Opportunities.md
Where a new product can win
The opportunities below are hypotheses to validate with customers, not claims of an empty market. Existing suppliers already cover parts of each workflow.
1. A production system for one recurring content job
Pick a narrow buyer and deliverable: for example, a small commerce team producing a weekly set of product launch videos. Accept the product assets and campaign brief, produce a controlled set of variants, collect approval, and export the formats the team uses.
The value comes from reducing the number of decisions the user must remake. Preserve the product, brand rules and approved references. Make the uncertain parts explicit. Let reviewers correct a shot without losing the rest of the project. Store rejected versions and reasons so an agent does not repeat the same mistake.
The advantage to test is a shorter path from brief to approved delivery. The risk is becoming an interchangeable prompt box. A paid pilot should measure customer editing minutes and repeat use, not just the number of exports.
2. An update-and-localize workflow
Treat a video library like maintained content. When a source document, product feature or policy changes, identify the affected segments, prepare revisions, request approval and regenerate the appropriate language variants.
The difficult product problem is traceability: which message, source version, language and approval produced this output? A useful system makes those relationships visible and can export them. It should not silently rewrite an already approved statement just because a model prefers a different phrasing.
This approach may build retention around an ongoing library. It requires learning a customer's real review process and integrating with where videos are delivered. Avatar vendors already compete here, so a new entrant needs a specific underserved workflow, not merely a claim of better translation.
3. Collaboration and asset memory for agent-made media
Agents can generate many drafts. Someone still needs to know which brief they belong to, which version was approved, who can see it, and which sources or references were used. A shared file layer can preserve originals, structured metadata, descriptions and access controls across agent runs.
This is directly relevant to Stuff. The initial useful role is to be the durable project record: briefs, references, storyboards, renders, review notes and final deliverables in a space. An agent should be able to read and update that record through the same user permissions as the UI. Previewing the assets lowers coordination cost.
The opportunity is an inference from the workflow, not evidence that the market wants another general-purpose drive. Validate it by watching an actual team retrieve and reuse past work. File counts and storage growth alone are weak indicators of product value.
4. Evaluation and routing for a specific task
A routing layer can choose models by acceptance rate, latency and cost on a customer's own workload. It becomes more useful when it records consistent evaluations and catches regressions after a supplier update.
Start with a limited task taxonomy rather than a universal “best model” switch. Product close-ups, talking presenters, location shots and stylized transitions need different checks. Include a human fallback for ambiguous results. Avoid pretending that an aesthetic score proves factual or brand correctness.
The risk is that routing becomes a feature of the generation platform. A standalone product needs workflow-specific evaluation data or a broader operational role that survives that bundling.
What to validate first
Recruit three to five teams with the same recurring job. Have them provide recent briefs and accepted deliverables, then run a paid assisted pilot. Track approval time, supplier spend, the work left outside the product, and whether they return for a second project. Ask what would make the tool impossible to remove from next month's workflow.
Do not build a marketplace, social feed and model platform simultaneously before this evidence exists. For a file product, useful spaces and easy retrieval are a clearer first test than a noisy stream of every generated artifact.
Open with JavaScript for the full viewer.