Scaling AI agents may be risky without an enterprise marketplace

From the perils of localized AI agents to the promise of full-blown ecosystems

Deloitte has predicted that 25% of companies that use generative AI will likely launch agentic AI proofs of concept in 2025, potentially growing to 50% in the next two years.³ Growing adoption could be proof that the performance of these agentic systems is significantly improving, as businesses tend to adopt technologies more readily when they observe improvements in effectiveness.⁴ As the trustworthiness of AI outputs continues to increase, agents will likely be systematically integrated into more and more solutions.

Right now, many teams across many organizations are building their own agents in silos. ⁵ One team might work with an existing agentic framework (used by developers to build and deploy agents), another with a different reputable framework, and yet another with a tool from a promising startup.

This is normal with emerging tech, but it’s not sustainable. As more AI agents start interacting with business systems, data, and even each other, risks increase: Agents might duplicate each other’s work and operate from different governance models, potentially causing misalignment between objectives. They could also mistake malicious software for customers, accidentally reveal sensitive information to clients, or pass need-to-know information (such as salary details) to one another that can easily fall into the wrong hands internally.

For these agents to work in harmony—and for organizations to ensure security—AI agents will need to exist within some form of controlled ecosystem. Prominent hyperscalers and agentic companies seem to agree. CrewAI,⁶ Google Cloud Platform,⁷ Amazon Web Services,⁸ Microsoft Azure,⁹ and LangChain or LangGraph¹⁰ all appear to be placing bets on agentic AI marketplaces.

What is an enterprise AI agent marketplace, and what does it do?

Think of a marketplace as the enterprise equivalent of an app store, where users can subscribe to deployable AI agents. It’s more than a place to publish and reuse AI agents across teams.¹¹ A well-designed marketplace gives you:

• Governance guardrails that are built in, not bolted on:¹² In the rush to innovate with agents, it can be tempting to build without guardrails and then add proper oversight, ethical behavior, and compliance protocols later when it’s time to scale. While this may seem like the faster approach, we’ve found in our experience working with clients that it’s ultimately not as efficient or as safe as taking the time to build guardrails up front and then scaling within those boundaries.
• Clear visibility into who’s using what, where, and how:¹³ An effective marketplace will offer administrators built-in monitoring, tracking, and analytics features to provide visibility into AI agent usage across the organization.
• A centralized management layer: An effective marketplace can become a foundation for scaling the agent-driven management, monitoring, and optimization of other AI agents while keeping everything running smoothly.¹⁴ Having an organization’s agents tethered to a centralized location will make it possible for other agents to prevent downtime and inefficiencies, and further reduce risks because they can proactively identify and address performance issues before they escalate at scale. Without a centralized approach, managing AI agents would be slower, more manual, fragmented, and riskier. Governance alone isn’t enough to prevent operational bottlenecks or ensure scalability.

External and internal marketplaces: How they’re different, and how they complement each other

Agent marketplaces can be built internally by enterprises or run by a third party (external marketplaces). Enterprises can use internal marketplaces to release and manage customized AI solutions that address specific objectives. External marketplaces provide complementary AI agents that can seamlessly integrate with internal solutions, enriching their capabilities and extending their effectiveness to meet broader business needs.

Just as mobile apps saw rapid proliferation after open-source app stores arrived, external marketplaces are showing signs of early growth as developers rush to create reusable agents for them.¹⁵ Many enterprises likely will soon source agents from trusted third parties—but only if they can run them securely within their own governed environment.

External marketplaces are starting to emerge as the open, go-to platforms for discovering, publishing, and subscribing to AI agents that can automate workflows, analyze data, and interact with customers.¹⁶ These marketplaces make it easier for companies to tap into innovation from third-party developers—whether by buying prebuilt agents, subscribing to specialized tools, or sharing internally developed ones with the broader ecosystem.¹⁷ As AI agents become more heterogeneous and are hosted in different environments (internal and external), they also need to become more interconnected through standard protocols. Adopting a marketplace approach could achieve this while also helping to reduce risks, which could in turn help to promote industry growth for AI agents. As these platforms grow, they’ll likely play a big role in how enterprises source and scale AI capabilities—but only if integrated into a secure, governed internal environment.

Internal marketplaces are either built from scratch specifically for an organization or purchased from external vendors and put through a rigorous vetting process before being released to employees internally. They’re curated environments where teams can publish, discover, reuse, and govern AI agents with full control. This concept is similar to a data products marketplace, which can enable organizations to exchange, share, and trade pre-vetted data to drive digital transformation and enhance algorithm training.¹⁸

With an effective internal marketplace, agents are vetted for compliance, tagged with metadata, version-controlled, and monitored through their life cycle. Teams can see what’s already available before building from scratch, which can help reduce duplication and speed up deployment. Internal marketplaces also provide a centralized approach to managing, monitoring, and optimizing AI agents, ensuring that performance, security, and accountability are built in from day one.

In short, the internal marketplace isn’t just a repository—it becomes the backbone of safe, efficient, and trusted AI agent adoption across the enterprise. For instance, when the compliance team seeks to automate regulatory reporting, they can access this marketplace to find pre-approved AI agents that can be easily integrated into their workflow. Each AI agent in the marketplace will already have been vetted for security and compliance, ensuring that risk is minimized while operational efficiency is maximized.

What makes a good internal enterprise marketplace?

Here are a few lessons we’ve learned by working with large organizations:

• Bake in governance from the start. Don’t make it optional. Every AI agent should meet security, compliance, and data-access rules before it can be published.¹⁹
• Centralize AI agent management, monitoring, and optimization. Think fully automated monitoring, version control, audit logs, and easy rollback.²⁰
• Make AI agents easy to find and trust. Tag them by function, department, risk level, and more. Implement transparent algorithms with robust ethical guidelines, continuous monitoring, and regular updates to ensure alignment with human values and societal norms. Include usage stats and feedback.²¹
• Encourage reusability. A good AI agent built for marketing might be useful in customer support too. Don’t reinvent the wheel every time. Design AI agents with modular, standardized interfaces to facilitate easy integration and adaptation across various applications.²²
• Support interoperability. Your marketplace should work with different frameworks and tools. Don’t force teams to conform to one vendor’s technology.²³

The importance of early adoption

If you want to scale AI agents safely and reliably, you’ll need a marketplace to manage them—just like organizations manage the data and apps their teams use. This will provide the foundation for making AI agents work at scale, with the right mix of control, flexibility, and visibility. Enterprises that take the time to get this right could be the ones that benefit from the AI agent wave—without drowning in it.