At first, the client considered using an off-the-shelf AI model to analyze its data and make optimization suggestions. After some experimentation, however, it realized that there was a problem: Most of the data resided in a complex ERP environment (specifically, one based on SAP), and generic AI models lack the ability to map ERP data onto business context. For example, they don’t fully understand SAP-specific data modeling and business semantics that vector engines and knowledge graphs enable. This lack of contextual insight meant that off-the-shelf AI solutions would be in a poor position to make truly optimal recommendations.
When genAI exploded onto the scene in 2022, enterprises worked through the typical calculus of whether to build their own AI solutions or buy technology from vendors. The vast majority arrived at a clear conclusion: Building LLMs in-house was too complex for most organizations, and “buy” was the better approach.
In short, although many businesses may assume they sorted out the build vs. buy decision for AI several years ago, the fact is that AI has changed in ways that can make solutions that are built in-house uniquely advantageous. This doesn’t mean that building is now always better than buying (it’s not), but it does mean that it’s high time for enterprises to reassess the value of building custom AI solutions, or working with partners to create them, as opposed to buying everything off-the-shelf from vendors.
What does it mean to build or buy agentic AI solutions?
I highlight this because it points to a third option, alongside building and buying. By working with a partner, organizations leverage the ability to create custom AI solutions, without having to do the hard work of building them on their own. Partners can also optionally help with operations and maintenance once custom solutions are up and running.
The company, a vertically integrated producer, refiner, and seller of agricultural products, wanted to use AI to help optimize its internal supply chain processes. It had tons of data available about those processes, but the company struggled to identify opportunities to improve crop yields by planting or harvesting at the ideal time based on weather patterns, for example, or to improve cash flow by prioritizing sales to customers with track records of settling invoices quickly.
Why the build vs. buy decision matters
Hence the importance of revisiting the build vs. buy conversation in the era of agentic AI (as opposed to the genAI-only age that preceded it). Read on for key considerations that business leaders should weigh, along with a real-world example highlighting the benefits enterprises stand to leverage when they choose to build agentic AI solutions instead of buying off-the-shelf offerings. We’ll also touch on a third potential path for businesses to consider alongside building or buying: Working with a partner to implement a custom AI solution.
That said, there’s another, more critical consideration at stake when thinking about build vs. buy for agentic AI: Business competitiveness.
To set the context: The build vs. by decision for agentic AI doesn’t mean that organizations either build all of the technology they use to deploy AI agents, or buy all of it.
To gain a real edge in the AI era, enterprises need to leverage AI technology in ways that their competitors can’t. And given that all businesses have access to the same off-the-shelf AI products, buying tends to offer less of an advantage than building a custom solution, especially for organizations that can’t gain an AI edge in other ways, such as by leveraging proprietary data to implement capabilities using off-the-shelf technologies that their competitors would not be able to emulate.
AI build vs. buy decision-making: A real-world example
Most take a more nuanced approach. They buy some components of their agentic stacks (like LLMs, which most enterprises source from the leading AI labs) while building others (like agents and orchestrators). Some take an even more nuanced approach by, for example, building some agents to address bespoke needs while purchasing off-the-shelf agentic platforms to support other, more generic use cases.
In the case of the example I just described, the enterprise didn’t actually build a custom AI solution entirely on its own. It partnered with my company to design and implement the product.
For businesses that choose to build AI apps or services and have the in-house expertise to do it, that approach is perfectly viable. But for those with limited data science and AI engineering resources, partnering offers a means of creating AI products that (unlike generic AI solutions) deliver a competitive edge.
The calculus that businesses should work through when they compare building vs. buying agentic AI technology includes the same considerations that they’d weigh when making any build vs. buy decision. They need to think about the TCO of an in-house solution versus one supplied by a vendor. They should evaluate the flexibility and customizability of each route. They must weigh how well positioned they are to manage the solutions they choose once they’re running.
Partnering instead of building: A hybrid approach
To drive home why it’s so important to get the build vs. buy decision right, and what the decision-making process looks like in practice, allow me to share the example of a client I recently worked with.
Fast forward four years, however, and companies are once again facing a “build vs. buy” decision around AI. Agentic AI has added a great deal of complexity and breadth to AI technology stacks. Although it’s still the case that almost no one is building their own LLMs, organizations are realizing that it can make sense to build other technology components themselves.
So, we set out with the client to build a custom AI agent optimized to support its unique data assets and use case requirements. The agent connects to third-party, off-the-shelf AI models, but it includes internal logic that allows it to interpret SAP data in a context-aware way, using MCP servers to interface between the models and the backend SAP environment.
Conclusion: Build vs. buy decisions as a linchpin of AI strategy
This is the most important reason why the build vs. buy decision for AI has assumed newfound importance today. Smart organizations have realized that, in a certain sense, AI is only as good as the unique advantages it confers, and achieving unique advantages often requires the development of custom AI applications or services.
