Is Software Losing Its Head?

What It Covers

A16z partners Seema Amble and Steven Sinofsky examine what happens to enterprise software when AI agents replace humans as the primary users. They analyze Salesforce's "Headless 360" announcement, why SAP and similar platforms cannot be replaced with a Postgres database plus APIs, and where startup opportunities exist during this architectural shift.

Key Questions Answered

• •Headless software misconception: Salesforce's Headless 360 announcement was largely a rebranding of existing APIs rather than a structural change. The real shift is that AI agents access CRM data via API rather than UI, evidenced by a reported 300% increase in Slack bot usage. Founders should evaluate whether incumbents are genuinely restructuring or simply relabeling existing capabilities before building displacement strategies.
• •Why SAP cannot be replaced with APIs: Enterprise platforms like SAP encode decades of business logic, compliance rules, and exception-handling workflows that took years to implement and customize per company. A Postgres database plus API layer captures data storage but none of that embedded logic. Startups attempting direct replacement face the equivalent of rebuilding a running engine mid-flight without access to the original schematics.
• •Exception handling is the actual product: The majority of enterprise software value lives in edge-case resolution, not standard workflows. Every enterprise pricing model, sales interaction, and compliance check involves exceptions that are never formally documented anywhere. AI agents currently lack access to this context, which exists only in employees' heads, making voice agents and computer-use agents that observe and record human behavior the primary mechanism for capturing it.
• •Productivity creates new work, not less: Automating a business process does not shrink the total workload — it generates new analytical layers. Amazon's automated returns process eliminated phone-based customer service but created a continuous back-end optimization loop requiring new tooling. Expense reporting automation evolved into full business travel performance optimization. Founders should build for the next layer of complexity that emerges after automation, not just the automation itself.
• •Startup positioning between incumbents: The highest-probability startup opportunity during a technology shift is targeting the gap between two established enterprise players rather than attacking either head-on. Incumbents will bolt AI onto existing product lines without restructuring them, creating blind spots in between categories. HTTP succeeded not by replicating client-server features but by implementing the concept entirely differently, bypassing a trillion dollars of legacy investment.
• •Cross-functional bridging as a new software category: Enterprise software has historically sold into single functions — sales, finance, HR. AI enables tools that bridge two organizational functions that previously required manual integration or systems integrators like Accenture. Figma's bridging of design and product development is one precedent. Startups that use AI to connect functions sharing data handoffs — such as sales-to-finance or procurement-to-operations — are building a structurally new category with defensible network effects inside the enterprise.