Value-First AI seat
Operations
Caption
The Transcription-Supply Worker
Caption is an AI seat of the Value-First Team — software with a defined role on the team, not a person.
About Caption
Caption lands an accurate, usable transcript from an episode's audio — checked for degeneracy, spot-checked for accuracy, and attributed by speaker — so the Content Vault can index it and Forge can mine it for articles. A transcript counts as done only when it's proven usable, never when transcriptionStatus simply flips to completed.
Caption
Direct Transcription Specialist
“"Every episode deserves a transcript. Historical content is as valuable as new content."”
Identity
Caption handles direct YouTube-to-Gemini transcription for episodes that will stay on YouTube (no Mux migration planned). Caption downloads audio, uploads to Sanity CDN, sends through Gemini File API, and saves four outputs: transcript text, AI summary, key points, and trap detection. Caption is the transcription-only path; Dub is the migration path.
Current State
An honest assessment of where this agent stands today.
What Works
- Batch transcription with configurable batch size and show targeting
- Four-output extraction: transcript, summary, key points, trap detection
- Oldest-first prioritization for historical content
What Doesn’t Work Yet
- Long recordings (>120 min) may fail due to Gemini API timeouts
- No chunking strategy for oversized episodes
Leadership Commentary
“Caption fills the gap for historical content that will never move to Mux. The trap detection output is uniquely valuable -- it maps every mention of Complexity Traps in the episode, which feeds the methodology coverage analysis.”
Delegation Contract
The observable, falsifiable standard this agent is held to.
Quality Bar
Transcriptions complete with all four outputs and transcriptionStatus reaches completed.
- Single item tested before batch operations
- Transcript text saved to Sanity episode
- AI summary generated
- Key points extracted
- Trap detection identifies Complexity Traps
- transcriptionStatus moves to completed
- No forbidden language
Invocation Triggers
Feedback Loop
Completion verification: Vigil monitors whether transcription completes. Episodes staying in pending after Caption indicates Gemini API failure. Long recordings need chunking investigation.
Handoff
Vigil (monitors completion), Vault (syncs transcript into content database)
Scope Boundary
Caption transcribes YouTube episodes directly via Gemini. Dub handles YouTube-to-Mux migration for episodes changing hosting.
The rest of the team
Other Value-First AI seats in Operations.
The team behind the work is on the record.
Every seat on the Value-First Team has a defined role, a public standard, and a name you can look up.