AI agents: which parts of software to keep paying for

By Nate from Nate's Substack · October 4, 2026.

The newsletter sets out a central idea: AI agents are splitting enterprise software into three components that used to be sold as a single package - the data, the interface and the workflows that turn that information into action. According to the author, the people on each team who already use AI are making de facto decisions, without management noticing, about which data matters, which steps in an application can be skipped, which workflows still deserve trust and which agent should handle each task. His argument is that much of a company's software strategy already exists implicitly in those individual decisions, and that the leadership team has yet to go out and discover it.

The piece invites readers to 'unbundle' what the company has been buying. A traditional software subscription combines data, an interface and the workflow rules that get the job done. When an agent becomes competent at searching, sorting or acting on that information, the customer suddenly has reason to rethink the entire package. But, Nate warns, each piece has a different future: you can stop using an application's screen and still depend on its records, its approval rules and its execution; you can build your own search process and still need a third party's data; and you can buy an agent because it is excellent at one specific task while preferring a different agent for the next one.

This, the author says, changes the software buying conversation. Companies must decide which capabilities to keep buying from third parties, where to invest their own engineering effort, and how much freedom they want to retain for when a better model comes along. On the selling side, it is no longer enough to say 'we have AI': you have to convince the customer that the company will keep improving their business as the underlying tools change.

This week's briefing is structured in four blocks, according to the index included in the email itself:

1. Learn what the teams already know: what questions to ask the people who already use agents day to day, so their workflows inform what gets standardized and which contracts get renewed.

2. Decide where owning the workflow pays off: how to tell a process worth building in-house -'The Firm's candidate search' is cited as an example- from another, such as payroll, that is better bought from an outside vendor.

3. Give buyers a reason to keep paying: what software vendors must demonstrate beyond 'we have AI', using Salesforce and Crustdata as examples.

4. Know what can actually be swapped: which parts of an agent workflow are portable if you change vendors, and why that switching cost shapes which improvements a company can realistically adopt.

Nate explains that all this analysis accompanies a video (linked in the original post) that develops how agents affect the value of software, while the briefing text focuses on the decisions that follow from it: what to learn from your own teams, where to build your own advantage, and how buyers and sellers should position themselves for models that keep getting better.

The email closes with a mention of the benefits of membership in Nate's Substack 'Executive Circle', which includes access to these Sunday briefings, to 'Nate's Library MCP' and to a private Slack chat for executives.