Perplexity Brain wrote 80,000+ skills for 20,000+ users
Perplexity says Brain created 80,000+ personal skills for 20,000+ users and made 100,000+ updates, while an eight-week test cut tool-call failures 11% across all activity.
By Nathan Brooks
5 min read

What's News
- Brain wrote 80,000+ personal skills for 20,000+ users from August 1 to November 14, 2026.
- An eight-week test put automatic skill creation at 72.9% of enabled users, versus 6.5% of controls.
- Tool-call failures fell 11%, or about 7% without background runs, with no measurable cost or response-time change.
- Perplexity recorded 100,000+ automatic updates to 30,000+ existing skills from September 4 to November 14, 2026.
- Automatically created skills are capped at 10 per user; built-in changes require paired replay and engineer approval.
By Nathan Brooks
Perplexity said on November 18, 2026, that Brain had written more than 80,000 personal skills for over 20,000 users between August 1 and November 14, 2026. The company said the personalization system for its Computer agent “can now create and update reusable agent skills.”
Perplexity says Brain captures workflow-specific preferences and procedures that a general-purpose model cannot know in advance. The company describes a skill as a package of reusable instructions and supporting resources, sometimes scripts.
Perplexity says Computer’s harness makes skills discoverable and loads their instructions when the agent requests them; users can also specify skills in a prompt. The company says personalization changes the agent’s context and instructions, not the underlying model’s weights.
What does this change for US Max subscribers?
For US Max subscribers, the practical change is a reusable layer for preferences and procedures that can carry across conversations. Perplexity says that design can reduce repeated explanations and corrections.
Perplexity says personal skill improvement is available to all US Max subscribers. It is enabled by default for eligible users, who can turn it off. The Plugins tab identifies Brain-authored skills and lets users choose which remain enabled.
Early test results
Perplexity reported more than 100,000 automatic updates to over 30,000 existing personal skills between September 4 and November 14, 2026. In an eight-week randomized production test, 72.9% of users with automatic improvement enabled received at least one new personal skill. The share of controls who manually added one was 6.5%.
Perplexity reported that a personal skill loaded in a user-started conversation for 53% of the enabled group and 13% of controls. Among customers who received a Brain-written skill, 63% used one again in a later conversation within four weeks.
Perplexity reported an 11% reduction in tool-call failures across all Computer activity in the same experiment. The decline was about 7% after background runs were excluded. Medium- or high-dissatisfaction threads per user fell 7.6%, with no measurable change in cost or response time.
Perplexity also ran a narrower offline experiment using SkillLearnBench. Computer generated one skill for each of nine task families. Across 16 paired instances, task success rose from 43.8% without skills to 62.5% with generated skills.
The full benchmark contains 20 tasks across six real-world categories and 100 verified instances, so Perplexity said its experiment covered only a subset of the suite.
How Brain improves skills
Perplexity says the shared improvement engine has four stages: gather evidence, reflect on it, verify a proposal and activate the change. The company says the harness logs skill loads and files structured diagnostics when it detects user frustration, broken behavior or clear inefficiency.
Perplexity’s stored session trajectories preserve messages, tool calls, reasoning and results. That lets the improvement agent inspect where a procedure went wrong instead of relying on an outcome or short summary.
During daily personal-skill runs, Perplexity’s background improvement agent receives recent sessions, skill-usage data and the existing skill catalog. It has no internet or connector access, although it can retrieve more of the user’s session history as needed.
Perplexity requires two independent occurrences of the same reusable behavior for a new skill, while one grounded occurrence can support an update. Each proposal is limited to one change per skill per run.
Perplexity runs deterministic checks to test whether a proposal is well-formed and applicable. An LLM judge audits sampled proposals for supporting evidence and useful instructions.
Perplexity writes the verified change to the user’s library for future sessions. Because Computer loads an index of available skills at the start of each session, an oversized library can consume context and make relevant skills harder to find.
Perplexity caps automatically created skills at 10 per user. At capacity, the agent must retire an existing skill and justify why its replacement is more useful before adding another.
Stricter controls for built-in skills
Perplexity says built-in skills are available to every Perplexity user, so their updates follow a stricter path than personal skills. The company uses sanitized evidence from sessions eligible for training, stripping credentials, other secrets, attached files and media.
Perplexity’s weekly revision job looks for failures recurring across multiple sessions rather than allowing one bad session to drive an edit. Its scenario-generation agent turns those failures into self-contained tests.
The scenario agent synthetically rebuilds unnecessary context and replaces connector-dependent inputs with synthetic files. Perplexity’s improvement agent receives root-cause descriptions and sanitized successful sessions, but not the held-out failure scenarios used for evaluation.
In Perplexity’s paired replay, every test scenario must pass with the revised skill and at least one must fail with the original. Perplexity uses control scenarios to check that previously working behavior has not regressed.
Perplexity’s LLM judge scores responses and trajectories against each scenario’s rubric. The company submits candidates and their evaluation results to an internal pull request, and a change takes effect only after an engineer approves it.
The team-skill test
Perplexity plans to extend improvement to user-authored skills, skills shared within projects and organization skills. The company said the open product question is how to apply automatic activation, replay testing and human approval as procedures move from an individual’s history into shared team workflows.
For teams, the next test is whether those safeguards remain workable when one skill affects multiple users. The open issue is whether automatic activation, replay testing and human approval can scale from private session history to shared procedures.
Source: unite.ai
More from Nathan Brooks
Show full bio
News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.
2,426 articles