Second Read: an AI editor that enforces human-in-the-loop

Most AI writing tools promise a better draft in one click. I wanted something different: a second reader that helps you decide what to change and why.

I build open-source tools that turn domain knowledge into explicit, repeatable guidance for AI assistants. The idea is to help people make better decisions, not just automate their work.

That philosophy came partly from my experience as a BI engineer working with marketing teams. Training sessions and dashboard walkthroughs weren't getting people to use data in their decisions. What worked better was documentation that explained concepts, assumptions, and practical applications, especially when people could explore that knowledge through AI on their own.

It made me realize that much of AI's value comes from making expertise accessible at the moment someone needs to exercise their own judgment. We need to make better decisions, not just automations.

Second Read applies that principle to writing.

Over the years, I'd accumulated extensive notes from my professional experience, academic studies, and personal reading on writing, communication, and editing. I wanted to turn that knowledge into something practical: a reusable editorial skill that could make even frontier AI models more discerning proofreaders, rather than simply asking them to "improve this draft."

I also needed it to understand the context of each project and adapt as my own writing style evolved. I didn't want to repeat my editorial preferences in every session or accept a fixed set of rules that would eventually stop reflecting how I write.

So I built Second Read for people who still edit their own drafts.

It reads a draft as a reader would, points to exact passages, explains proposed edits, flags claims that need verification, and sometimes places a question directly in the text where a reader might get lost or draw a different conclusion.

It distinguishes necessary corrections from stylistic suggestions and aims to make the smallest useful change. Sometimes, the right recommendation is no change at all.

You can accept, reject, or rewrite suggestions yourself. Your decisions and draft versions are preserved, so the review process doesn't erase the thinking behind the final text.

The editorial method is available as an open-source skill, which you can use with AI assistants.

I'm also building Second Read Studio, a hosted workspace that keeps track of your drafts, decisions, and project-specific context. Unlike the standalone skill, the cloud version can propose additional editorial rules based on how you revise your writing, including which automated suggestions you accept or reject.

The goal is an editorial assistant that doesn't just follow a predefined rulebook, but helps you develop one that evolves with your work.

Team workspaces are coming next. The challenge I'm particularly interested in is preserving individual editorial preferences while allowing teams to collaborate, establish shared standards, and keep a clear record of who made each editorial decision.

The free cloud beta is open for applications from content and email writers, technical writers, product marketers, and founders.

There's an interactive demo and beta application at https://secondread.studio.

I'm curious: when you revise something you've written, what's more useful—a proposed edit, or a question that helps you recognize the problem yourself?