Build the AI-Services Buy Box Before You Buy

31 Aug 2026 · 1 min

Watch on YouTube

1:00

In this presentation Eric Siu outlines a strategic framework for acquiring AI‑driven service businesses. He contrasts two business models – “Copilot,” which sells AI tools, and “Autopilot,” which sells AI‑executed work – and explains how companies can start with outsourced, intelligence‑heavy tasks and gradually move to in‑source, judgment‑heavy work as AI capabilities compound. Siu argues that the rise of vertical SaaS embedded with AI, combined with cloud and fintech, creates a third wave where software can now perform tasks previously too complex for automation. He highlights the emerging opportunity of vertical large‑language‑model (LLM) agents that can serve specific functions such as finance, marketing, or sales. Crucially, Siu stresses the need for an internal investment thesis and a well‑defined “buy box” within the next 12 months, detailing underwriting standards, diligence processes, integration models, founder compacts, and performance scoreboards, to ensure disciplined and rapid acquisition decisions.

Chapters

Copilot sells the tool and Autopilot sells the work. Its playbook starts with outsourced intelligence-heavy tasks, then expands towards in-source judgment-heavy work as the AI compounds. You can do this for marketing, you can do this for accounting, you can do this for a series of different services. Uses 2 distinct formulations. You have vertical SaaS now with AI inside, argues that AI lets vertical software take on tasks previously too complex for software and describes a 3rd wave of cloud plus fintech plus AI. Software is eating labor is one of their arguments, right? And then YC finally, not last but not least, the accelerator calls the opportunity vertical LLM agents, right? You can have agents Again, for finance or for marketing or sales and things like that. And the whole idea here is that still, if we don't establish the thesis on our own first and establish over the next 12 months or what things should look like internally, then we cannot define what the buy box looks like, the underwriting standards, the diligence process, integration model, founder compact, scoreboards, all these things, right? And at least for us, as quickly as I want to move, it's more prudent to establish it more aggressively because we're learning so much as we're getting on these calls right now.