Beta

AI Rollout Plan

Sequenced plan for AI features (foundation → guidance → collaboration) with guardrails and dependencies.

See also Claude overview (feature ladders) and Phase 0 checklist for immediate guardrails.

Context-aware prompts

Foundation

AI suggestions scoped to current session/section (e.g., “Summarize this session”).

Owner
Product + Infra
Status
Prototyping
Success signal
3+ workflows successfully inject context without hallucinations

Guardrails

  • User-initiated only
  • All AI text badged
  • Logs stored per user

Dependencies

sessionRoomsaiContextHints

Signal title/summary suggestions

Foundation

Single button in signal composer to propose title + abstract.

Owner
Signals pod
Status
Scoping
Success signal
Writers accept 60%+ of proposed titles

Guardrails

  • No auto-save
  • Show original vs suggestion diff
  • User confirmation required

Pattern extraction panel

Guidance

AI scans selected resources/signals to highlight common threads for sprints.

Owner
Research pod
Status
Backlog
Success signal
Sprint leads cite panel in retros

Guardrails

  • Limited to user-selected nodes
  • Shows sources for each pattern

Dependencies

resourceBucketsuniversalBlocks

Session recap generation

Guidance

After session close, AI drafts summary (decisions, questions, action items).

Owner
Sessions pod
Status
Backlog
Success signal
50% of sessions publish recap block within 24 hours

Guardrails

  • User edits required before publishing
  • Stored as block with AI badge

Dependencies

sessionMemory

Grid cell AI assistants

Collaboration

Specific grid cells (e.g., AI helper in translation) run commands inline.

Owner
Studios pod
Status
Backlog
Success signal
Measurable time saved on translation/content sessions

Guardrails

  • Per-cell opt-in
  • Rate limit per user
  • Audit log for runs

Dependencies

gridComposeraiContextHints

Auto-pilot workflows

Collaboration

Allow AI to run multi-step recipes (e.g., pull signals → draft outline → populate insights) with explicit approvals.

Owner
Platform pod
Status
Backlog
Success signal
At least three recipes reach 80% acceptance without rollback

Guardrails

  • Human approval between each step
  • Scenario-based rate limits
  • Explainability log per recipe

Dependencies

aiContextHintsworkflowEngineauditLog