TwoTail AI
Analytics for AI Agents
Last year I'd been using tools to do LinkedIn Outbound.
But I got banned!
So, out of fear of losing my account, I decided the agent had to live on my machine. That led me to creating a scheduled job in Cowork.
The agent works from ICPs I define, sources profiles, and connects with them. The way it sources and the connect messages are both built as bandits - so I can try new approaches and it biases towards what works. This means it self-improves!
And that got me to a 50% rate of connect requests being accepted!
Well - I've subsequently found out there are tools that don't get you banned!
I had a lead lined up for a pilot in my sales pipe, and their product is a marketing campaign agent.
I'd not analyzed that type of agent before, so I wanted to check how TwoTail would handle it.
I decided to built a simple replica of their product - spinning up a demo agent with approximately the same features.
Then I integrated that with my product TwoTail AI to see how it would analyze it.
Along the way, I noted down areas where the analysis could be improved.
This entires process took 1 day - now I'm much better prepared to onboard the the real one!
And a side benefit - I can use this agent when running demos for future marketing agents.
-
2
The interesting part wasn't building the replica—it was using it to reduce uncertainty before a real customer ever arrived.
Work that improves both onboarding and future demos usually compounds in ways that aren't obvious when you first build it.
Starting an early access program for TwoTail, open to anyone building an AI agent!
1 Comment
The interesting part isn't the 50% accept rate—it's that the system keeps learning instead of relying on one fixed outreach strategy. Most automation stops at execution. Letting it continuously test and bias toward what actually works is a much stronger long-term advantage.