Wrote this up for OpenAgents substack. This was after the 3 hour livestream we did for the afterparty at TEMS. Also I beat everyone in poker that night and no we were not playing for money. Enjoy.

After wrapping up a three-hour livestream, the OpenAgents team gathered around the table at Austin’s AI Frontier lab to decompress from a packed few days at TEMS.

What started as a casual afterparty conversation quickly turned into something much more interesting.

The discussion moved beyond the usual AI talking points and into territory that most people are not paying attention to yet. Not because it is obscure, but because it sits beneath the layer where most AI conversations happen.

Most conversations I hear around Austin around AI still happen at the surface.

People talk about apps, agents, model rankings, benchmarks, product launches, and whichever company happens to be winning the news cycle that week.

Those conversations are not wrong. They are just increasingly incomplete.

What emerged from the OpenAgents AfterParty at TEMS was a different picture entirely.

These builders, operators, and researchers in the room kept returning to the same conclusion…as AI becomes more useful, the bottlenecks move downward.

Read the rest on OpenAgents Substack.