Mashery Founder's Barndoor: AI Agents are the New APIs

If Application Programming Interfaces (APIs) served as the vital connective tissue binding disparate websites and applications during the Web 2.0 era, then AI agents are rapidly emerging as their counterpart in the burgeoning age of artificial intelligence. These intelligent agents are poised to become the primary conduits through which users access and leverage data from a multitude of internet sources. While APIs will undoubtedly retain their significance, AI agents are undeniably capturing the spotlight.

This paradigm shift forms the core thesis of Oren Michels, CEO of the new AI company Barndoor. Michels, a familiar name in the tech landscape, previously founded Mashery, a pioneering API management firm he steered from its inception in 2006 until its acquisition by Intel in 2013. He embraces the comparison between the two eras with characteristic wit, suggesting that removing the “P” from “API” (implying a move beyond programmatic interfaces to more intelligent, autonomous ones) unveils a far more intriguing future.

Barndoor positions itself as the “control plane for agentic AI.” Just as Mashery empowered companies to systematize and manage their APIs, Barndoor aims to equip enterprises with the tools to tame and govern their AI agents, establishing crucial guardrails for their operation. Michels contends that existing solutions for managing AI agents, including traditional Identity and Access Management (IAM) and API management platforms, fall short. He argues that if these solutions were truly adequate, the enterprise adoption of agentic AI would be far more widespread than it currently is. Drawing a parallel to his Mashery days, he likens the situation to assuming Cisco, merely by managing network traffic, was also handling API management. It was precisely because they weren’t that companies like Mashery found their essential niche.

The typical users of Barndoor, according to Michels, are not necessarily CIOs or security personnel. While these roles are increasingly impacted by AI, their core function isn’t to directly deploy or manage AI solutions. Instead, the “AI problem” resides with business roles, such as sales or marketing professionals, whose objectives involve faster and more efficient operations powered by AI. He illustrates this with an analogy from the API era: the person at Marriott tasked with enabling hotel room bookings via a mobile app was the one with an “API problem.” Similarly, in the AI era, those whose job is to sell products and require AI to achieve that more effectively are Barndoor’s target audience.

Though Barndoor has recently launched with a waiting list, early use cases are already emerging. Michels describes these as agentic workflows that necessitate interaction with various enterprise tools. He envisions agents as a “robot workforce,” requiring the same rigorous identity management for tool access as their human counterparts. Just as human employees gain access to platforms like Salesforce, Notion, or Gmail to perform their duties and generate work product, clients are seeking similar capabilities for their agentic tools. However, Michels stresses the importance of a cautious, incremental approach. Allowing agents access to enterprise tools requires starting small, being deliberate, and imposing strict limits on their “blast radius” to ensure close monitoring of their actions and intentions.

The nascent nature of AI agents, despite intense hype, prompts questions about the types of tools early enterprise customers want them to use. Michels notes that the initial wave of use cases resembles “Robotic Process Automation (RPA) on steroids”—automating repetitive, rule-based tasks with greater sophistication. Yet, the Barndoor team has found limited long-term value in merely accelerating human actions. More compelling use cases are emerging around the Model Context Protocol (MCP), which Michels likens to Representational State Transfer (REST) in the API world—a widely adopted architectural style for APIs in the 2000s. He describes MCP as a more efficient and rapid means for computers to communicate and accomplish tasks. The process often involves accessing an API, converting it into an MCP server, and then granting AI access to it. Barndoor effectively acts as a proxy to these MCP servers, orchestrating the workflow between human workers and agents, mirroring the OAuth authorization flows common in API management.

Michels holds an ambitious vision for the agentic AI market, predicting it will rival the scale of the Software as a Service (SaaS) market, a multi-trillion-dollar industry. He even anticipates that agentic AI could supplant certain SaaS categories, particularly those in areas like customer service. However, for other SaaS categories where human oversight and control remain paramount, such as core business systems like Salesforce, Notion, or Gmail, coexistence is the likely path. Michels emphasizes that humans will continue to use these tools as a “source of record” and “one truth” for organizations. The role of AI, he suggests, is not to entirely replace humans but to provide “superpowers,” enabling them to achieve far more, faster. Ultimately, these core SaaS tools will increasingly be utilized by agents alongside their human counterparts.