The Agent Platform Explosion
Every major enterprise software vendor has announced or shipped an "agent platform" in 2025. Salesforce, Microsoft, ServiceNow, and dozens of startups are selling the promise that you can deploy AI agents without building them from scratch. The pitch is compelling: configure, do not code. But the reality is more nuanced than the demo suggests.
Why Agent Build vs. Buy Is Unique
Traditional SaaS build vs. buy is relatively straightforward. The software does a defined thing. You evaluate whether that thing matches your needs. If yes, buy. If no, build.
AI agents are different because they interact with your systems, your data, and your business logic in deeply integrated ways. The "buy" option requires so much customization that it often feels like building anyway, but with someone else's constraints.
- Agent behavior is your business logic. When an agent decides how to route a request, what to prioritize, or how to handle an exception, it is encoding your business rules. Buying a platform that imposes its own logic means your competitive differentiation lives in someone else's configuration layer.
- Integration depth determines value. An agent that can only interact with the vendor's own product suite (Salesforce agents that only work with Salesforce, for example) is useful but limited. Real agentic value comes from cross-system orchestration, and that almost always requires custom work regardless of the platform.
- Control over model selection matters. Many agent platforms lock you into a specific model provider. Given how fast the model landscape is moving, this lock-in can be strategically dangerous.
- Observability requirements are high. When an agent makes a decision, you need to understand why. Vendor platforms often provide limited visibility into the reasoning chain, making debugging and compliance difficult.
When to Buy
Buy an agent platform when: the workflow is common across your industry (support, IT helpdesk, simple procurement), speed of deployment matters more than differentiation, and you are comfortable with the vendor's model and integration constraints.
When to Build
Build when: the agent's behavior is a competitive differentiator, you need deep integration across multiple internal systems, model flexibility is important, and you have the engineering capacity to maintain it.
The question is not "build or buy." It is "where is the boundary between commodity and differentiation, and does the vendor's platform respect that boundary?"
The Evaluation Framework
For each agent use case, score it on four dimensions: how differentiated the workflow is (1 to 5), how deep the integration requirements are (1 to 5), how important model flexibility is (1 to 5), and how critical observability is (1 to 5). Scores above 12 lean build. Scores below 8 lean buy. Everything in between requires honest analysis, not vendor demos.