At the 2025 Norm Ai Central Park AI Forum (CPAIF), two conversations captured key dimensions of the buy vs. build choice. First, Henry Moniz, Chief Compliance Officer at Meta, in a fireside chat with John Nay, founder and CEO of Norm Ai, reminded us that enthusiasm to build doesn’t translate into long‑term ownership. Directly following that, Philippe Laffont, founder and CEO of Coatue, took the stage with Lawrence H. Summers, former U.S. Treasury Secretary, and continued on themes related to talent, focus, and the high-stakes nature of enterprise AI deployments.

“If you're an engineer, you might be onto the next sexy thing you want to build. You might not necessarily be pushing [...] updates and improving. It's not as fun to maintain as it is to build.” — Henry Moniz, CPAIF 2025

Building software is fun. Maintaining software is just as important, but not nearly as enjoyable. Similarly, building a prototype is easy, but building production-grade software that can be used in mission critical situations, e.g. for legal and regulatory compliance, is a totally different endeavor. The optimal incentive structure to maintain and constantly improve production-grade software is only found inside of growing technology companies that have bet their future on a vision that requires that specific thing. Only then can you attract world-class talent and keep them driving toward a prime directive to build and support that focused mission.

“There's a cost benefit, right? Like if you have engineers, we have a lot of engineers. All these companies have a lot of engineers. But do you want them working on products that are accretive and commercial and they can really build the bottom line?” — Henry Moniz, CPAIF 2025

The ability to build high quality production-grade software is surprisingly scarce. Companies need to allocate this scarce resource like anything else: where it is likely to generate the highest differentiating value for them. Building legal and compliance software is rarely a high potential return on investment because it is unlikely to be a differentiating asset for any company other than a law firm or compliance consulting firm.

The ideal setup is to find a third party firm that is all in on Legal & Compliance for in-house teams, but at the same time, is not a small point solution: that is the other trap to avoid. If a company uses point solutions that are too narrow, coming from a plethora of small vendors, then they lose the ability to consolidate data, relationships, change management, and focus around a more unified platform. Ultimately, they end up needing to change things up in a painful way. Furthermore, if a vendor is too niche and narrow, they will not be able to attract and retain world-class talent through profitable growth.

An AI-specific buy vs. build trap that will grow in size in the coming years is working with vendors that are a thin wrapper on top of the foundation models. Buying access to OpenAI or Anthropic from a vendor that is not themselves going deep into specific workflows on a company’s behalf is worse than simply calling upon OpenAI more directly. These underlying foundation models continue to rapidly improve and the most effective way to index into their chatbot modality deployment is to go direct.

The optimal solution for specific AI-driven workflow automation is to collaborate with a partner that is (1) fully focused on a certain category, e.g. legal & compliance AI for in-house teams at large enterprises; and (2) adaptive and open to implicitly co-developing additional functionality alongside their clients. That latter adaptiveness point is what enables companies to capture the benefits they would have with building in-house solutions, namely, influence over the broadening roadmap of integrated solutions, without the drawbacks of building something outside of their core wheelhouse. To sustain that adaptiveness, the partner needs to be growing. They need world-class talent, and they need to strike the right balance between focus and flexibility.

Lawrence H. Summers (OpenAI board member; former U.S. Treasury Secretary; former President of Harvard) emphasized many of these points at the Forum.

“I was Treasury Secretary during Y2K, so I was kind of in charge of making sure that the economy didn't collapse on January 1st my general observation was that the people who had proud IT departments that thought they could figure out how to do it themselves and they would restructure their own systems because only they knew their own systems tended to spend five times as much money not do as good a job as the people who went to the established vendors who had checklists, had seen it all, seen all the problems in three, seven other places, and were in a position to work with those companies.

But I think there's a big danger in trying to do too much of this yourself, and it's reinforced by two other considerations. It's reinforced by what was mentioned on the previous panel [by Henry Moniz, Meta CCO] that your talented person who built a great system is not gonna think it's that interesting to hang around and maintain the great system.

And if they really did build a great system, somebody else is gonna hire them to build another great system for them for a lot more money than you're gonna want to pay them to maintain your system. And this is something you learn a lot about when you're in Washington.

And if you screw up and you screw up using one of the four systems that everybody uses, then it's really too bad. You screwed up. But you used one of the four systems that everybody used. Screw-ups sometimes happen. If you invented your own system 'cause you thought you were really clever and your company was different and your culture was different and you had an IT guy who understood your specific culture and they did it and it screwed up, it's a lot harder to go and get sympathy.”

The higher-stakes an enterprise deployment, the more critical it is to lean on teams that are seeing across the entire industry. For functions like customer support AI agents and marketing AI agents, the buy vs. build equation can net out toward building. As an AI touchpoint gets closer to mission critical, turning to those that are tapped into the broader industry and regulatory landscape increases in importance. This is analogous to why high-end law firms and strategy consulting firms will always have a place in the Fortune 500 arsenal of business tools. As a concrete example, it is common for large enterprises to build a chatbot in-house with a foundation model provider, but the regulatory and legal checks on the highly regulated touchpoints that chatbot would have are rarely implemented by in-house builds.