OpenAI DevDay 2026: Dots, Agents, and the $100B Opportunity
Greg Isenberg watches sixty minutes of Sam Altman's Dev Day keynote and picks the four launches, out of more than twenty, that he thinks hand outside builders a real business.
Posted
yesterday
Duration
Format
Review
hype
Views
112K
1.4K likes
57 · 43
Big Idea
The argument in one line.
Four of OpenAI's twenty-plus Dev Day 2026 launches, personal agents, a decisions API, a computer-use agents API, and sign-in with ChatGPT, hand independent builders a distribution channel and a billing model they didn't have before.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A developer or indie builder deciding whether to build a ChatGPT plugin, agent, or AI-native app before the space gets crowded.
A SaaS founder trying to figure out how to make an AI app free to try without eating the inference bill.
Someone hunting for a defensible AI business idea instead of a thin wrapper around a model.
SKIP IF…
You want a hands-on build tutorial; this is a keynote summary and business analysis, not a build-along.
You already watched the full Dev Day keynote and know the announcements in detail.
TL;DR
The full version, fast.
OpenAI's Dev Day 2026 keynote produced twenty-plus feature launches, but four matter for builders: Dots, a personal agent platform routing user requests to plugins; the Decisions API, a lightweight classifier for real-time routing; the Agents API with computer use, a managed harness for agents that operate software; and Sign In With ChatGPT, which lets a site bill against a user's existing plan. The mechanism: plugins and login now let a builder ship a free core app funded by the user's own subscription, then charge for advanced features once they're hooked. The actionable framework is trigger, decision, action, feedback, and the strongest businesses own the trigger, the action, and the feedback rather than just the AI decision in between.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Isenberg explains he wasn't planning to cover Dev Day, then narrows twenty-plus launches down to the three or four he thinks matter for builders.
01:48 – 03:29
02 · Dots, the personal agent platform
OpenAI's answer to Meta's Muse and Instinct: a personal agent that understands a user's request and hands it to a plugin able to deliver it, with access to 1.2 billion weekly active users.
03:29 – 05:02
03 · OpenAI doubles down on plugins
Sam Altman describes past plugin reception as mixed and reframes ChatGPT as an app ecosystem, with plugins competing to fulfill a job end to end.
05:02 – 06:43
04 · Decisions API
A limited-preview API that returns a classification from a fixed set of predefined answers, positioned as OpenAI's answer to a decision-routing tool Isenberg calls 'Jev'.
06:43 – 08:01
05 · Agents API with computer use
The Agents API adds computer-use and Codex's multi-agent tooling, giving vertical-specific builders a managed harness instead of infrastructure to run themselves.
08:01 – 12:36
06 · Sign in with ChatGPT
Users can bill a third-party app's token usage against their existing ChatGPT plan, which Isenberg compares to Facebook Login's effect on early apps like Yelp, unlocking a free-core-app model and a wave of previously uneconomical niche AI categories.
12:36 – 13:46
07 · The trigger, decision, action, feedback framework
A four-step mental model for any agentic business, with the advice to own the trigger, the action, and the feedback, since owning only the decision leaves you replaceable by a better model.
13:46 – 15:04
08 · Idea: real-world work APIs
Route structured real-world jobs surfaced by personal agent platforms to vetted human specialists, permit expediters, customs brokers, for a transaction fee.
15:04 – 17:05
09 · Idea: analytics for agent discovery
Build the measurement layer that tells a plugin builder which agent platform selected them, why, and whether it converted, the Mixpanel or Profound opportunity for the agent era.
17:05 – 18:56
10 · Closing: the rest of the launches
A fast rundown of GPT 6.1 Sol, Ultrafast, Private Intelligence, and Spaces, plus a note that Isenberg reads and replies to comments himself.
Atomic Insights
Lines worth screenshotting.
OpenAI's Dots gives ChatGPT plugin builders a distribution front door to 1.2 billion weekly active users.
OpenAI is reviving plugins after admitting mixed reviews the first time, betting ChatGPT becomes the app ecosystem for AI.
The Decisions API returns a classification or routing decision from a fixed set of predefined answers, which is what makes it fast enough for real-time use.
The Agents API now ships computer use plus Codex's multi-agent tooling, so a developer can build software-operating agents without running the infrastructure.
Sign In With ChatGPT lets a website bill against a user's existing ChatGPT plan, the same unlock Facebook Login gave companies like Yelp.
A free core app funded by the user's ChatGPT subscription removes the biggest objection to trying a new AI tool: paying another $20 to $30 a month.
Tiny AI categories that couldn't support a standalone business when builders had to pay for inference now can, because the user's login pays for the tokens.
The winning business model shifts from selling tokens to selling a workflow: proprietary data, a specialist process, or a network of suppliers.
A four-step framework, trigger, decision, action, feedback, describes any agentic business, and the strongest ones own two of the three non-model steps.
If a business only owns the decision step, a better model from OpenAI can replace it overnight.
One open idea is a real-world work API that routes structured jobs from personal agent platforms to vetted human specialists for a transaction fee.
Another open idea is an analytics layer that measures which plugin gets selected across every agent platform, the Mixpanel or Profound of the agent era.
Takeaway
Four launches, four different business models to build on.
WHAT TO LEARN
Dev Day shipped a distribution channel, a billing model, and a decision engine in the same keynote, and the businesses that win will own the steps around the AI rather than the AI step itself.
02Dots, the personal agent platform
Dots gives outside developers a front door to ChatGPT's 1.2 billion weekly active users, provided a plugin gets selected for the job.
The model watches a job after the first hand-off: a rejection reopens it, a missing file triggers a request, an approval sends a notification, so a plugin has to handle a full lifecycle, not a one-shot answer.
03OpenAI doubles down on plugins
OpenAI is reviving plugins after admitting they had mixed reviews the first time, betting a better selection model fixes the discovery problem plugins had.
The plugin path only pays off if it resolves the customer's request end to end, since ChatGPT is routing on job completion, not just relevance.
04Decisions API
The Decisions API is a narrow classifier, not a general chat model: it returns one answer from a fixed set of predefined options, which is what makes it fast enough for real-time routing.
UI and conversion-rate optimization, picking the highest-converting screenshot or landing page, is flagged as an underused use case for this class of decision API.
05Agents API with computer use
The Agents API's managed computer-use harness means a builder doesn't have to run the agent infrastructure, sandboxing, tool calling, context management, just the vertical logic on top.
The opportunity favors builders who already have a proprietary dataset, an established brand, or deep knowledge of one vertical, since OpenAI is commoditizing the generic harness underneath.
06Sign in with ChatGPT
Sign In With ChatGPT lets a user's existing plan pay for a new app's token usage, the same unlock Facebook Login gave apps like Yelp: instantly lower friction to try something new.
A free-to-use core app funded by the user's own subscription, monetized later through enterprise features, team sharing, or extra data, is now viable because the builder isn't footing the inference bill.
Categories too small to be worth OpenAI building natively, a CAD file cleaner, a single-scientific-field researcher, a franchise-specific contract reviewer, are now economically viable as standalone products.
The defensible version of these products sells a workflow, not tokens: proprietary information, a specialist process, access to outside systems, auditability, a supplier network, or a measurable result.
Every agentic business reduces to four steps: a trigger fires, a decision gets made, an action executes, and feedback closes the loop.
The strongest position is owning two of the three non-model steps, trigger, action, feedback, because if a business only owns the decision, a better model can replace it.
08Idea: real-world work APIs
One open opportunity is a real-world work API that routes structured jobs surfaced through agent platforms to vetted human specialists, like permit expediters or customs brokers, for a transaction fee.
The moat in a real-world work API isn't the software, it's the supply side: the quality and reliability history of the human network doing the last-mile work.
09Idea: analytics for agent discovery
A second open opportunity is an analytics and optimization layer for agent discovery: measuring which plugin gets selected, why, and how invocation turns into completion, across every personal agent platform.
Isenberg compares this to how Mixpanel, Amplitude, and Google Analytics won the web-analytics era, citing Profound's outsized valuation as proof that visibility tooling for AI answer engines already commands real money.
10Closing: the rest of the launches
OpenAI also shipped GPT 6.1 Sol, a faster Ultrafast mode, Private Intelligence for data-retention-sensitive enterprises, and Spaces, a cloud docs and slides surface for AI agents, in the same keynote.
Glossary
Terms worth knowing.
Dots
OpenAI's personal agent platform inside ChatGPT that understands what a user is asking for and routes the request to a third-party plugin that can deliver it.
Decisions API
An OpenAI API in limited preview that returns a routing or classification decision from a fixed set of predefined answers, based on text or image context a developer supplies.
Agents API
An OpenAI API that lets a developer's application use computer-use, multi-agent, and tool-calling capabilities, with OpenAI running the underlying infrastructure.
Sign In With ChatGPT
A login option that lets a user authenticate on a third-party website with their ChatGPT account and use their existing plan's token allowance there.
Plugin
A mini app built by an outside developer that ChatGPT can select to complete a specific job a user asked for inside a personal agent platform like Dots.
Vertical harness
A managed agent framework built around one industry or workflow, where an outside builder's domain data and brand differentiate the product from OpenAI's own generic tools.
“I don't think it's crazy to say that in 10, 15, 20 years that there's trillions of dollars of transaction volume happening within personal agent platforms.”
big, quotable number claim that frames the whole thesis→ TikTok hook↗ Tweet quote
08:01
“But basically, you can now sign in with your ChatGPT login and use your token allowance on those websites.”
the low-key announcement Isenberg says 'blew his mind', clean standalone reveal→ IG reel cold open↗ Tweet quote
04:14
“Because if you can do that, you can be hired before the customer knows your name.”
tight, punchy line that matches the on-screen whiteboard text word for word→ newsletter pull-quote↗ Tweet quote
11:31
“Yeah, sell workflows instead of tokens.”
four-word thesis distilled to a slogan→ TikTok hook↗ Tweet quote
16:55
“Who's going to be the profound for this personal agent era? Huge opportunity.”
open-ended challenge to the audience, good discussion-starter→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphoranalogy
I got to be real with you. I didn't think I'd be making this episode of the podcast. But I just watched almost 60 minutes of Sam Altman up on stage.
And he launched a few things that I think are going to make people, in aggregate, billions of dollars. And I don't think that's a stretch. I think that there were 20 launches, feature launches today.
But I think there was probably three or four. that are so big that I think the people who figure out how to use those features and to grab a slice of those billions of dollars of transactions that are going to be happening within the ChatGPT platform, I think they're going to be able to create some really, really big businesses.
I had to come on here and just break it down for you. If you're new here, my name is Greg Eisenberg. I'm the host of the Startup Ideas podcast.
I started and sold three venture -backed companies. I was an advisor to Reddit, TikTok. And I come on here because I think that this is an incredible time to be building your ideas.
And I think that in the age of AI, you can use tools like this to go and do that. So by the end of this episode, you're going to learn about the most important things that happened today that Sam Altman launched at Dev Day 2026. I'm going to give you some ideas and some frameworks for thinking about how you could come up with your own ideas so that you can grab a slice of that pie.
And I'm just going to cut through the noise. So by the end of it, I'm just going to really focus on... For people interested in building businesses, making money around AI, this is the episode for you.
Yes, like I said, there are 20 plus launches, feature launches today. The most important ones to me, the ones I think are the most interesting, and I think maybe for you, Let's start with Dots.
So we have to talk about Dots. Dots is the personal agent platform. And it's their answer to Muse by Meta and their answer to Instinct.
And what's really cool about it is they now have this. front door where you're going to have access to the 1 .2 billion weekly active users. That's what Sam Altman said today.
People are going to go and use their personal agent platform, DOTS, and they're going to be asking questions. They're going to say, I need this permit fixed, or I have this request, or I'm looking for a restaurant that serves this. ChatGPT is going to understand that job.
And your plugin that you have made, which is essentially like a mini app, is going to get selected for the work. And we're going to talk about ways to surface that. But they've basically announced that plugins, they're doubling down on plugins.
I think Sam said something around like, he didn't say it verbatim that it's... plugins were a failure. He didn't say that.
But he did say something to the degree, I'm paraphrasing, that they had some mixed reviews on plugins. They're now doubling down on plugins. And I believe, Sam, that he's going to make plugins a thing and he's going to try to make ChatGPT the app ecosystem for AI.
And that's a huge... announcement of today. And we're going to talk about some of the things.
We'll go deeper on that. So your business, you want your business to go and like fix the customer's request, file the permit, you know, if it's an example of like a permit company, and then you're going to want to do the result. So this is the new app ecosystem.
You need to drill this into your brain. You, you know, have to figure out how you could focus on the right customer.
ChatGPT is going to understand the job, so they're going to do that. You're going to have to create the right plugin that does the right thing and actually creates a good result. Because if you can do that, you can be hired before the customer knows your name.
You can have access to the 1 .2 billion weekly active users. And we're talking about... I said in the intro, billions of dollars of transactions.
I'm even being conservative. I don't think it's crazy to say that in 10, 15, 20 years that there's trillions of dollars of transaction volume happening within personal agent platforms. But let's be conservative and say that there are billions of dollars that will be in transaction volume.
So now that we understand that, let's go back to Dev Day's website. So you can see here's their personal agent platform. The other thing that they announced today is decision to API.
I did a video, I think it was last week on Jev. And this is their answer to Jev. So you saw their answer to Muse.
This is their answer to Jev. Decisions API enables real -time decision -making by focusing Luna's intelligence on a specific set of user -defined questions with finite predefined answers. So developers supply context using text or images and they get back answers they can use to classify content, route requests, or choose an agent's next action.
So it's in limited preview today. Excited to play around with it. I've been hooked.
on Jev. Hooked on Jev, I think there's a bunch of use cases for Jev that people are not talking about. It's underserved.
One that I think is really interesting is just UI design. Essentially, Jev is a decisions API. How do you use a decisions API to get you the most highly converted converting app store screenshots?
direct response websites and apps, things like that. So there's a ton of opportunity there. Question here is, will OpenAI's decision API be worse, as good, or better than Jev?
TBD, we will find out really soon, but I think it's worth paying attention to. The other big launch of today that I think is really interesting is the Agents API is with computer use. So the Agents API now supports computer use so developers can build agents that interact with software to complete tasks.
It also brings Codex's multi -agent capabilities, tool search, tool calling, and context compaction into your application. OpenAI runs the underlying infrastructure so your team can focus on building applications. I've talked about this before around vertical harnesses.
You can see here, OpenAI, it's a managed Conex harness. If you have great data, if you have a great brand, if you understand a vertical better than other people, there's this... Huge opportunity to build apps.
And that's why I would be playing with the agent API today. Go and figure out what you can be building, how you can extend yourself, extend your business to this world. And this is just like a new surface area.
It's like thinking about how you can create AI native apps. And it's just, I think there's a lot here. I don't think a lot of people are doing it.
And obviously that excites me. The other low -key, huge announcements, I don't even know if it's on this page. That's how low -key it is.
But when I saw it, it sort of blew my mind. But basically, you can now sign in with your ChatGPT login and use your token allowance on those websites. This is going to open up thousands of business opportunities in the same way that when Facebook launched their login, it opened up thousands of business opportunities and actually created many billion dollar companies.
Think about Yelp as an example. I remember when Yelp came out, you could go check out what their early product looked like. multi -billion dollar company that was basically log in with your Facebook and you're going to see local reviews and reviews that your friends have given.
And it had that Facebook graph so it allowed you to do more. Now, where do I see the opportunity? We're going to go into depth into this in a bit.
But where I see the opportunity with sign in with ChatGPT is now you can create a free app. You don't charge people, so your top of the funnel becomes bigger. It's connected with your chat GPT, and then you can upsell them to other products and services from there.
That is a huge, huge, huge thing. Let's go back here. Here's my word dump of chat GPT login.
I think a lot of people look at AI apps and they're like, do I really want to spend $20 or $30 a month for another AI app? It's like, no. You can just sign in with chat GPT.
And start using the plan you already have. So I think a lot of people, if you're smart, you create your core app free and then you charge for advanced features. So you can charge for enterprise stuff, you can charge for sharing with your team, you can charge for more data.
But now you have this opportunity to create the core app for free. In the past, you couldn't create the core app for free because you were paying for inference. But now you can just focus on, your business, building the distribution, building the brand, getting the data sets, obviously getting those people from the brand into the app, get them to use the app, and when they start seeing value, upsell them to other things.
A few ideas around... What you can do with the login, an agent that cleans architectural CAD files, a researcher that monitors one specific scientific field, a Shopify catalog cleanup desktop app, a video production tool built around one repeatable editing workflow, a contract review tool for one type of franchise agreement.
Basically, what's the opportunity? These tiny... ai categories um that weren't viable weren't economically viable today as of today are so you want to build these products that are too niche to become you know major ai open ai features but you know you you you have enough of an audience that you can charge you know for upselling and for that workflow.
So I think this is a really big deal and I'm really excited about it. My mind is going absolutely, I can't stop thinking about it actually. And now we can focus on selling workflows, right?
Yeah, sell workflows instead of tokens. So if you're ideating, if you want to build a business in this agentic age, what's some proprietary information? that you have.
A specialist workflow, access to outside systems, collaboration and auditability, a network of human or suppliers, a measurable business result. These are the things you can focus on and we don't need to really focus on tokens. And I think you're also just going to get a lot more users of your app.
You get their emails, you might get their phone numbers. And then you could email them, you can sell ads in those emails.
There's just so much more that you can do now that I think it's really, really interesting. I wanted to just talk about the mental model I have of this new world where we have decisions API, we have login with ChatGPT, we have dots, we have the agents API. How am I thinking about building a business in this world?
To me, it's this four -step framework. Trigger, decision, action, and feedback. There's going to be a trigger, an invoice becomes due, the inventory runs low, or it's tax time.
Something happens such that I'm going to ChatGPT and I'm asking for some help, or it does it autonomously. Then there's a decision. ChatGPT is like, okay, what should I do next?
Then there's an action.
There's a correct, there's a submit, there's an ability. for an application to do something. And then there's feedback.
Was it approved? Was it rejected? And I think that the strongest businesses are going to own two of three.
So they're going to own trigger, they're going to own action, and feedback. So when you're thinking and you're brainstorming around what you can be building, I would be focused on those three things. I wanted to give you a couple ideas just to get your creative juices flowing around, okay, What other businesses could you build now that Sam Altman and OpenAI team have launched this?
One is like a real world API. So in a world where you're going to have millions of people basically going to dots and asking it for stuff, some of those stuff are going to be digital. But some of them...
Some of those are going to be digital. Some of them are going to be structured jobs. So if you can figure out a way to get vetted specialists in the real world, I think that that's a huge opportunity.
So like permanent expeditors, custom brokers, and then you make a transaction fee. So I think my point is there's all this stuff happening in the physical real world. A lot of people are going to be focusing on the digital stuff.
How can you create... real -world work APIs in specific niches that you can connect dots or connect chat GPT to real -world workers who can finish the last mile of the transaction. Huge opportunity.
Throwing it out there. The second is, who's going to build the analytics and optimization layer for agent discovery?
I'm not the only one who's thinking about building plugins right now. At this very moment, there's probably thousands of companies thinking about, how the hell do I get surface within dots? Not just dots.
How do I get surface within Muse? How do I get surface within Gemini? Google's platform, I think it's called Spark.
There's all these personal agent platforms, Instinct. There's all these platforms, Grokbot. How do you get your plugin surface?
Some of these have plugins, some of them don't. But I think eventually they all will. If you can create an analytics platform that generates realistic, indirect prompts, which measures which plugin gets selected, which tracks activation share by customer intent, which finds metadata and tool definition problems, which measures whether invocation turns into completion, tests every plugin update.
There's a huge opportunity here. In the same way that... In the web era, you had huge companies.
Mixpanel came out. Amplitude came out. Google Analytics was a huge product.
I think there's tons of analytics products that are going to come out around helping you be seen. You actually saw that with AEO, GEO. I think Profound raised a $2 billion valuation or a billion dollar valuation.
Some absurd large amount. They deserve it. They're generating a ton of buzz and revenue and they're solving a real problem.
My point is, who's going to be the profound for this personal agent era? Huge opportunity. I hope someone takes that.
I'll end with, they obviously launched other stuff. They've obviously launched GPT 6 .1 Sol, which was huge, which is obviously a major upgrade to GPT 6, but we kind of expected that. They obviously launched UltraFast, which is amazing, by the way.
If you haven't tried UltraFast, it's absolutely amazing. You're going to love it. It's just way faster.
Of course, it's a lot more tokens, but of course they launched this. And they obviously launched private intelligence, I think a lot of companies are scared to use Frontier AI because their data isn't really protected.
I'm not surprised that they're doing things like that and doing stuff like Codex Security Cloud. But I think the point of this whole episode was there's a few main features, Decision API, Agents API, the login with ChatGPT and DOTS that I think... If you're listening to this, you're a founder, you're someone who wants to build a company, wants to bet on yourself.
That's going to be huge.
They also launched Spaces, which is basically their version of having cloud slides and docs. You can bring your AI agent to work with it. It's cool and it's going to help make people more productive.
For people listening to this, for my people, for the idea junkies, the people who want to build businesses, I thought this would be most interesting to you. Hope it was. Let me know if it was.
I read every single comment, YouTube comment. I reply to a lot of them too. Have a creative day and I'll see you next time.
The Hook
The bait, then the rug-pull.
Greg Isenberg watched sixty minutes of Sam Altman's Dev Day keynote so you don't have to, then picked the four launches, out of more than twenty, that he thinks can turn into real revenue for outside builders.
Frameworks
Named ideas worth stealing.
12:36model
Trigger, Decision, Action, Feedback
Trigger
Decision
Action
Feedback
A four-step model for any agentic business: something triggers a request, ChatGPT or an agent makes a decision, an application takes the action, and feedback closes the loop. The strongest businesses own the trigger, the action, and the feedback, leaving only the decision to the model.
Steal forframing any AI-agent product pitch, or stress-testing whether a niche AI idea has a real moat
10:04list
What Makes a Workflow Worth Charging For
Proprietary information
A specialist workflow
Access to outside systems
Collaboration and auditability
A network of humans or suppliers
A measurable business result
Six qualities Isenberg says separate a chargeable AI workflow from a thin wrapper OpenAI could clone or absorb.
Steal forevaluating whether a niche AI app idea has enough of a moat to charge for
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A two-person breakdown of OpenAI's top-tier model that skips the game demos and goes straight to code audits, nine money-making agent prompts, and a Raspberry Pi speaker built and shipped in about 30 minutes.
A solo walkthrough of five open-source GitHub repos getting traction right now, an AI writing editor, an agent-run CRM, a video-editing agent, a skill security scanner, and a phone-controlling harness, each with the exact install command and the first small workflow to try.
Greg Isenberg and developer Vinny break down WebMCP, the experimental browser feature that lets any AI agent search, compare, and buy on a website without scraping the page, then price two startup ideas built on it.
Cloudflare quietly built the payment rail for AI agents to pay tiny fees for web resources, and that rail is about to spawn a whole generation of businesses selling clean data, agent-readiness, and packaged expertise.
Cody Schneider maps the exact infrastructure — pipeline, warehouse, agent — behind a Facebook ads system that researches, creates, publishes, and kills its own losing ads.
Nick Vasilescu spins up an AI co-founder running on Grok 4.5 and, in one live session, takes it from idea to landing page, thumbnail, and cold-email sequence.
00:15
00:39
00:43
01:03
01:14
01:31
01:42
01:55
02:09
02:38
02:52
03:07
03:21
03:35
03:48
04:15
04:28
04:42
04:55
05:05
05:23
05:38
05:52
06:21
06:35
06:50
07:06
07:37
07:53
08:11
08:17
08:37
08:51
09:06
09:20
09:34
09:49
10:18
10:32
10:47
11:01
11:16
11:30
11:45
11:59
12:14
12:28
12:43
13:11
13:25
13:39
13:53
14:25
14:40
14:56
15:10
15:37
15:51
16:04
16:17
16:31
16:44
16:58
17:06
17:25
17:55
18:01
18:21
18:34
18:49
22:52
24:43
28:58
34:10
Video of the Day37:47
Video of the Day56:00