Skip to main content

AI Skills v2 is here

We’re launching AI Skills v2 to give you a more consistent, representative picture of how your engineers work with AI and where they can grow.

What’s changing

Each day, v2 randomly selects one eligible session per engineer for analysis. For plugin-captured sessions, the Weave plugin must be installed and the engineer must send at least five human prompts. Claude Enterprise Compliance API transcripts are also supported. A weekly fallback selects one session with at least two human prompts for engineers who had no daily pick that week. This approach helps ensure all engineers with eligible sessions are represented, instead of letting a few people with many sessions dominate the picture. We’ve also updated the AI Skills experience with more detailed insight breakdowns and score trends over time.

How v2 analyzes skills differently

V1 estimates skills from summarized and classified conversation turns. V2 assesses the engineer’s actions against specific criteria, considering the full conversation for planning and nearby turns for other skills. V2 is designed to produce more precise, actionable assessments in three ways:
  • Scores relevant opportunities: If there is no multi-agent work, coordination isn’t scored.
  • Focuses on the engineer: V2 assesses the person’s decisions and actions, rather than crediting the agent’s work to them.
  • Shows its reasoning: Ratings distinguish basic from expert execution and include a rationale and conversation location.
Together, these changes make assessments more precise and easier to understand.

What this looks like

  • Planning: Break down the initial objective into concrete, actionable milestones before starting implementation to give the agent a clearer roadmap.
  • Verification: Ask the AI to check its code against the agreed behavior. Following up on an important finding matters, too.
  • Context: Share a relevant constraint or reference to help the AI work effectively. A longer prompt isn’t automatically a better one.
  • Leadership: When an important trade-off arises, set priorities, make a decision, or clearly delegate it to demonstrate ownership.
  • Parallelization: When multiple agents are involved, clear responsibilities and attention to dependencies help keep work coordinated.

A smooth transition

For the two weeks before launch, we ran v1 and v2 in the background for your organization. With the November 2026 cutover, v2 is the default moving forward. If you’d prefer to opt out of v2, reach out to us directly and we can keep your organization on v1. We’re excited to hear your feedback and bring you a more representative view of AI skills across your team.
Last modified on November 10, 2026