Deterministic document comparison at scale — now within Claude, ChatGPT, and via REST API
Joined October 2009
- Tweets753
- Following1.2K
- Followers449
- Likes9.5K
Last night I built an app to cover regression tests from my keyboard, superhuman style.
I saved for the end, “Claude, add a button that locks the baseline, confetti shower. Don’t press the button. I want to press the button.”
Claude pressed the button.
Thanks @mitchduncombe - and I have the utmost respect for Vector. It’s clear you’re merging the high standards of elite law practice with the speed and efficiency ai-native enables.
@VersionStory is an awesome redlining product - highly recommended. As a firm, we care about quality in every detail. Version Story ups our game.
Vector Legal (YC W26) embedded our comparisons in their OS for ai-native law practice. Founder @mitchduncombe on why:
“Beautiful redlines matter to a law firm as a business. When we present to a company, we want polished deliverables, and Version Story’s PDF comparisons embed right in our client portal.”
Kevin O'Connell retweeted
@VersionStory - it's seamless and been great for me and my team!
In 2021 we joined YC to build git for lawyers. Five years and tens of thousands of edge cases later, our tech is mature and AI-native firms like Blackletter are asking us how we handle rebase and merge.
Kevin O'Connell retweeted
"The question of whether a computer can think is no more interesting than the question of whether a submarine can swim."
-Edsger Dijkstra
“Feel,” even “want,” I can get behind as things for which evidence is thin. But AIs *clearly* think and understand. If your definition of “think” excludes entities that can autonomously resolve Millennium Problems, it is your definition that is flawed.
This desire to avoid anthropomorphic language has become a pathology.
Small changes can improve quality of life for lawyers – like making it easy to set a file naming convention for your redlines.
We’ve rolled out a developer portal where you can create keys for our deterministic comparison REST APIs, manage your retention policy, and monitor usage. It’s been fun working with AI native firms on our developer experience.
To reduce token spend, we return a redline serialized to Markdown, alongside Word and PDF redlines. This way, agents and lawyers can review in parallel. Our evals found substantial improvements in both token usage and speed.
Kevin O'Connell retweeted
Over the weekend, our team built a novel agentic document processing system using Jev. The initial results are very promising. Our system completed the same editing task 97% faster than ChatGPT's Word plugin.
Redlines have been blue and red for decades, unreadable for the one in twenty people who are colorblind. At Version Story, accessibility has become core to our operating model. When a colorblind lawyer needed to review a redline, we partnered with one of our customers to ship support for accessible color modes that same day. We are committed to pulling legal software forward for everyone.
Just say "compare these docs". Claude identifies the original and modified versions, then runs them through our best-in-class deterministic redline engine via MCP.
One lawyer told me it's a five-hundred percent improvement in his redlining experience.
Ping me if you'd like to try it out.
Follow this guide to try it out: versionstory.com/developers/…
PSA: If there’s anything you want to get done, tonight would be a good time to do it. Anthropic and OpenAI are in a model war right now, and the rate limits are OFF. AI might not be this good again for months.
Kevin O'Connell retweeted
This is the UI (or lack thereof) that AI-native law firms should want for specialized legal tech tools:
I wonder if trees feel about paper data storage the way turkeys feel about Thanksgiving
Remembering the night I got fixated on paper data storage and learned you can fit about 140 pages of (encoded) plaintext on a single sheet of paper. The page below includes the encoder legend and its source code repeated 26 times
It's illegible at a glance and breaks when photographed or shared on social media, but is totally reconstructable without guidance if printed, scanned, and handed to an LLM for decoding
Do with that what you will!
