A fast decision before the agent moves
Look at a screen, offer one action, and wait for a click before a slower agent takes over.
- Category
- Developer tools
- Who it is for
- A developer adding one suggested action inside an app they already ship.
- Possible model
- API
- Difficulty
- Medium
- Inspired by
- Decisions API · OpenAI
- Original launch
- Watch the launch · Oct 6, 2026
Difficulty is an editorial estimate of technical and operational complexity, not a score.
Startup idea analysis by Startup Ideas. Last reviewed Oct 7, 2026. How we read a launch.
The startup idea
River Mist, who lists Emerging Products at OpenAI, posted that the Decisions API is in public beta. The demo in the post watches the screen, and on a pause it suggests an action from what is on screen. A click hands control to an agent loop. The post says that task finished about 10 times faster than a traditional computer-use agent. That 10x is their claim about this demo. It is not a benchmark this page reran. OpenAI's own docs describe the API differently: it answers questions about text or images, as a probability, a choice, or a score, and they say it is about 10 times faster than the Responses API. Same brand, two different comparisons. Do not mix them.
The opportunity is the pause. A computer-use agent that always drives is slow and easy to distrust. A product that looks, offers one action, and does nothing until a person clicks is a smaller company, and it can sit on top of someone else's decision API. The buyer is an app that already knows the next three actions a user might want, or a team that wants an agent to wait.
Do not rebuild OpenAI's endpoint. Do not put a watcher on a screen full of someone else's mail until they have said which apps are in bounds.
The problem
A computer-use agent that always drives the mouse is slow, and it acts before anyone asked it to.
Who would pay for this?
- Primary
- A developer adding one suggested action inside an app they already ship.
- Secondary
- A team that wants an agent on the desktop but will not let it click first.
- Early adopter
- One app, one pause, one suggestion.
How could it make money?
Possible models. Nothing here is a claim about what OpenAI charges.
- Charge for the shortcut inside a product, or per decision if you are the one calling the API.
- OpenAI's docs list a price for the model behind the API. That is their price. Do not copy it as yours, and do not assume the demo in the post is free.
- The launch post does not sell a separate app. The product in the film is a demo of the API.
What would the MVP look like?
- One app, or one site.
- A pause.
- One suggested action, written in plain language.
- A click required.
- A log of what was on screen when you suggested it.
How to validate the idea
- Pick one screen you use every day. Write the three actions you wish were one click. Ship one of them without any model.
- If people use the click, add a model that chooses among a list you wrote. If they do not, the decision API will not save it.
- Time the agent that drives the mouse against the version that only suggests. The post's 10x is a claim to test on your own task, not a number to put on a landing page.
How to differentiate
- Suggest, then wait. Auto-click is off.
- One app, not every window on the machine.
- The decision is a label you can audit. The agent is the second step.
- Use the API. Do not pretend to be the model lab.
Why this could be interesting now
A full computer-use loop is still slow and still needs a person nearby. The Decisions post is a public example of deciding first and moving second. The company is the suggestion a named user agrees to.
Original product launch
This launch is inspiration and evidence that someone shipped, not a partnership. The idea above is our reading of the broader opportunity. OpenAI did not write it.
- Product
- Decisions API
- Company
- OpenAI
- Film
- Decisions API, public beta
- Posted
- Oct 6, 2026 · 0:46
Product that inspired this idea
A public beta of an API that turns a screen, or text, into a suggested action. The post says a click then hands the task to an agent, and claims that demo was about 10x faster than a traditional computer-use loop.
Posted by @RivMist. Watch the original launch. No separate product site is listed unless the post itself is the source.
- api
- agents
Other launches from OpenAI
Related startup ideas
- Sign-in that belongs to the agent
Let an agent sign in as itself, with a human owner on the record, instead of borrowing a password.
- Steer an agent without cancelling its current work
A way to add a message to an agent’s next turn while a subagent keeps going.
- The passage, not the whole page
Give an agent the paragraphs that answer a question, with the surrounding lines, instead of a scraped page.
- One bill for the tools an agent calls
Let an agent discover a tool and pay per call, instead of a subscription to every vendor it might need.
- A cloud machine that builds the mobile app
A remote computer whose job is shipping a mobile app with the coding agent you already pay for.
- A drive that files the document and cites the page
Upload a pile of files, have them land in a folder a person would recognize, and answer a question with the page it came from.
Looking for more ideas?
The full list is startup ideas for 2026, each one tied to a launch you can watch.
Browse startup ideas for 2026