At a glance
WHAT IT’S REALLY ABOUT
How AI agents reshape enterprise stickiness beyond UI and APIs
- “Headless software” is largely a rebranding of API-first access, but it reflects a real shift: agents (and chat interfaces) increasingly interact with systems of record without using the traditional UI.
- Enterprise software stays sticky less because of screens and more because it encodes business logic, compliance requirements, and organization-specific exception handling that is difficult to replicate or replace.
- Agents introduce a three-part capability spectrum—lookup, do/modify, and analyze—each with different enterprise constraints around identity, permissions, liability, and hallucination/verification.
- The “replace SAP with Postgres + APIs” narrative is rejected; the hardest and most valuable layer is the embedded process logic and customization that effectively defines how large companies operate.
- The biggest startup openings are “in-between” incumbent categories and across organizational handoffs, using AI to create new action loops (collect→decide→act→learn) rather than trying to rip-and-replace incumbents head-on.
IDEAS WORTH REMEMBERING
5 ideasHeadless is less about removing UI and more about shifting the primary interface to agents.
Salesforce’s “Headless 360” is framed as a marketing wrapper on existing APIs, but it signals a broader change: bots, chat, and agents increasingly become the entry point to CRM/ERP data and actions.
Don’t conflate “agent,” “API,” and “MCP”—they imply different enterprise risks.
Sinofsky proposes a practical lens: lookup is easy, “do” requires identity/seat/credentials and audit trails, and analyze is powerful but must address verification and hallucinations—especially across multiple systems.
The moat in enterprise is the encoded business logic, not the database schema.
SAP’s durability comes from years of customization that codifies how a company buys, plans, hedges, hires, and closes books; “Postgres + APIs” misses the depth of rules, controls, and process dependencies.
“Vibe coding” into enterprise fails because maintenance and edge cases dominate.
It’s feasible to prototype internal tools, but production enterprise systems must survive long-term ownership, permission models, audits, integrations, and constant process evolution—where the long tail of exceptions is the real work.
Exception handling is the entire game for agentic automation.
Most meaningful enterprise work is non-standard: regional policies, special customer handling, pricing exceptions, and approvals; agents need context beyond fields—often trapped in tribal knowledge, emails, and documents.
WORDS WORTH SAVING
5 quotesLike, first you have this, you have a- agent, which as far as I can tell right now, is also a new word for program that takes a very long time to run and might not finish.
— Steven Sinofsky
Almost everything interesting in an enterprise is an exception.
— Steven Sinofsky
The minute that you can get something easier with automation, and you can actually automate it, which I do think is happening right now with agents and with, with language models, well, then we're gonna dream up a whole bunch of new stuff to do.
— Steven Sinofsky
The minute that you automate the most mundane thing and think you have it all squared away, whole new things appear.
— Steven Sinofsky
There's this wild underestimation about like you could vibe code your way into enterprise software.
— Steven Sinofsky
High quality AI-generated summary created from speaker-labeled transcript.
