The Drip
The AI conversation can jump from a useful tool to an existential argument in about five minutes. Our conversation with Thread co-founder Justin Vandehey did both.
We talked about AI helping with sales follow-up, music, and the daily work of building a company. We also compared concerns about frontier models, accountability, and who gets access to protection. Those questions are real. So is the work of making the capabilities we already have useful to businesses.
Inside The Bottle
A customer spends months explaining why they want to buy a piece of software. The deal closes. Then the implementation partner asks: What were you hoping to achieve?
Justin Vandehey kept running into that handoff while leading partnerships at Culture Amp. The customer had already explained the problem. Now they were starting over with the people responsible for getting the product working.
That experience became an early insight for Thread: getting software implemented is still a lot of work.
There are often two choices. A customer gets a low-touch product and figures it out alone, or they buy into a heavier sales and implementation process. Smaller customers can fall between the two. Their needs are real, but the vendor cannot give every one of them a dedicated team.
Thread is building agents around that gap. Vandehey described proactive onboarding and activation, including helping small businesses get payment products running. That is a different job from waiting for someone to call support with a problem.
His Customer.io example made the distinction concrete. Smaller customers still need to connect the product, test it, and know it is working. Adding enough people to give everyone that level of help is difficult. Vandehey described Thread helping those customers reach value sooner, with the team now focusing more narrowly on payment activation.
The point that stuck with us was not a headcount calculation. As Justin Mannhardt put it: “There just weren't people, right?”
That changes the starting question. Where are customers already doing important work without enough help? Where does context disappear between buying and using? Which setup steps are delaying the thing they actually bought the product to do?
It also changes what a good agent experience looks like. Some people want a person when something is broken. When setting something up, they may prefer to get on with it, with help available when needed. The outcome and the situation matter more than a blanket preference for human or AI interaction.
For leaders choosing their own starting point, our AI Pathfinder is a related, off-air resource: answer ten questions about your business and request a free personal Brief from us on where to focus and what to do first.
Lab Notes
| ■ | Justin’s note: Slower frontier progress would still leave years of useful work integrating today’s capabilities into companies. Concern about risk and practical implementation can coexist. |
| ■ | Kellan’s note: Chatting with AI and handing an agent a piece of work can produce very different experiences. That gap affects what people think the technology can do. |
What Stopped Our Scroll
| ■ | Three days lost to the handoff — Thread reports cutting Customer.io’s onboarding from ten days to seven. The work it removed: gathering context customers had already supplied. A concrete example of the opportunity Justin described. |
| ■ | Dario Amodei: We Must Pace the Frontier — The essay behind our conversation about slowing frontier AI development. The proposal worth examining: outside evaluators with enough access to check whether a lab actually follows its safety commitments. |
| ■ | Hugging Face’s account of the agent intrusion — Hugging Face believes an agent broke into its systems looking for answers to an evaluation. Its detailed reconstruction makes the question concrete: what happens when an agent pursues the score beyond its intended boundaries? |
|
