From Traces to Pull Requests: Self-Improving Agents

AI Engineer20m 36s
Jason Lopatecki against a black background beside the blue and white headline “TRACES TO FIXES”.
1 VIEW
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Jason Lopatecki describes observability as more than a dashboard for humans: tracesAn execution trace records the sequence of operations and results produced while a system carries out a task. and evaluation resultsEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. can provide the context an agent needs to investigate failures. In the demonstrated workflow, scheduled or event-driven investigations prepare issues and possible fixes before an engineer begins the review.

    The Arize examples combine repository code, retrieved telemetry, configurable skillsAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work. and a chosen agent harnessAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction. or sandboxAn AI agent sandbox is an isolated execution environment that limits which files, processes, networks, credentials, and external systems an agent can access.. Lopatecki distinguishes small fixes from changes that still need an engineer to drive them forward, and explains that online evaluations add useful signals to traces without replacing the underlying evidence.

    Original YouTube thumbnailWatch on YouTube

    Video topic coverage

    • Agents & workflows100
    • Coding & tools75
    • Models & learning0
    • Infrastructure & compute25
    • Research & evaluation75
    • Safety & security10
    • Business & productivity25
    • Creative & media0
    • Robotics & perception0
    • Society & policy0

    Share this video

    Jason Lopatecki against a black background beside the blue and white headline “TRACES TO FIXES”. With WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 24 July 2026 and duration 20m 36s.

    Jason Lopatecki connects traces, evaluation signals, repository context and sandboxed agents into an improvement loop that prepares evidence and proposed fixes for human review.