The Agent Graveyard Is No Longer Reality

Enterprise adoption of AI agents is rapidly moving beyond the experimentation stage into actual operations and purchasing. In the past, about 95% of generative AI pilots failed to deliver measurable financial results, but the situation has changed as model performance and purchasing methods have matured.

According to a survey of 500 US enterprise buyers by Menlo Ventures, 47% of AI deals converted to operational stage. The conversion rate for traditional SaaS was 25%, and 76% of enterprise AI use cases are being adopted through external product purchases rather than internal development. This is a significant change from the nearly 50-50 build-versus-buy split in 2024.

While traditional SaaS sold features, AI sells work outcomes. When AI deployment works properly, value becomes apparent immediately after signing or even within the same week, reducing the chances of opportunities disappearing during lengthy procurement processes.

AI buyers no longer make judgments based on demos alone. Since most AI demos look good, actual evaluation takes place within the product itself.

• Shadow mode targeting real-time traffic
• Evaluation using actual historical conversation data
• Running a portion of traffic that includes real customers

Because products are verified in real environments even before contracts are signed, the evaluation itself becomes proof of performance. This approach raises the verification burden on vendors, but it is also a factor that boosts contract conversion rates.

The build vs. buy decision for AI agents has also changed. While building a customer service agent from scratch is now achievable for any capable engineering team, the ongoing improvements needed to keep pace with policy and product changes, analyzing millions of conversations, simulation and regression testing, and QA for edge cases are far more difficult and never-ending.

As a result, enterprises are moving toward purchasing an operational layer for iterative improvement rather than the agent itself. Enterprise business teams can directly adjust behavior and deploy changes without engineering tickets, meaning that buying no longer means giving up control.

On the other hand, not every project succeeds. A representative cause of failure is when ROI is not clearly defined. Even if an agent supports workflows or improves team efficiency, if that impact is difficult to connect to the bottom line, it becomes hard to secure investment from management.