Interview, Podcast
The New Rules of Enterprise Software with Steven Sinofsky
- The market is shifting toward headless data access via APIs for AI agents, reducing reliance on traditional user interfaces as evidenced by a reported 300% increase in Slack bot and agent usage.
- Future value is expected to reside in logic and data storage rather than workflow software, with sticky applications identified as those handling financial transactions, long-tail processes, or complex business rules that are difficult to replace.
- Established enterprise systems like SAP, Salesforce, and Workday are predicted to persist rather than decay, with vendors likely adopting new AI functionality alongside existing legacy code rather than replacing it entirely.
- Agents are projected to handle "analyze" tasks and unstructured data via new usability layers, while systems designed for specific "do something" actions will face persistent challenges regarding credentials, impersonation, and exception handling.
- A misconception that legacy systems can be replaced by simple Postgres databases and APIs is expected to persist, as these systems contain customized, years-in-the-making logic that defines core business operations.
- The "SaaSpocalypse" is predicted to be overblown, with certain industries like insurance seeing 50-to-75-year-old COBOL software remain in place due to replacement difficulty and critical operational dependencies.
- Automation will increase productivity but expand rather than reduce workload, potentially creating new job categories, longer contracts, and increased demand for services as seen in radiology and expense reporting evolution.
- Challenges in capturing organizational context and exceptions remain significant, with the "80-20 rule" implying that 20% of edge cases often drive the majority of value and require human concurrence to resolve.
- Mid-term risks include potential degradation in customer service quality if AI stops defaulting decisions in favor of the customer, alongside instability in the emerging middleware layer as incumbents resist disintermediation.
- Startup opportunities are identified in vertical physical-world software, bridging two organizational functions, positioning between established players, and building agentic loops that collect benchmark data through outbound messaging.
- Three primary development paths are anticipated: building on top of incumbents like Salesforce, adopting a completely DIY approach, or constructing agents alongside existing logic, with the "New Way" exclusive focus recommended to avoid legacy constraints.
- Network effects in enterprise software are expected to occur primarily within individual companies rather than across them due to security and compliance requirements, driven by viral loops where users discover efficiency gains through AI tools.
- Data extraction from major platforms like Workday will become increasingly difficult without using the native platform, and the ability to extract all data cleanly will diminish as systems integrate AI directly into their core.