Field Notes
I was explaining Niche to my brother recently and caught myself saying something that would make most SaaS founders wince: the human interface is not the most powerful way to use the product.
That is a strange thing to admit after spending thousands of hours building that interface. The normal founder story is that you obsess over every screen and every empty state. You cut features, hide complexity, collapse choices, and polish the flow until a person can move through it without feeling like they are operating machinery. That work matters, and I still believe in it. A creator should be able to open Niche, type a niche, read a few strong stories, pick an angle, and publish something worth reading. That is a real product, and it works.
But if I am honest, the more powerful version of Niche is not the one where a creator clicks through the UI. It is the one where an agent uses the toolset on the creator's behalf. That is the uncomfortable truth, and the more I use the product myself, the more convinced I am that it is the right shape.
Niche has a simple promise on the surface. You tell it your niche and it helps you find what is worth saying. It scans live conversations, news, and pre-news signals, surfaces stories, proposes angles, and generates platform-native content. It can render assets, revise drafts, and prepare work for review.
For a human, all of that has to be packaged carefully. You cannot expose every story, every score, every source, every alternate angle, every cost estimate, and every branch in the workflow. That would be exhausting software. A human interface has limits, not because people are not smart, but because attention is finite. Past a certain point, more buttons and more fields and more data just make the product worse.
This is what I mean by buttonology: what happens when powerful software tries to expose all of itself through a human interface. Every new capability becomes another button, every edge case another toggle, every option another setting. Eventually the product is powerful and miserable to use. Creators do not want to fly a 747 to publish a good post. They want momentum. So the UI has to compress, recommend, and hide complexity until it is needed. That is good design, and it is also a ceiling.
An agent does not get overwhelmed by a dense payload. It does not need a beautiful card layout or an interface that protects it from reading too much. It can inspect the full state of a session, weigh the options, hold the creator's instructions in context, and come back with a recommendation.
That changes what I should build. For a human, I might show three stories because five or ten would feel like work. For an agent, I can return all of them, including the ones I rejected, along with source quality, freshness, brand fit, conflict flags, platform fit, estimated cost, and the reasoning behind each. None of it clutters anything, because there is no screen to clutter. The simple human path of type your niche, pick the story, pick the angle, review the draft becomes something much richer: scan the niche, compare everything surfaced, prioritize what fits the creator's point of view, avoid anything that conflicts with the brand, down-rank the overexposed, pick the strongest angle for LinkedIn, draft it, check source faithfulness, revise the hook, prepare a second cut for X, and hand back a recommendation. That is not just faster. It is better.
A lot of people assume agent-native software just means an agent clicking the same buttons a human would have clicked. I do not think that is the opportunity. An agent operating the human UI is trapped inside the human UI. It inherits all the simplifications and hidden data and missing branches that exist to keep the product usable for a person.
The real opportunity is giving the agent the underlying primitives. Not the buttons. The tools. A button says "Generate." A tool says: here is the session state, here are the stories, here are the angles, here is the brand profile, here are the source warnings, here is the revision function, here is the cost telemetry, here is the dry-run publish check, now decide what to do next. That is a completely different level of capability, and it creates a different relationship between the creator and the software. The creator stops operating every step and starts directing the work.
That is already how serious people work with human teams. A founder does not ask a strategist to click five buttons. They give context, constraints, and judgment, and the strategist comes back with options and a recommendation. Agents make that same pattern possible inside software. The creator can say: find the strongest story in defense tech this morning, skip the generic hype, lean into my view that decision support matters more than automation, draft it for LinkedIn, but show me the angle first. That is a better interaction than any button sequence I could design.
None of this removes the person. It does the opposite. The creator brings the taste, the judgment, the constraints, and the final approval. The agent handles the mechanical complexity. Niche provides the editorial intelligence layer underneath. Creator, agent, and engine, each with a clear job: the creator knows what they believe, the agent knows how to operate across tools, and Niche knows how to turn niche-specific signal into credible content.
This is why I keep coming back to the word headless. Headless software is not software without a UI. It is software whose value is not trapped inside the UI. For Niche, the core product is not the dashboard. It is the editorial engine beneath it. The dashboard is one way in. An agent is another, and for power creators, founders, analysts, and anyone already building AI into their daily workflow, the agent path may be the better one.
That is a humbling thing to accept as a founder. You spend all this time making the product friendly for humans and then realize your most powerful customer might be a machine acting on behalf of one. I do not read that as a failure of the interface. I read it as a sign that the product is deeper than its interface. The job of the UI is to make Niche approachable. The job of the agent surface is to make it fully usable. Both matter, and they are different jobs.
The future of SaaS is not prettier dashboards. It is products with an intelligence layer that humans can trust and agents can operate, where the best software stops forcing every capability through a button and instead exposes real tools beneath a simple human experience. That is what I am building with Niche: a content desk for creators, a toolset for agents, and one editorial engine for both.
More field notes