Perplexity Brain automatically updates agent skills from experience
Perplexity Brain turns recurring patterns in session history into personal agent skills that it can create and update; the feature is available to Max subscribers in the US.

AT A GLANCE
Key points
- Brain creates and updates reusable personal agent skills from session history.
- Personal skills change instructions and scripts loaded into context, not model weights.
- A new skill requires two independent occurrences, and the automatic cap is 10 per user.
- The feature is available to US Max subscribers and can be controlled from Plugins.
What is changing?
Perplexity announced on November 18, 2026 that its Brain system can automatically create and update personal agent skills from usage experience. The feature is available to Max subscribers in the United States, is enabled by default, and can be turned off. Skills authored by Brain are labeled in the Plugins tab, where users can enable, disable, or remove them.
Skills are reusable instructions and scripts loaded into context when needed; they do not change model weights. The system follows a loop of evidence, reflection, verification, and activation. A personal improvement agent reviews recent sessions daily. Creating a new skill requires the same need to appear independently twice, while updating an existing skill requires one session-grounded example.
Proposals are not activated directly: the pipeline uses deterministic checks and a sampled language-model judge, and permits only one change per skill per run. Automatically created personal skills are capped at 10 per user. Built-in skills use eligible sanitized aggregate evidence, paired replay tests, and approval by a human engineer before merging. That human approval applies to the built-in-skill pipeline, not to daily personal-skill updates.
Perplexity says it created more than 80,000 personal skills for over 20,000 users from August 1 to November 14, and made more than 100,000 automatic updates to over 30,000 existing personal skills from September 4 to November 14. In the provider's eight-week randomized test, 72.9% of enabled users received a new skill versus 6.5% of controls; tool-call failures fell 11% overall, dissatisfaction fell 7.6%, and no measurable change in cost or response time was detected. These are Perplexity's own experimental results, not independent validation.
This report is prepared with AI assistance and checked against sources. Automated checks do not constitute human editorial review. Editorial approach
Explore the original sources
Source date: 18 November 2026.
Published by AI Radar: 19 Nov 2026, 07:09 (Istanbul, UTC+3). Updated: 19 Nov 2026, 08:01.
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.






Loading comments…