# VS Code Live: Agent Sessions Day Source: https://www.youtube.com/watch?v=Osvu53c0yYg Recap page: https://rapidrecap.app/video/Osvu53c0yYg Generated: 2026-02-19T20:34:19.054+00:00 --- ## Quick Overview VS Code Agent Sessions Day showcases significant advancements in multi-agent development within VS Code, allowing developers like PMs and engineers to delegate complex tasks across local, background, and cloud agents for increased velocity, better decision-making through parallel task execution, and automated workflows like issue triaging and code review. **Key Points:** - Copilot Pro Plus and Enterprise users can now run Claude, Codex, and GitHub Copilot as coding agents on GitHub, VS Code, and GitHub mobile, offering 'three frontier agents, one subscription.' - The Agent Sessions view acts as a 'single pane of glass' for managing local, background, and cloud agents simultaneously, enabling developers to compare competing approaches to a task. - The VS Code team uses AI extensively for adjacent tasks, such as Harold using a Dogfood agent with Playwright to critically review generated slides from a learner's perspective, documenting rough edges. - Custom agents and skills integrate tools like Work IQ (for meeting summaries) and GitHub actions/Chrome extensions for automating issue grooming, including duplicate detection and owner assignment with confidence scores. - The team emphasizes a new operating model: 'Stay hands on locally for visual and exploratory work. Delegate clear longer running tasks to background and cloud,' addressing developer anxiety about waiting for results. - PMs are shifting from writing long specs to creating prototypes or PRs, exemplified by a PM sending a PR for a feature idea which was iterated on with engineering feedback and merged, drastically changing the feedback loop. - Automated processes include Copilot Code Review Agent (CCR) automatically reviewing new PRs in a dedicated Slack channel, requiring engineers to resolve its comments before final review proceeds. **Context:** The video captures 'VS Code Live: Agent Sessions Day,' an event focused on demonstrating the evolution of multi-agent development capabilities integrated directly into Visual Studio Code and the broader GitHub ecosystem. Key participants include hosts Burke and Olivia, and speakers like Kyle Daigle (CEO of GitHub) and Harold Kershner, who showcase how developers, product managers (PMs), and engineers are using these new agent features for daily tasks, ranging from feature development to administrative overhead reduction. ## Detailed Analysis The Agent Sessions Day highlights the maturity of multi-agent development, positioning VS Code as the 'home for multi-agent developments' covering agent orchestration and model selection. GitHub CEO Kyle Daigle announced that Copilot Pro Plus and Enterprise users gain access to Claude, Codex, and GitHub Copilot as coding agents across VS Code, mobile, and CLI, providing developer choice without separate accounts. Harold Kershner demonstrated using multiple agents concurrently—local for visual work (like animating a workshop factory CLI output with ASCII art), background for defined tasks (like fixing terminal rendering breaks), and cloud for exploratory work (like migrating code to Open UI via a PR). Guardrails are critical, including hooks on tool calls, terminal sandboxing, and CI checks on cloud PRs. The VS Code team, represented by Pearce and Peng, detailed how they use agents for self-hosting and daily operations: Peng uses custom agents to summarize daily priorities from Obsidian/Work IQ, assign issues, and summarize the last 24 hours of commits in the VS Code repo. PM workflows are also transforming, moving from writing lengthy documents to creating executable prototypes (PRs) based on user feedback, which agents can then validate using tools like Playwright and the 'demonstrate' custom agent. Furthermore, agents automate administrative tasks; bots use Copilot CLI and Agent SDK to triage hundreds of daily GitHub issues, suggesting duplicates and owners with confidence scores, preventing the team from drowning in triage work, although accountability for code quality ultimately remains with human engineers. ### Agent Platform Expansion - Copilot Pro Plus/Enterprise users run Claude, Codex, and Copilot as agents - Agent Sessions view centralizes local, background, and cloud agents - Focus is on developer choice and ecosystem building. ### Live Demonstrations and Workflows - Harold demonstrates local agent use for visual polish (ASCII art banner) and remote agents for large refactors (Inc to Open UI migration) - Agents are used for generating structured workshops quickly via CLI. ### Quality Gates and Control - Developers use deterministic controls like hooks, terminal sandboxing, and CI on cloud PRs to ensure quality - The Dogfood agent acts as an opinionated learner persona, using Playwright to drive a browser and critique generated content like slide decks. ### Internal Team Velocity and Automation - VS Code team engineers use custom agents daily to summarize commits and triage issues - Issue grooming bots automatically check for duplicates and suggest owners using past assignment data and ownership instructions. ### The Maker vs. Manager Schedule Shift - AI empowers parallel work, allowing managers to fire off agents between meetings instead of sequential note-taking - PMs create prototypes/PRs instead of long specs, sending direct code contributions for faster feedback cycles. ### Automated Code Review Process - A Slack bot picks up ready PRs, automatically assigns Copilot Code Review (CCR) to check changes - Engineers must resolve CCR comments before the PR moves forward for final human review, ensuring quality checks are not skipped.