Jason M. Lemkin explains why context engineering matters most for AI agents. SaaS products have rich internal data—audit trails, user roles, contact details—that lets agents take real actions. An incomplete answer is worse than an inaccurate one because users can't see what's missing.
Chapters
I've gotta give it the right context. Walden Yan, the co-founder of Cognition Labs, the makers of Devin, an incredible AI tool, says context engineering, context engineering is effectively the number one job for engineers building AI agents. So you have all this content. It really can act as context for an AI agent. You are so well positioned for the future. Your SaaS application is like Cher's closet. The agent helps take action based on the content inside the closet. So why in the world would AI natives want to work for your old SaaS company? Besides wanting to work for cool AI companies, the good ones actually wanna work for companies where they have more access to data. Let me tell you a little bit, a little story. Uh, in our world, there are a ton, legal tech is such a hot area for AI today. There are so many competitors that have said, hey, give us your documents. Let me give you a simple suggestion or a simple explanation. Let's say that you wanted to ask our system, what's going on on this case and what should I do next? Now, your application might, might be, what should I do? You know, tell me about this opportunity for this customer and what should we do next? It's a very common question that you would use a chat application, a Copilot, to ask and get an answer. But to get a good answer to that question requires way more information. It requires you to know who's done what. Those are audit trails. It requires you to know who is who, who is doing the things in your application. For us, it requires conflict checks and contact information. You have to know so much to give a complete answer. And in the world of legal tech, and I believe almost for all of you in this room, an incomplete answer is actually worse. It's worse than an inaccurate answer because the customer doesn't know what they didn't see, and that is deadly. But the good news is, because you have such rich data, you can sell this to the AI natives you're recruiting, and then you can sell them on your distribution. And here's what happened. This is distribution from a real product that we launched with, uh, just recently.
