source and proprietary communication protocols will compete to lead the way.
As companies integrate multiagent systems—where different AI reasoning engines interact seamlessly across domains—agent orchestration (the effective coordination of role-specific agents) will be essential to help unlock their full potential. Thoughtful orchestration unleashes intelligent workflows by enabling multiagent systems to interpret requests, design workflows, delegate and coordinate tasks, and continuously validate and enhance outcomes.1Conversely, poor agent orchestration can significantly limit this business value.
On average, market estimates suggest that the autonomous AI agent market could reach US$8.5 billion by 2026 and US$35 billion by 2030 (figure 1).2 Deloitte predicts that if enterprises orchestrate agents better and thoughtfully address the associated challenges and risks, this market projection could increase by 15% to 30%—or as high as US$45 billion by 2030. According to an estimate, more than 40% of today’s agentic AI projects could be cancelled by 2027, due to unanticipated cost, complexity of scaling, or unexpected risks.3 These projects could drive significant revenue growth if enterprises remediate the potential pitfalls preemptively.
To leverage multiagent systems fully, businesses will likely work on their readiness to orchestrate agents with a specific degree of autonomy and address the early potential pitfalls. At the same time, multiagent systems will likely work for those businesses that focus on agent interoperability and management and implement the required changes in workflows and talent, effectively.
Making businesses work for multiagent systems
As businesses work through decisions related to their agent orchestration preparation, these three guideposts will likely be pivotal.
From single-purpose agents to multiagent systems: Are enterprises ready?
Enterprises today could leverage single-purpose AI agents to carry out multiple steps autonomously.4 Increasingly, they’re realizing that the benefits of agentic AI also extend to multiagent systems, unlocking broader and exponential enterprise value.5 However, tech implementations could be far from maturity for many organizations.
In Deloitte’s 2025 Tech Value Survey of nearly 550 US cross-industry leaders, 80% of respondents believe their organization has mature capabilities with basic automation efforts, whereas only 28% believe the same with basic automation and AI agent–related efforts. Furthermore, among those pursuing each strategy, 45% expect that their basic automation efforts could yield the desired return on investment within three years, whereas only 12% expect the same for basic automation and agents, within a similar time frame.6
How can they get there faster? Step one is to consider the three potential multiagent approaches (figure 2).7
The human layer in agent orchestration
In 2025, businesses have been implementing relatively simple yet promising agent orchestrations in specific domains, like financial investment research and health care for critical illnesses.8 In such applications, agents often work together under the purview of human supervision or a dedicated “supervisor agent” to provide insights for human professionals to act on. More complex and autonomous agent orchestration spanning across multiple business domains has been limited, for the most part, to select industry leaders.9 As such efforts intensify, businesses will increasingly need to balance agentic autonomy and human oversight—carefully weighing innovation against risk, accountability, and trust.
Research suggests that today’s emerging multiagent systems can perform better with humans in the loop, as they benefit from human experience and remain aligned with the nuanced organizational expectations.10 We predict that, in the next 12 to 18 months, more businesses will accelerate experimenting and scaling of complex agent orchestrations, keeping humans in the loop. They will likely adopt frameworks and solutions to integrate human judgment into agentic workflows for higher confidence, quality, and accountability.11
Additionally, a progressive “autonomy spectrum”—humans in the loop, on the loop, and out of the loop—will emerge based on task complexity, business domain, workflow design, and outcome criticality (figure 3). While the humans out of the loop approach will still need continuous monitoring—human-in-the-loop and human-on-the-loop approaches will rely more on platforms and agent telemetry dashboards offering outcome tracing, orchestration visualization, and other details to guide human interventions. We predict, in 2026, the most advanced businesses will begin to lay the foundation of shifting toward human-on-the-loop orchestration.
Taming the fragmented AI agent proliferation
In 2026, AI agent sprawl is likely to increase across different programming languages, frameworks, infrastructure, and communication protocols. To add complexity, some agents might need multimodal capabilities (the ability to interpret different information types and formats like text, audio, and images) to reach peak intelligence. Additionally, web protocol developments for agents, like Massachusetts Institute of Technology’s project NANDA, can define how agents coordinate on digital interfaces, external to businesses.12 In the longer term, it can enable strategic agent orchestration across internal and external networks of businesses, unlocking new capabilities.
These variables will make multiagent interoperability critical yet challenging. Additionally, businesses will increasingly look for ways to direct, observe, and manage disparate AI agents through a unified platform. Lack of digital workforce operational standards may make building, configuring, and deploying AI agents decentralized and uncoordinated. This, in turn, will likely increase potential risks and costs of performance degradation and ethical, cyber, and regulatory compliance issues.
Businesses can draw inspiration from previous technologies that shaped today’s information technology and business architecture, like cloud and microservices. Standardized protocols (like HTTPS, JSON, etc.), clear application programming interface blueprints, and domain-specific microservices enabled interoperability, stability, and ownership. Service registries, distributed tracking, and centralized logs improved discovery of capabilities, error resolution, and service management. Governance, service catalogs, and “zero-trust” security ensured robust systems and prevented confusion about versions. All these measures could offer lessons for building resilient and scalable multiagent systems. However, businesses should also adopt a fresh approach and focus on creating unique layers in their enterprise architecture.
Enterprise architecture for resilient and scalable multiagent systems
- Context layer: This robust knowledge engineering foundation is important for scalable AI agent architecture. It translates raw and diverse data into structured and well-governed knowledge representations (for example, knowledge graphs, ontologies, domain taxonomies, etc.) to provide agents with a “small world” model of the problem space. Optimized context retrieval techniques can empower agents with precise and timely access to relevant information, while context shaping can refine inputs to reduce noise and conflicts, enhancing agent accuracy and efficiency.
- Agent layer: This component leverages the underlying context layer to enable agent operations, focusing on safety, autonomy, and interoperability. Central to this layer is a modular and composable architecture that can integrate and adapt to new technologies. Strategies emphasizing tool relevance and abstraction help prevent agent overload. Additionally, thoughtful memory strategies optimize access to the right blend of factual, experiential, and procedural memories to enhance context awareness. This layer also selects appropriate AI models (ranging from compact, specialized models to expansive, powerful ones) to optimize agent performance across orchestration tasks. Robust security measures and comprehensive observability via advanced telemetry help ensure secure, transparent, and reliable agent activities.
- Experience layer: This primary interface between enterprise users and agents helps to control and course-correct agent actions. It provides users with relevant information like agent status and contextual data. It also enables prompt suggestions and comprehensible results in easy-to-review formats. Intuitive controls for human oversight, advanced feedback capabilities, and explainability features like displaying agent reasoning help make the outcomes more transparent and trustworthy. Additionally, when errors or ambiguous situations arise, it provides clear explanations and options to recover.
Making multiagent systems work for businesses
As businesses master the technical foundations, these three guideposts can help enable better alignment with business imperatives.
Flexible, scalable, and secure communication protocols
Multiagent orchestration requires a standard form of communication among agents and between agents and other tools or platforms. It’s essential for predictable messaging on agent capabilities, insights, and actions. Over the last year, several inter-agent communication protocols have emerged, each promising coordination among agents built on different frameworks or models. These include Google’s A2A, Cisco-led AGNTCY, Anthropic’s MCP, and others.13 Tech providers are rallying their partners, alliances, and customers to achieve dominance in this category. Additionally, some of these protocols are being extended for trustworthy agent interoperability in specific domains like financial transactions.14
Excessive competition across protocols could risk the development of “walled gardens,” where companies are locked into one communication protocol and agent ecosystem.15 It’s likely, however, that, by next year, these protocols will begin converging, resulting in two or three leading standards that other tech providers will need to align with to remain competitive.
Which select protocols rise to the top will likely depend on multiple parameters and how businesses prioritize them according to their multiagent use, industry, and orchestration maturity. For example, lightweight protocols with standard application programming interfaces and developer tools for testing and simulation can ease experimentation. Support for peer-to-peer and hub-and-spoke agent interactions with shared context and memory and built-in negotiation, delegation, and conflict resolution can enable diverse orchestrations. Agent registries for trusted discovery and workload balance, asynchronous messaging, high throughput, low latency, and support for chained and nested workflows can help scale up agent orchestrations. Additionally, authentication, secure messaging, and access control can help mitigate security risks, while inter-agent messages and explanations can ensure auditability and error traceability.
Management platforms and observability tools
As multiagent systems scale, businesses will increasingly need to manage agents and understand the decisions being taken by them. They can leverage the unified and scalable platforms available, with supervising capabilities or “supervisor agents”—to interpret requests, route tasks, grant and manage access, and execute parallel or multi-step processes.16 It’s likely that, in the next year, tech companies will launch new capabilities here, leaving businesses to decide how they want such orchestration platforms set up. For example, central in-house platforms can limit vendor dependency and increase data and agent control. However, off-the-shelf platforms can help accelerate testing and manage the cost of innovation.
Whatever businesses choose, agent orchestration platforms will be important to track operational metrics, enhance performance, and manage cost. Currently, some platforms are developing ways to integrate monitoring of agent telemetry such as latency, error rates, token usage, and other tool insights.17 Guardrail assessments and capabilities to detect unusual behaviors can help mitigate risks. Over time, such platforms will likely bring innovative features, such as layered business insights and additional control mechanisms. For example, an emerging category called guardian agent can both own tasks and govern other agents to sense and manage risky behaviors.18
Agent orchestration platforms will also need to incorporate regulatory compliance, an area where international efforts are advancing. The European Union AI Act sets requirements around risk assessment, transparency measures, technical safeguards, and human oversight.19 In addition, the EU’s standards bodies are working to develop harmonized legal standards as per the EU AI Act.20
Business process and workforce changes
Gartner® predicts that, by 2028, “33% of enterprise software applications will include agentic AI, up from less than 1% in 2024, with at least 15% of day-to-day work decisions being made autonomously through AI agents.”21 To get there, more businesses will likely begin reimagining their workflows in 2026, defining concrete and unique modules. This will help determine the kinds of agent orchestration needed, depending on criticality, dependencies, task predictability, and targeted resilience. For example, some modules may benefit from agents working sequentially—where one agent’s output becomes another’s input—while other modules might leverage agents operating in parallel or collaboratively.
Another major consideration is how humans will collaborate with multiagent systems. A global survey of 200 human resources leaders found that 86% of chief human resources officers see integrating digital labor (that is, technologies performing intelligent work) as central to their role.22 Early models show humans acting as “agent bosses,” or working alongside agents.23 In 2026, businesses will likely delve deeper into these collaboration models across more roles, functions, and tasks to identify where agent orchestration can enhance efficiency and where human strengths and collaboration can bring more meaningful value.24
By next year, enterprises will also likely start reimagining how existing roles can unlock higher-value outcomes with multiagent systems.25 For example, human contributions can include more creative prompting and guiding multiagent systems while solving problems and taking strategic decisions efficiently. At the same time, businesses will also likely focus on defining the new human skills and responsibilities for agent training, orchestration, oversight, and governance.26 Tailored training programs and developing leaders to manage both human and digital workers will be important—to embed higher quality, accountability, and resilience in multiagent decisions while leveraging uniquely human skills.27
The bottom line: 2026 could be an inflection point for agent orchestration
Agent orchestration will likely shape the next era of intelligent enterprises. Next year, we expect businesses to start scaling multiagent systems, bringing additional complexity to their IT and business environments. Agent communication protocols will likely consolidate around those offering ease of experimentation, flexibility, scalability, and security. Enterprise workflows will likely start becoming more modular, powered by agents—built internally or acquired through software as a service and other third-party providers. New and modified roles for human workers will begin emerging, facilitating effective collaboration with multiagent systems.
However, businesses and technology providers should act decisively to shape that journey.
Considerations for businesses adopting multiagent systems
- Define ownership and accountability. Businesses should identify who in the C-suite will own their company’s AI agent vision, strategy, and execution with aligned incentives and accountability. This role could most naturally align with those leading strategic technology initiatives and driving innovation, but an integrated function can demonstrate more holistic impact and risk management.
- Design for evolution, not just deployment. Agents and orchestration capabilities are advancing fast. Modular “plug-and-play” orchestration frameworks can help businesses boost flexibility, cost-efficiency, and innovation, while minimizing disruption to system architectures.
- Stress-test orchestrations rigorously. Before scaling, businesses should simulate agent orchestration with real complexities of businesses—incomplete data, conflicting goals, or adversarial scenarios. Controlled environments can reveal hidden failure points and strengthen safeguards before enterprisewide deployments.
- Take governance and measurement seriously. AI agent governance will be critical to help ensure secure, compliant, and reliable orchestration on a scale. Setting clear rules for AI agent roles, defining their accountability, designing fallback routes to address errors, and oversight can help prevent misuse, ensure auditability, and build trust. Beyond technical readiness, enterprises should identify and track metrics that connect agent orchestration to value creation—such as quicker decisions, better customer experience, or faster innovation.
Considerations for tech providers
- Build with interoperability. Besides adhering to inter-agent communication standards, tech providers should design solutions that are modular, and where agents understand each other’s intent and context of actions, to enable seamless coordination.
- Rethink trust. Insight delivery won’t be enough; the ability to understand or validate AI agent output is essential for trust and adoption. Novel security measures like digital identity for agents will also be pertinent to build and run trustworthy multiagent systems.
- Make governance inherent. Learning what businesses will need over time, to align with human values and organizational policies, could be key to providing relevant governance frameworks. Future solutions should have innovative agent monitoring and advanced governance, and ethical guardrails to enable compliance and efficacy.
- Expand the ecosystem. Tech providers should continue forming and strengthening industrywide alliances to achieve necessary standards in communication protocols, trust, and governance. Innovative and cross-platform orchestration tools are gaining traction, signaling opportunities for new and established tech players to strengthen their market position through acquisitions, partnerships, and collaboration.28
by
Sayantani Mazumder
China Widener
Gillian Crossan
Girija Krishnamurthy
Baris Sarer
Diana Kearns-Manolatos
The authors would like to thank Prakul Sharma, Rajib Deb, Mark Szarka, David Jarvis, Abhinav Chauhan, Michael Steinhart, Ankit Dhameja, and Iram Parveen for their contributions to this article.
Cover image by: Jaime Austin; Adobe Stock