Headless SaaS and using my own product less(!)
A totally weird and fascinating quirk of the current “agentic moment” is that I am now actually using my own product’s UI less and less. Suffice to say, this is a pretty weird thing to say.
Let me explain. Today I had to spend some time or so working through some analytics data inside our product with a customer (very B2B SaaS-coded), but instead of what I would normally do, which is use our own reporting and insight features (that we’ve built inside the product), I instead turned to my agent of choice, and did all the analysis with the help of a trusty agent, powered by the Komo REST API and CLI.
Working with an agent on work like this, does truly feel like working with a person with superpowers (not to mention endless patience). “Hey, can you go and pull all those records, sort them by this and that, then see if there are any interesting patterns in there?”. I even find myself interrogating the work too, just like you would a real person. “How did you get to this number? Can you show your working?”. It’s fascinating to watch an agent work on a task like this. To reason through the problem, the task and the path to goal. To explore different solutions and tools that may be use, and even all the dead ends and trial and error along the way.
The implications of all of this are pretty wild. Those beautiful analytics and insights features that we put so much effort into, may suddenly actually not always the best way to interface with the data in our product. Now, not all users are going to want to work with agents for every task, clearly, so I’m not here to declare the death-of-the-interface, but there is little doubt in my mind that products that don’t enable this kind of headless workflow, are going to put themselves in an extremely vulnerable position moving forward, as power users increasing look to agents to do the work with and for them.
It’s beginning to become clear to me what people mean when they talk about a future where the “user” is an increasingly an agent, not a human. What a time.


Agents are excellent to explore, cut, slice data… may be yet not for modifying records. Which tasks do you think are still more suited for traditional UI?