At DevDay 2026, OpenAI outlined changes that could turn ChatGPT into a place to discover, publish and use software, according to reports. TechCrunch's Sarah Perez framed the news as a possible blow to the traditional app store model, and it became a Techmeme headline on October 4, 2026. Whether or not every prediction holds, the direction is worth understanding, because it points at a different answer to a very ordinary question: do you still need to install an app?

01What OpenAI changed at DevDay 2026

The reporting describes a shift rather than a single feature. Instead of asking you to leave ChatGPT, download something and come back, the direction is to let software run inside the conversation. Third-party services become capabilities the assistant can call while it answers you.

That reframes what an app means on this platform. The unit is no longer a screen full of buttons you tap; it is a task the assistant can finish on your behalf. Coverage of the event treats the announcement as a competitive threat to the established app stores, though how far it goes depends on details — publishing rules, pricing and availability — that are still developing.

What is confirmed versus reported

The event and the general direction are covered widely. The specific mechanics — who can publish, how money moves, which countries get access first — are described in reporting, not confirmed product documentation. Treat the details as reported until OpenAI publishes rules you can read.

02What using software inside ChatGPT really means

When you ask an assistant to do something — split a bill, pull a value out of a document, check a flight — it has to decide which capability to use and how to use it. It makes that decision by reading a structured description of each tool: a name, a short summary, and a schema for the inputs the tool accepts.

That description is usually JSON. The model does not see your app's buttons or screens; it sees a contract that says what the tool does and what arguments it takes. This is the part many people miss about an AI app store. The discovery surface is text, so the quality of your description becomes the quality of your store listing.

A tool with a vague description gets skipped or called with the wrong arguments. A tool with a precise schema gets chosen reliably. On this kind of platform, the packaging layer matters as much as the code behind it.

03Do you still need to install apps?

Mostly yes, for now — but the number of apps you open directly may fall. The app still exists; what changes is the front door you use to reach it.

The pattern is the same in every row. You stop choosing an app before you know whether you need it, and you start describing the outcome you want.

04From a request to a finished task

Here is an illustrative example — not a real DevDay release — of how a small tool introduces itself to an assistant, and what the user sees as a result.

The tool publishes a description of itself as JSON:

{
  "name": "split_bill",
  "description": "Split a restaurant bill evenly across a group, with an optional tip.",
  "parameters": {
    "type": "object",
    "properties": {
      "total": { "type": "number", "description": "Bill total before tip" },
      "people": { "type": "integer", "description": "How many people share the bill" },
      "tip_percent": { "type": "number", "description": "Tip percentage" }
    },
    "required": ["total", "people"]
  }
}

When you make the request, the assistant matches it to that contract and reports the result:

You: Split a $96 dinner among 4 people with a 20% tip.

ChatGPT: The bill is $96.00, the tip is $19.20, and the total is $115.20.
That is $28.80 per person.

The schema does the quiet work. It tells the model that people is a whole number and total is numeric, which is why the answer comes back without guesswork. Change the schema and the behavior changes with it.

Read the schema, not just the description

A friendly one-line description can hide a tool that asks for far more data than it needs. The parameter list is the honest part of a tool's listing, so check it first.

05What changes for everyday users

The upside is friction. Fewer downloads, one account, one interface, and a box that understands a sentence instead of an exact keyword. For a task you do once a month, that is a real improvement: you no longer install an app you will delete next week.

The trade-offs are just as real. When software runs inside a conversation, you cannot see the screens it would have shown you, so you rely on the assistant's summary of what it did. Permissions become harder to notice. And an assistant that reads web pages or documents can be steered by text it should ignore — the problem described in our guide to prompt injection. A tool marketplace widens that surface rather than closing it.

06What it means for independent developers

For a small team, the promise is simpler distribution. There is no store ranking to fight and no need to build a separate version for every platform just to test an idea. If discovery happens through a model, being chosen depends on how clearly you describe what your software does.

That is a different skill from polishing an icon and writing screenshots. The practical work shifts toward precise schemas, honest capability descriptions and dependable error handling, because the assistant decides whether to call your tool before a human ever sees it. The same discipline shows up in coding agent harnesses, where a bad tool description quietly degrades every run.

The open questions are substantial. Reporting does not settle how publishing, review or payment will work, or whether all of this is open to every developer. The opportunity is promising but unconfirmed, and the winners will be the people who prepare for the rules rather than assume them.

07How the app store model could shift

App stores earn from a mix of distribution, discovery and billing. A conversational layer can insert itself into all three at once.

If even part of that shift happens, the contest stops being about which app looks best and becomes about who controls the front door. That is why incumbents rarely treat a new discovery channel as a neutral feature.

08What to watch next

A few signals will show whether this is a novelty or a genuine shift:

1. whether ordinary developers can publish tools, or only selected partners can;
2. whether users can see and control what each tool is allowed to do;
3. how payment, refunds and disputes are handled when the assistant mediates the sale;
4. whether the same tool reaches other assistants, or stays locked inside one.

While you wait, keep one habit that pays off either way: make the structured data around your work clean. Formatting and validating a JSON description is a quick way to see whether a schema says what its author thinks it says, and this site's JSON Formatter is built for exactly that check.