VS Code: Voice Stack Gets a Hard Look, One Rollback Included
A cluster of fixes to voice, dictation, and Agent Host reliability landed today, but the headline is a same-day revert: the new on-device dictation runtime broke minimum Linux distro support and had to be pulled from insiders.
Duration: PT2M37S
Episode overview
This episode is a short developer briefing from VS Code.
It explains recent repository work in plain language.
- Show: VS Code
- Published: 2026-07-21T13:07:09Z
- Audio duration: PT2M37S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good day. It's July 21st, and today's build activity centers on voice, dictation, and the Agent Host — with one notable reversal in the mix.
Start with dictation. Megan Rogge shipped PR 326678, migrating on-device speech recognition to Microsoft's Foundry Local streaming runtime for faster, more accurate dictation. Hours later, Martin Aeschlimann merged PR 326769, reverting it — along with a follow-up model-setting change — because the native addon…
The voice pipeline itself got several independent bug fixes converging on the same theme — state that doesn't get cleared. PR 326405 fixed hands-free narration getting stranded because a mic-latch never released. PR 326696 cleared stale routing and pending-confirmation state on disconnect, so reconnecting no longer…
The second theme is Agent Host hardening, largely from Rob Lourens. PR 326768 prewarms the Agent Host earlier in startup. PR 326767 fixes a startup race specific to Remote-SSH. PR 326711 stops subagent messages from corrupting root conversation state. And PR 326720 adds broad end-to-end coverage with shared replay…
Elsewhere: sandy081 continued restructuring session ownership and layout commands across several PRs, xterm and…
Wh…