Baxter

Why we built this

We built it for ourselves first. Every time we needed a new type of content, we'd define the format, set the voice, wire up the constraints, and build a repeatable system around it. Venture posts, case studies, social content — all produced consistently because the rules were already encoded.

It worked. But the setup required thinking like a prompt engineer. You had to know how to structure a prompt, how to feed the right context and build a feedback loop that actually improved output over time. If you couldn't do that, you were stuck with the blank chat window. Type what you want, hope it comes back right, fix what doesn't.

That's the real problem with AI writing tools. Not that AI can't write. It can. A Cornell study published at CHI 2025 showed something more specific: AI writing assistants actively homogenize output. They strip the specifics and replace them with generic descriptors. Across tools and across users, everything starts sounding the same.

The blank-page approach makes that worse. Without structure, without voice rules, without a format to follow, the AI defaults to its training data. And its training data sounds like everything else on the internet.

Baxter started as our own writing companion. We rebuilt it so everyone could have one.

How it works

You start by defining your content types. Blog posts. LinkedIn updates. Case studies. Newsletters. Whatever you create regularly, each one becomes a content type in Baxter.

Each content type gets three things. A format, which is the blueprint for how that content is structured. A voice — the tone, style, and rules that make it sound like you. And a strictness setting that controls how closely Baxter follows the format. Tight for a press release. Loose for a thought piece. You decide.

Then you tell Baxter what to create. It drafts. You review. You approve, redirect, or reject. That's the core loop. Baxter handles the production. Your job is editorial: setting the direction and deciding what meets your standard.

When you redirect, Baxter learns. It tracks what you correct and adjusts — globally across all content types when the feedback is universal, or locally for a specific content type when it's not. Fifty blog posts in, Baxter knows your voice better than a new hire would.

When something is ready, you publish. WordPress, a headless CMS, markdown export, ebook format. Baxter handles the distribution. You decide what makes the cut.

What makes it different

Most AI writing tools give you one of two things: a blank page with a chat box, or a library of templates someone else designed. The first puts you in the prompt engineering seat whether you know it or not. The second forces your content into someone else's mold.

Baxter does neither. You build the content types. You define what a blog post looks like for your brand, what a case study should include, how a newsletter flows. The formats are yours. The voice is yours. The AI operates inside your system.

The strictness control matters more than it sounds. A weekly newsletter might need tight structure — same sections, same tone, every issue. A thought piece might need room to wander. No major writing tool exposes that dial. They either follow a rigid template or give you a blank page. Baxter lets you set the constraint anywhere between those poles, per content type.

And the feedback loop creates something that templates can't: a system that gets better with use. Every correction teaches Baxter something specific about your preferences. That learning doesn't transfer to another tool. A hundred trained content types, each tuned by your real feedback, is a writing system built entirely around how you work.

Where it's headed

Baxter starts on the web. But writing ideas don't wait for you to sit down at a computer.

A browser companion lets you grab inspiration in context. You're reading an article, a competitor launches something interesting, a data point catches your eye. Select it, tell Baxter what sparked your interest, and it turns the inspiration into a draft in the right content type. No context switching. No rebuilding the prompt from memory.

A mobile companion captures ideas by voice. You're driving and something clicks. "Hey Baxter, this should be a blog post." Baxter asks if you want it atomized into smaller pieces for other channels — a LinkedIn post, a tweet, a Reddit thread from the same idea. You say yes. By the time you park, drafts are waiting.

The next step is visual. If Baxter knows the format, the voice, and the content, it can generate the images that belong alongside the writing. Blog headers and social cards. Ebook covers. The writing and the visuals come from the same understanding of what the piece needs to be.

The goal is the writing companion that shows up where the ideas happen, not just where the writing happens.