How AI agents use harnesses to act: Gamma's Jon Noronha on design and startup edges
Greg Isenberg · · 55-second read
November 22, 2026 · AI Tech
In this video, Greg Isenberg interviews Jon Noronha, co-founder and chief product officer of Gamma, about how use harnesses—the tools and interfaces that let agents act, inspect their work, and ask users for guidance. Noronha explains that Gamma spent 2024 rebuilding its presentation tool as a creative agent powered by Claude, shifting from preset workflows to agentic decision-making.
Noronha walks through the mechanics of a single Gamma request: when a user provides messy notes and asks for 20 slides, the system spins up a cloud sandbox, pulls context from Notion or Slack, generates an outline, selects a template using vision models, and produces text, layouts, images, and charts in parallel. One finished deck uses roughly 20 different models and 200 calls. The harness—the wrapper around the agent—is where most of the competitive work happens, and even OpenAI and Anthropic invest heavily in harness design.
Noronha discusses how creative agents differ from code agents: taste and human feedback matter more than evaluation metrics, and testing new models can happen within an hour. He shares examples of harnesses he wants to see, including a Slack agent that learns a user's voice and a shopping tool that checks furniture fit in a room.
This is an AI-written summary. The original reporting belongs to Greg Isenberg. How we write these · Spot a mistake?
Get the AI brief every morning.
The five AI stories that matter each morning: startups, money, jobs, tech and what's coming next. Free, and one click to unsubscribe. See our Privacy Policy.