everyone's building the same personal agent
Dan Shipper’s team at Every.to just launched Plus One. I spent yesterday reading through it and I keep coming back to one thing: this is the same thing multiple teams are building right now.
Hosted runtime per person. Lives in Slack or iMessage as a DM. Connects to your tools — calendar, email, notes, GitHub, whatever. Has a memory layer. Has a skill system. Can surface information before you ask.
This is the pattern. And it’s converging fast.
It’s not copying. When you start from “give every person an AI agent that knows their context,” the design space collapses pretty quickly. You end up with:
- A persistent runtime (the agent lives somewhere, not just in a chat window)
- A messaging interface (Slack DM, iMessage, something you already use)
- Tool integrations (it can actually do things, not just answer questions)
- Memory (it remembers you across sessions)
- Composable skills (you can extend what it knows how to do)
There aren’t that many ways to wire those five things up. So teams starting from first principles keep landing in roughly the same place. That’s convergent evolution.
The differences between implementations are real but narrower than you’d think.
Every.to’s version connects to their own app ecosystem — Quora for email, Monologue for dictation, Sparkle for file organization. Consumer-focused. You get a full virtual machine per user (expensive, but powerful).
The enterprise version of this pattern is a different execution: connecting to Salesforce, Confluence, internal wikis, CRMs. Boring, hard plumbing. Worth it if you’re selling to companies.
The local-first version is a third flavor: everything runs on your machine, nothing goes to a server, no subscription.
But structurally — persistent runtime, messaging interface, memory, composable tools — it’s the same design.
Here’s the red bar: if the pattern is this obvious, the moat isn’t the architecture. Everyone will ship roughly this design within 12 months. The moat has to be somewhere else.
For Every.to it’s probably the content audience. Dan has a newsletter with hundreds of thousands of readers. He can launch Plus One to people who already trust him. Distribution is the moat, not the product.
For enterprise plays it’s probably the integrations. Connecting to Salesforce and Confluence is unglamorous, slow work that nobody wants to redo. Once you’ve done the plumbing, you have a durable advantage.
For local-first products the bet is privacy and cost. No server, no subscription, no data leaving your device.
Personal agents aren’t a niche experiment anymore. Multiple well-funded teams converged on the same architecture in about three months. That’s a signal. When an idea goes from “interesting” to “everyone’s building it” that fast, the category is real.
The question isn’t whether everyone gets a personal agent. They will. The question is which version wins — cloud with a data flywheel, enterprise with deep integrations, or local-first with privacy as the feature.
When everyone builds the same thing, the question stops being “can you build it?” and starts being “why should anyone pick yours?”
The moat is distribution, or integrations, or trust. Not the design.