Tuesday, August 4, 2026

The Agent Myth: Why Governance Wins the AI War

The hot take in B2B SW right now: You don't need enterprise software, you just let agents access the database and let them do the work. 

But several signals that are pointing to something different and more nuanced.






Jason Lemkin's team runs SaaStr on 3 humans and 20+ AI agents. His conclusion isn't that CRM is dead. It is the opposite: 20 agents are writing directly to a database and produce 20 different definitions of a qualified lead. No shared forecasting logic, no audit trail. It turns out the agents need the shared system of record as much as the humans do.

Salesforce's answer is to go headless where CRM acts as an API/MPC substrate, with Slack, voice, and custom UIs as interchangeable surfaces on top. 

Andre Wenz of SAP Signavio pushes the argument one layer up. He observes that today's application screens of pipelines, stage views, and approval forms aren't the process. Instead, they are cloud era artifacts, born when software was too expensive to rebuild per task. With AI, the interface itself is generated per objective and is disposable. The canonical records, permissions, policies and controls are the bedrock underneath. 


And then there is Celonis. They are already living the 'foundation, not the app' thesis on top of  any single ERP stack. Their Process Intelligence Graph mines the real processes across SAP, Oracle, Salesforce and other B2B appls and feed that context to agents via MCP. 

Put together, here is my read on where this goes: 

Generated, fluid, disposable UIs will become the norm; fixed screens will be limited to very few use cases and demos. 

The system of record and governance layers become more valuable because ungoverned agents are more dangerous than ungoverned humans. But rebuilding a system of record is hard and will take to time get deployed and distributed. 


The real battleground is the governance layer. 

No comments:

Post a Comment