Personal news: I joined Every

I joined Every earlier this year as GM of Spiral, their co-writer product. Here's why and a little about my experience so far.

About a year ago I wrote a post called "Every (the company) knows what's up". To summarize:

• Every started as a media company, built a direct following, and then expanded to a product studio and AI transformation consultancy.
• Every came to many of the same conclusions I have here at Subtle Software and put them into practice.
• I think they're ahead in terms of using agents in production effectively and thinking about products in an "agent-native" way.
• LLMs are changing software so much, and I knew I'd have to join a startup that was embracing this new mode of building software. Every fit the bill.

During my year of building software as a team of one, I found that my main limitation was distribution. I don't have a big public presence, let alone a media operation with a large audience. This meant that no matter how good my products were, there was still a formidable distribution challenge. Substats earned a handful of business customers and has continued to grow, but getting the word out remains a challenge.

Every benefited from being a premium media company for years before launching its own software products. They don't have the distribution challenge that I – and many other new software startups – faced. This makes me bullish on the company overall, but it's also attractive as a software creator to tap into a built-in distribution channel for any new product I cook up.

Another reason I joined Every is that I could only make so much progress navigating this new world of software development on my own. I texted with other product/engineering people during my year building Subtle Software, but actually working alongside colleagues is the only real form of collaboration. As a practical example, I recently was exploring how to make it easy for a user to install my product (Spiral) in an agent. I pointed Claude Code at my colleague's repo (for Cora, Every's email assistant) and asked it to review his implementation. That saved me a bunch of time, and I can trust I'm using cutting edge best practices (my colleague created the popular Compound Engineering plugin).

Spiral itself is an interesting product. To date, it's been positioned as an AI writing partner with taste. The reason you'd use Spiral instead of Claude or ChatGPT is that those chatbots typically rush to a first draft. The output is generic-sounding, and they don't take the time to do what a professional ghostwriter does with their clients, which is to interview the writer and draw out further material and message clarity before writing. Spiral is designed to interview the user to reach material sufficiency and message clarity, before writing a draft that mimics the user's writing style (gleaned from writing samples).

I've previously worked on CMSes at media companies and even an AI writing product, Axios HQ. Spiral is appealing for a few reasons.

1. I genuinely think LLMs can draw out the interesting stories and information that lives in the heads of people who aren't confident writers. As an example, I've been emailing with a Spiral user who's a retired musician. He'd dreamed of writing a memoir, but didn't think he was a strong enough writer to put pen to paper. He found Spiral and he's been using it to build his memoir chapter by chapter. He's not only put together dozens of chapters of personal reflection but also told me the experience of using Spiral has taught him how to be a better storyteller. I think this is AI at its best: allowing us to express our creativity and unique personal experiences better by augmenting the skills we lack.
2. Spiral has only been accessible via the web, but clearly there's an opportunity to integrate it with agents. Most agent writing today uses the same writing style, the kind of generic politeness you'll see in ChatGPT or Claude responses. (I wrote more about this for Every here.) If Spiral learns your writing style, you'll want your agent to use Spiral for any writing request. That's a big opportunity and very sticky if we really nail the writing style. In general, I think much of software usage will become intermediated by agents, so I want to explore and help drive a part of this shift myself.
3. The notion of using AI for writing is fraught. The word "slop" is ubiquitous. The novel Shy Girl was yanked from U.S. publication recently because passages were revealed to have been likely generated by ChatGPT. To me, slop is a two-dimensional axis of reader expectations vs. writer effort (more on this in an upcoming post). As mentioned above, I do think LLMs can be harnessed to give the world more good writing, but we need to tread carefully. With humility, I think I have the right combination of reverence for great writing and enthusiasm for new technology to approach this opportunity in the right way.

I also love how Every has embraced the "two-slice" team, as they call it. I've written a bunch about how LLMs allow a single person to accomplish what previously took several or dozens of people. Every's organizational structure reflects this. Each one of its five software products is run by one person. Aside from saving money on headcount, this operating model allows people to move quickly, avoid coordination drag, and be more creative. Some of the best ideas die in committees.

I'm a couple months in. I'll be writing for Every every few weeks, but I'll continue to post here about random observations as I continue to navigate this brave new world of AI software development.