6–9 months · live (authenticated)
Some jobs need one model call. Others need a brief, an image, a video, and an evaluation step. Authenticated execute already expands a bounded DAG, waits on catalog jobs, and hands off image-to-video inputs. The planner is not a general-purpose agent. The sequence in the demo is illustrative.
Bounded plan · no jobs submitted
Create a 15-second product ad
Product analysis
vision brief
Hero still
image generation
15s 9:16 clip
image-to-video
Caption pass
copy step
Quality check
evaluation layer
Authenticated execute runs a bounded DAG with wait and handoff. This page does not start jobs.
Each step has a tool, expected input, expected output, and a stop condition. Open-ended tool use is out of scope.
The agent should see one locked price for the completed job, not a surprise charge after step three. Outcome quotes are the paired 6–9 month surface.
Retries and ambiguous submissions reconcile against that identity so the same plan cannot be charged twice. Authenticated execute is the charge path; this page does not start jobs.
QA is a step in the plan, not an afterthought. Failed checks are stored on the receipt the requesting agent can retrieve.
Customer asks for a defined creative job. Studio chooses the sequence.
Every step is inspectable. The planner is not an unbounded agent.
Execute runs the DAG. This page does not start jobs.
Stripe Checkout, wallet env, and seller payouts stay off this page. Authenticated outcomes execute is the live charge path. Named-model MCP and /ai-router remain available.
Do not call FAL or other provider credentials. Candidate scores and outputs on these pages are illustrative.
Do not grant credits, settle USDC, or pay seller revenue share from this surface.
Do not wire Stripe Checkout, card forms, or human top-up URLs on these plan pages. Settlement stays a later integration.
Yes, through authenticated POST /api/v1/outcomes/execute or MCP execute_outcome. Catalog steps reserve Studio credits. This page does not submit those jobs.
No. The strategy is to be the specialized creative execution layer that other agents trust — not a general-purpose agent that browses the web or spends without a task boundary.
If the sequence cannot be audited, an agent cannot trust the charge.