Other People's Interfaces

Last month Salesforce did something I think will look obvious in hindsight: it launched Headless 360, exposing its entire platform as APIs, MCP tools, and CLI commands. No browser required. Sixty-plus new integration points so AI agents can run the CRM without a human ever touching a dashboard.

When the world’s biggest SaaS company tells its customers “you don’t need our interface anymore,” that’s a confession about where the industry is heading.

Seema Amble at a16z at A16Z is clear on what’s happening: when the primary user of software is a machine, every moat built on human habit — frequency of access, training investment, the undocumented how-we-do-things-here that lives in workflow rules — starts to erode. An agent needs an API, context, instructions, and the ability to act. It doesn’t care about your dashboard’s color scheme...

The numbers make this feel less theoretical than it did even six months ago. MCP, Anthropic’s Model Context Protocol, has hit 97 million monthly SDK downloads with 78% of enterprise AI teams running at least one MCP-backed agent in production. Microsoft’s MDASH system — over 100 specialized agents coordinating across multiple AI models — just topped a major cybersecurity benchmark. No human in the loop. The agents scanned code, debated whether findings were real, and built proof-of-concept attacks.

But even individuals are now liberated from the constraints of Other People’s Interfaces (OPIs).

I have been using Monday.com for my personal task management for the past two months, entirely because it has a native integration with Claude. This means that I can ask Claude to interact with Monday.com from any of its interfaces (mobile, Claude code, web, etc), and it can view and manipulate my tasks. I had hooked up other systems before but they wouldn’t work from my phone, because Claude couldn’t access them if I was using the mobile app.

But the Monday interface is so comically horrible it almost has to be intentional trolling.

So I gave up and vibe coded some tools so my agents can manipulate Trello directly via the API instead. I can use Claude dispatch if I want Anthropic’s models to see my Trello board. Our team at Dent is also frustrated with the Monday interface, so our operations lead vibe coded a web interface to use for our team meetings that then syncs back to Monday.com with a button.

So if the UI disappears, what’s left? How is a headless CRM different from a well-designed Postgres database with an API?

Defensibility could live in the long tail. I’m using Trello instead of a postgress database because I don’t want to also build and manage a mobile app for quick entry and task management, and I don’t want to handle whatever server nonsense needs to go on to keep things up and running.

Or defensibility might live underneath the interfaces with well-thought out database design, connectivity, and agent-friendly documentation. Vibe coding means you don’t have to use Other People’s Interfaces anymore, but you still want good infrastructure.

Software has always been a little bit of a shoehorn experience for the user, so the idea that we can all have bespoke interfaces is kind of exciting.