Why AI Agents Can Beat the Incumbents

The complexity of aircraft manufacturing highlights the need for sophisticated procurement processes that go beyond simple transaction recording.

Early enterprise adoption of AI agents is cautious due to a lack of trust, necessitating a human-in-the-loop approach for gradual integration and learning.

An AI-native startup's advantage over incumbents lies in tackling end-to-end work, not just within a single system of record.

The discussion begins by framing the competitive landscape between incumbents and AI-native startups, introducing the idea that incumbents can layer AI onto existing platforms.

AI-native startups have an advantage because legacy incumbents are limited to their existing systems of record and do not complete the end-to-end job.

The complexity of procurement lies not just in the final negotiated price but in the extensive behind-the-scenes work involving numerous stakeholders, emails, and spreadsheets.

Seema outlines four types of AI agents: retrieval, process, policy, and principal, differentiating them by their level of judgment.

Incumbents are primarily operating at the retrieval and basic process agent levels, lacking the advanced judgment capabilities of policy and principal agents.

Incumbents are "held back" by their distribution advantage and customer trust, but also by internal incentive conflicts that hinder the development of more complex AI agents.

Leo's agents span all four categories (retrieval, process, policy, principal) and can operate fully autonomously or with human oversight depending on complexity and risk.

Leo quickly shifted from focusing on simple invoice process agents to handling exception handling, demonstrating an ability to adapt and address real problems.

The concept of internal versus external trust is highlighted, with external trust being crucial for convincing customers to let AI agents handle sensitive tasks like negotiation.

Procurement is described as an intense, multi-stakeholder process involving legal, finance, and cost engineering, which Leo's agents aim to coordinate.

Leo convinces customers by being ahead of the technology curve, identifying problems, and quickly shipping solutions, with trust being a key people component.

Leo's agents handle negotiations ranging from smaller deals to multi-million dollar contracts, often involving experts in the loop for complex analysis.

The human loop in negotiation involves cost engineers and other experts who provide insights into contract design, cost structures, and legal aspects.

Procurement's scope has expanded beyond simple transactions to include legal, finance, and various software systems, requiring coordination among specialists and generalists.

The complexity of procuring parts for industries like aircraft or robotics involves thousands of suppliers and meticulous coordination, where small delays can have massive financial consequences.

The operational back-office work in procurement, such as managing supplier confirmations, is critical, and missing even one email can lead to hundreds of millions in damage.

AI agents can intervene in physical world problems by predicting supplier reliability and protecting goods based on comprehensive context, rather than changing unchangeable events.

Agents need context from within an enterprise and the outside world (news, supplier information) to make informed decisions and help throughout the procurement process.

Leo's multi-agent system is designed to handle end-to-end tasks by enabling agents to share information and communicate effectively in a specific order.

Leo simplifies procurement by allowing users to upload photos or quotes, with agents handling the complex back-end tasks like checking inventory and drafting RFQs.

Direct procurement focuses on automation for strategically important, high-value items, while indirect procurement, like MRO parts or laptops, offers opportunities for autonomous negotiation and savings on previously unnegotiated spend.

Leo uses multiple AI models from various providers as a commodity, focusing on the "harness" (proprietary data and workflows) to achieve specific outcomes, especially for complex tasks like cost modeling and price benchmarking.

While general-purpose models can be helpful, certain use cases like negotiation and cost engineering require fine-tuning and outcome-based models that go beyond traditional LLMs.

The difficulty in codifying procurement intuition and experience highlights the need for specialized AI agents trained on outcome-based data.

The concept of "moat" is discussed in the context of AI, emphasizing defensibility through customer trust, deep integration, and owning more of the end-to-end work.

Customers are convinced to use Leo by demonstrating how it solves complex problems that can't be easily replicated by internal teams or simple model integrations, emphasizing the significant business value of their solutions.

While building basic AI tools is becoming easier, achieving high performance (70%+) often still requires specialized expertise and can inadvertently create more work.

The increased familiarity with AI tools is beneficial for the market, and enterprise companies are realizing that focusing on their core competencies is more efficient than building all internal tools.

Leo believes agents will eventually operate on both buyer and supplier sides, creating a more coordinated transaction ecosystem, and uses its influence over suppliers to encourage AI adoption.

The opportunity exists for Leo to own both sides of the transaction, extending into legal and other areas where agents can facilitate coordination.

While price negotiations can be zero-sum, the majority of procurement tasks have aligned incentives, allowing agents to automate these processes and accelerate time-to-market.

AI's ability to enable customization without slowing down the business is a key advantage, with forward-deployed teams automating deployment and customers having more self-service capabilities.

Leo's high proportion of engineers reflects its focus on building a product that reduces customization and maximizes self-service, automating even their own deployment processes.

The Bots and Buyers Summit highlighted that procurement tools have historically only made processes more efficient, not fundamentally changed how people work, leading to a focus on AI-enabled agents.

Procurement's pain points, combined with the massive business impact and its perceived boring nature, create a unique opportunity for AI to deliver remarkable value.

Excitement around procurement comes from its emotional nature, its significant business impact, and the fact that AI is a revolutionary advancement in a long-stagnant field.

Procurement presents a trillion-dollar business opportunity due to its critical role in building complex products, improving competitive landscapes, and driving P&L savings.