What happens to software when the primary user is no longer a human, but an AI agent? For decades, the “stickiness” of enterprise giants like Salesforce and SAP has relied on user interfaces, muscle memory, and the complex ways humans interact with screens. But as we move toward a “headless” world—where agents read data, update records, and complete workflows via APIs—the value shifts entirely from the visual interface to the underlying business logic and data integrity.

The conversation clarifies that while “headless” may sound like a simple rebrand of existing APIs, it represents a fundamental shift in how companies operate. True enterprise stickiness isn’t just about a pretty UI; it’s about the codification of complex business rules and regulatory requirements. For instance, replacing a system like SAP isn’t as simple as moving data to a new database because SAP contains years of customized logic that defines how a global company actually functions. Attempting to “vibe code” a replacement without understanding these deep operational dependencies is a common startup pitfall.

Looking forward, the real opportunity for new software isn’t in replacing these incumbents head-on, but in managing the “long tail” of exceptions. Most enterprise software handles the 80% of standard cases well, but the actual business is run on the 20% of exceptions—the edge cases and manual workarounds. AI agents have the potential to synthesize unstructured data and automate these exceptions, creating new “systems of record” by observing how humans actually solve problems in the field rather than how they are told to enter data into a CRM.

Surprising Insights

• The “SaaSpocalypse” is exaggerated: The idea that AI will simply replace all SaaS is unlikely because productivity gains usually create new, more complex demands rather than eliminating the need for software.
• The “Export to Excel” paradox: The two most used features in enterprise software are often the ones that allow users to leave the software (CSV and PDF exports), proving that native UIs often fail to provide the specific analysis users actually need.
• Inertia as a product feature: Software stickiness often comes from “arcane” details—like how Outlook handles recurring meeting exceptions—which aren’t planned by PMs but become impossible for a company to migrate away from.
• The “Middle” Strategy: The biggest opening for startups is not to compete directly with a giant, but to sit between two established players or two different internal corporate functions that don’t communicate well.

Practical Takeaways

• Focus on the “Long Tail”: If you are building AI for the enterprise, don’t try to automate the mundane 80%. Instead, build tools that handle the complex exceptions and edge cases where humans currently spend most of their time.
• Avoid “Vibe Coding” Enterprise Logic: When disrupting a legacy category, remember that the value is in the business logic and compliance rules, not the data storage. Map the undocumented “Standard Operating Procedures” before attempting to replace the tool.
• Leverage Unstructured Data: Use LLMs to tap into the “untapped resource” of a company—the Word docs, Slide decks, and emails—to create a layer of visibility and intelligence that incumbents are too slow to implement.
• Build Bridges, Not Just Tools: Look for opportunities to create software that facilitates a handoff between two different organizational functions (e.g., Finance and IT), as these integration points are typically underserved and highly valuable.