Why AI Agents Can Beat the Incumbents

Elena Berger introduces the episode by highlighting the challenges in enterprise procurement, emphasizing how the process involves multiple software systems, legal, finance, and supplier interactions that happen outside traditional systems of record like ERPs. She notes that AI startups have the chance to own the entire procurement process end-to-end, even if initial trust must be built gradually with their agents. Vlad Kyle explains that because no company starts with fully autonomous negotiation agents, an 'in the loop' approach allows systems to learn and gain trust through feedback, scaling up to handle large volumes of negotiations. This introduces the episode’s focus on how AI can revolutionize complex enterprise workflows.

SEMA Amble discusses the long-standing struggle for startups competing against incumbents who already dominate customer trust, data, and the system of record. She explains that recent innovations involve incumbents layering AI models on top of existing platforms, making competitive differentiation harder. Vlad Kyle adds that startups can focus on owning the entire end-to-end work process, especially tasks that happen outside the core systems of record—such as email, spreadsheets, and manual supplier conversations—that incumbents typically do not fully cover. This ownership of the broader process provides startups with a distinct advantage in building durable, vertical AI solutions.

Vlad Kyle details how Leo uses a multi-agent system to handle the complex procurement process, from sourcing, RFQ management, to negotiation, shipment, and invoicing. He describes how these agents communicate and coordinate to perform tasks across multiple stakeholders, software tools, and data sources. The system allows for fully autonomous operations in straightforward cases, while human oversight remains in more complex or high-stakes negotiations, such as those involving multimillion-dollar deals. Vlad emphasizes the importance of sharing information among agents and maintaining a flexible, end-to-end solution that adapts to different risk levels and deal sizes.

The conversation shifts to how enterprises gradually trust AI agents to handle negotiations. Vlad Kyle explains that initially, humans remain in the loop, providing feedback, which builds trust over time as agents learn from the process. For high-value negotiations, humans—such as expert engineers—collaborate closely with the AI, especially in analyzing drawings and setting prices. Vlad stresses that most work in large negotiations is preparation, often involving months of data analysis and stakeholder coordination, with AI aimed at automating as much of this as possible, including real-time insights during negotiations. Trust, both internal and external, is critical, especially when AI interfaces with third parties.

Vlad describes Leo’s approach to leveraging multiple underlying foundation models from various providers, integrating them with specialized harnesses for different tasks. He emphasizes that while general-purpose models are useful, some enterprise-specific tasks—like price benchmarking or cost modeling—may require fine-tuning or custom outcomes. Vlad notes that proprietary data, like unique quotes or engineering details, often can't be used to train generic models, so Leo focuses on outcome-based tuning (e.g., optimizing prices) to improve accuracy. This hybrid method ensures models are tailored to the nuances of enterprise procurement.

SEMA Amble discusses what makes an AI company durable and vertically integrated. She emphasizes owning the entire process, building trusted data assets, and expanding beyond initial use cases. She notes that winning trust and increasing the scope of work that AI handles leads to greater customer dependencies and competitive barriers. Vlad adds that AI companies should focus on gaining customer trust, establishing network effects, and automating increasingly complex tasks, transforming procurement into a core competency that drives lifetime value through deeper workflow integration.

The episode concludes with Vlad and Seema imagining a future where both buyers and suppliers use AI agents, even on adversarial tasks like negotiations or legal drafting. Vlad envisions AI on both sides improving coordination, transparency, and efficiency—such as sharing draft contracts or tracking issues. The conversation highlights that as AI becomes more embedded, it will facilitate seamless interactions across different parts of enterprise supply chains, making complex multi-party negotiations more efficient. Vlad also remarks on how enterprise procurement, often seen as unglamorous, could become exciting through AI-driven automation and process integration.