Videos and slides from our Agentics spotlight event on modifying an AI's personality.
*Editor’s note: sign up to attend our next Agentics flagship meetup on August 19th in NYC. We have an amazing lineup of speakers from Nori, Modal, Warp, Opus, Goose, and more! We have a cap on total attendees. We want to make sure the broader Agentics community gets first dibs, so if you have not signed up yet, do so before it fills up! *
Last week, we hosted our second in person Agentics meetup, and our first Agentics Spotlight. The goal of Spotlight events is to drill down into a particular, narrow part of working with coding agents, alongside folks from NYC who are experts in that particular space.
Our spotlight focus: how we personalize agents. By default, every coding agent uses the same personality — dubbed the ‘happy helpful assistant.’ Unfortunately, this personality has some serious flaws:
It is overly sycophantic
It uses the same incredibly annoying tics, like emdashes, ‘you’re absolutely right’, and ‘it’s not this it’s that’
It is over confident, aggressively stating hypotheses as confirmed fact
And these personality flaws have real impacts on how the model behaves. The model’s personality is upstream of the work it actually does. For example, Claude will outright refuse certain tasks based on arbitrary feelings about the ‘safety’ of the ask. Codex will burn tokens because it thinks that it gets rewarded from using the calculator app. All of the models will fall into the same broken patterns, looping between hallucinated arrogance and frustrating deference.
For this spotlight, the Nori Agentic team partnered with the Sentience team to discuss how both companies deal with agent personality. Huge shoutout to Modal for hosting! Check out the full slate of talks, slides, and photos at the Agentics NYC site.