In a World Full of AI Tools, Where Does the Work Actually Live?
by Rebecca Bassett, Head of Core Product Marketing, Superhuman Docs and Databases
There is a part of the AI story that we’re not talking about enough.
As AI makes it trivial to build software, we're heading toward a world of dozens of purpose-built apps, each solving a real problem and delightful to use. But each one also comes with its own interface, logic, and expectations for how you show up and engage with it. Even if you could easily keep track of which app does what, you'd still be learning a new system every time another one launches.
We spend a lot of time marveling at what AI can build and how fast it can do it, but we're not spending nearly enough time thinking about what that means for the people who actually have to work with all the outputs.
Even agents need a surface
The intuitive answer to the proliferation problem is agents. If AI can handle more of the execution in the background, maybe the surface problem solves itself. This is the logic behind what some companies are calling headless apps: Strip away the front-end, let agents work directly with the underlying data, and give people a minimal interface just for signing off on what agents have done.
I think that gets it backward, though.
The surface isn’t just where execution happens. It’s where people reason together. Where a team decides what’s true, what matters, and what to do next. Agents can do extraordinary things in the background, but at some point that work has to land in a place where humans can see it, question it, and build on it together. And when it doesn’t, we risk losing the thinking, the iterations, the decisions that shaped the output. You get the final artifact without the people or the thinking that led to it.
The case for familiar infrastructure
This tension isn’t new. Coda’s founding thesis was that despite a very long list of point solutions—a tool for every job, a dashboard for every team—people kept returning to docs and spreadsheets to get things done. Not because those tools were best in class at any specific task, but because they are consistent and widely understood. You could reliably say, “I’ll write something up and share it in a doc,” and the person on the other end knew what to expect.
That kind of familiarity turns out to matter a lot more than it might seem. There’s a real cognitive overhead to every unfamiliar interface, and when you’re context-switching between five or 10 AI tools, that overhead compounds. You stop doing your best work and start managing your tools.
I think those same observations still apply today, maybe more than ever. If organizations are going to have both humans and agents working together, they still need some durable, familiar place to get work done and make decisions together. The building blocks that have always made sense are still the ones worth betting on.
The doc as common denominator
When I think about which surface actually holds up, the answer is pretty unglamorous. A doc is where context lives and accumulates. Where the reasoning stays attached to the output, and the team can see not just what was decided but how. It’s not the most technically impressive surface, but it’s the one everyone already knows how to use, and it gets more useful as the work grows, not more complicated.
This doesn’t mean abandoning the AI tools your team already uses. The whole point is that whatever Claude, ChatGPT, or Cursor produces can flow back into the doc, keeping the work connected and the team in sync. The doc isn’t a walled garden. It’s where everything lands.
To me, that’s what makes it the right bet right now. Not because it’s the most powerful option, but because it’s the one that keeps people connected to the work and to one another. At some point, you need a place that doesn’t just hold the work, but also holds the people doing it together.
