$100M ARR Founder: The Next Big AI Opportunity
Greg Isenberg talks with Jon Noronha, co-founder and chief product officer of Gamma, about agent harnesses: how Gamma built one, how it tests models, and where vertical harness startup opportunities remain.
The gist
-
Noronha explains a harness as the tools an agent uses; one Gamma deck can take about 20 models and 200 calls, including visual QA.
-
Since taste is hard to score with evals, Gamma leans on A/B testing; a fast mode that skipped steps lost on satisfaction and paid conversion.
-
Noronha says 98% of Gamma's gains come from prompts, not fine-tuning, and he launched eight tests of Claude Haiku 5.5 the morning after release.
-
He wishes for a Slack harness that learns his voice and drafts replies, and uses Muse and Instinct because they make personal agents easy.
-
Noronha calls image models underhyped and sees room for vertical tools, like a couch visualizer that checks real measurements, especially for professional users.
"Because we have a super diverse user base who are all different, we actually rely a lot more on A/B testing. We will actually run two models or two prompts at very big scale and measure across that wide user base what's actually performing better, because that captures subject..."
▶ Play at 17:522 creators recommend it, including Mauro Botanes.
See all on the A/B Testing page"This is something that's gotten easier than ever. Like, absolutely, Claude is my data scientist. And so I would say now everybody on our team is constantly armed with very fine-grained analysis of failure modes and opportunities."
▶ Play at 33:54250 creators recommend it, including Riley Brown, Chris Koerner, and This Week in Startups.
See all on the Claude page"If I could build a harness for anything, it would be Slack. And I'm a little disappointed that Slack themselves have not built the harness for Slack... I have sent approximately a billion Slack messages. That should be all you need to know how I'm going to reply."
▶ Play at 20:4153 creators recommend it, including Riley Brown, Jason Calacanis, and Guillermo Rauch.
See all on the Slack page"I'm going to wait a few months for them to just make it easy for me. And now I got Muse and Instinct blowing up my phone because they made it easy for me."
▶ Play at 33:2317 creators recommend it, including Claire Vo, Peter Yang, and Riley Brown.
See all on the Muse page"I'm going to wait a few months for them to just make it easy for me. And now I got Muse and Instinct blowing up my phone because they made it easy for me."
▶ Play at 33:2318 creators recommend it, including Harry Stebbings, Kevin Rose, and Shaan Puri.
See all on the Instinct pageWhat they recommend · 5 picks
"Because we have a super diverse user base who are all different, we actually rely a lot more on A/B testing. We will actually run two models or two prompts at very big scale and..."
"This is something that's gotten easier than ever. Like, absolutely, Claude is my data scientist. And so I would say now everybody on our team is constantly armed with very fine-..."
"If I could build a harness for anything, it would be Slack. And I'm a little disappointed that Slack themselves have not built the harness for Slack... I have sent approximately..."
Greg, This Week in Startups, Riley and Chris agree on 40 things
See what they all use, in their own words. Paid ads are left out.
See what they all swear by Or pick your own peopleMade by someone on this episode
Also mentioned
Ads in this episode
Claude. They're paid, so we keep them out of the picks above.
See what Greg recommends without being paid →Related episodes
Episodes covering similar topics
More from Greg Isenberg