I build networks for a living.
I came to the US in 2010 with a question: how will technology reshape the way societies are governed? I arrived on Stanford campus just when platforms like Facebook were becoming societies of their own, run by code instead of law.
The question kept finding me. Peter Thiel - Facebook's first investor - was one of my professors at Stanford. I lived next door to Mark Zuckerberg after graduating. I became friends with Ilya Zhitomirskiy, one of the Diaspora founders, when I moved from Palo Alto to SF to live with my cofounder Tony at a place we lovingly called The Hive.
I did not plan any of this. But I have spent a strange amount of my life adjacent to the people who decided what the social internet became, while sharing a kitchen with the person who tried to build the alternative.
These questions at the intersection of law and technology aren't new to me; more than fifteen years in, they're the work. As CEO of Legal.io, I connect legal professionals to each other, and to the world's best companies. What's new is the leverage - AI changed what one person can build, enough to run the real experiment in a weekend. So I did.
Just a task list
Months ago I built a small app to automatically label my email ("need response", "FYI") and draft replies using a built-in chat assistant. On July 4th the quiet of the holiday weekend allowed me to pick it up to try Anthropic's newly released Fable model.
Cursor IDE booted and Fable enabled, the experiment started:
Investigate this app and add a task list feature to help me track email follow ups organized by functional area.
After a fabulous 10 minute Fable run - my first - it ended up looking like this:
.png)
"Nice!", I thought, "it even added filters and a share feature". It was clear that my task list prompt was just scratching on the model's potential.
So I fired off the next one:
build an MCP so you can interact with lists
This addition would allow the app's chat assistant to view and update my task list. Mere minutes later, I could describe a task and watch it appear on the appropriate task list.
An interesting loop
As I played around with the newly delivered updates, I found myself writing down minor annoyances, improvements and feature ideas to a "Roadmap" list. Things like:
- Add a tab of the contacts I regularly email
- Look at calendar to write better drafts?
- Access LinkedIn contacts/messages (probably not; limited API's)
By then I had set up the credentials to access my task list in our Slack instance, which allowed me to use external agents (ChatGPT, Claude) to interact with my task list. As soon as I started testing the feature, this convo happened:
It dawned on me: the Cursor agent in Slack can also access the code of the app itself (not just the tasks in the database) - it could read my notes, ask me what to build next, and do it. In other words the app was able to build and maintain its own roadmap.
The fact that it was a quiet long weekend wasn't the only treat; the timing helped: Anthropic had made its new Fable model free to subscribers that weekend - a promo that kept getting extended.
So I kept adding items to the roadmap, unbothered by any cost constraints. "Allow me to create recurring tasks", "add avatars to contacts.", "allow me to send drafts from the app", "show a tab with my important messages". Suddenly, it was midnight.
I added some more ideas to the roadmap. "Figure out a way to make LinkedIn work"; "Allow me to import my facebook contacts (birthdays?)", "Can we support messaging?"
Then I gave the agent its final prompt for the night, and went to bed:
Build the next roadmap item. Add bugs, improvements and features to the roadmap list as you go. Once done, build the next item. Also, build a fully featured social network, lol - You have an unlimited budget 🤣
Fourteen days later
Two weeks on, the agent hasn't stopped: hundreds of completed roadmap tasks, 1,200+ commits, over 100 pull requests, $20,000 worth of free tokens consumed.
What came out the other end would have taken a team and serious money to build just a couple of years ago. I spend a lot of my time with these models. This sure felt different.
That roadmap task list had become Jane: a personal assistant that sits on top of my network, communications, and tasks. Here's some of what it built:
A really friendly assistant
Jane evolved from a simple email classifier into a full assistant - aware of my schedule, on top of my tasks, and in touch with my network.
One place for all my contacts
Fable built features to import and sync existing contacts from LinkedIn, Facebook and other sources, providing me with a single view of my network for the first time.
It also created various ways to view my network, including an org chart for Legal.io, a family tree, and a cool 3d view that shows mutual connections.
A unified inbox
By July 6th, Fable had discovered the Unipile API, and used it to expand coverage from just email to LinkedIn, WhatsApp and other messaging apps.
Keeping track of messages across different channels is a challenge I had never found a good solution for, so this feature in particular was rather exciting to me.
Jobs
But it didn't stop there! By July 14 Jane had decided that a job board would be an important feature, and Fable pulled on some historical work to get it going.
News
And wouldn't it be nice if Jane could give you updates about important things that happen to the contacts and companies in your network? So it built a news aggregator!
Posts
How about user posts? NO PROBLEM! It added "Reflections", a feature that allows users to "share updates with their circles".
It only took a small nudge to integrate the feature with Diaspora - and I was excited to see that the project continues to be actively maintained.
All good things
These are only some examples of what a $20,000 Fable budget buys you. And like all good things, the free ride had to end - Fable was free for a while, and then it wasn't.
But that's the experiment starting, not ending. This isn't a new startup, and that was never the point - the point was whether it could build itself, and it could.
The open question now is whether it can keep running now that the tokens aren't free. Jane keeps her own feedback list, her own roadmap, and writes her own updates - and today she got her own budget, with a mandate to make herself self-sustaining. Whether she can is the next thing to find out.
If an app can build and run itself, what's left for the intermediaries in the middle? That's the part that feels new.
Ilya believed a network could belong to the people in it. After the past two weeks I believe he was right, just too early.