Greg Isenberg watched almost 60 minutes of Sam Altman on stage and came away saying one thing: a handful of these launches will make people billions of dollars in aggregate.
Not OpenAI. Builders like you.
I just went through his full breakdown of OpenAI DevDay 2026, and I think he's right to be loud about it. There were 20+ feature launches on that stage. Greg says only 3 or 4 actually matter, and they all point at the same shift: ChatGPT is turning into an app ecosystem with 1.2 billion weekly active users behind the front door.
If you're a bootstrapped founder, that number should stop you mid-scroll.
Here's the whole episode. Watch it, then read on for what it actually means if you're building toward $100K MRR.
Dots: the personal agent platform with 1.2 billion users behind it
The headline launch is Dots, OpenAI's personal agent platform. It's their answer to Meta's Muse and to Instinct.
The mechanic is simple and brutal. A user asks ChatGPT for something real: "I need this permit filed", "find me a restaurant that does X", "clean up my product catalog". ChatGPT understands the job. Then it picks a plugin to do the work.
Your plugin, if you built the right one.
Greg's phrase for this stuck with me: you can get hired before the customer knows your name.
That's a different distribution game than anything we've had. No landing page, no ad spend, no SEO wait. The model routes intent straight to whichever mini app does the job best.
And yes, OpenAI tried plugins before. Greg's paraphrase of Sam on stage: the first version got mixed reviews. This time they're doubling down, and with Dots as the front door to 1.2 billion weekly active users, Greg thinks the transaction volume flowing through personal agent platforms reaches billions of dollars soon, and he calls that estimate conservative.
I covered the other side of this fight when Meta shipped its own version. If you want the full picture of the platform war, read my breakdown of Meta's Muse AI connectors next to this one. Same battle, two giants, and the builders in the middle win either way.
Sign in with ChatGPT changes your cost structure overnight
This one was so low-key Greg couldn't even find it on the DevDay site. He still thinks it's the biggest deal of the day, and I agree.
Users can now log in to your app with their ChatGPT account and burn their own token allowance inside your product.
Their tokens, not yours.
Until now, every AI app carried inference costs from day one, which is why everything charges $20 to $30 a month before delivering a single result. That pricing wall killed thousands of niche ideas that could never support a subscription.
Now you can ship the core app free, let users bring their own compute, and charge for the layer on top: team seats, enterprise features, more data, saved workflows.
Greg's comparison is Facebook Login. When Facebook opened its login and social graph, it created multiple billion dollar companies. Yelp rode that wave early: log in with Facebook, see reviews from your actual friends.
His idea list for this new wave is deliberately tiny:
- 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
Notice the pattern. Categories too niche for OpenAI to ever build as a feature, but plenty big enough to feed a bootstrapped founder. Ideas that were not economically viable last week are viable today.
The Decisions API and the Agents API: pick your infrastructure fight
Two more launches matter for builders.
The Decisions API is OpenAI's answer to Jev. You feed it context (text or images) plus a finite set of predefined answers, and it returns a decision you can act on in real time: classify this, route that, choose the agent's next step. It's in limited preview.
Greg's favorite underrated use case: conversion. Point a decisions engine at your app store screenshots or your landing page variants and let it pick the highest converting option. Whether OpenAI's version beats Jev is an open question, but the category is real.
The Agents API now supports computer use, and it inherits Codex's multi-agent capabilities: tool search, tool calling, context compaction. OpenAI runs the infrastructure. You bring the vertical.
That last part is the opportunity. Greg calls these vertical uses. If you have proprietary data, a strong brand, or you simply understand one industry better than a generalist ever will, you can now ship an AI native app without building any of the agent plumbing yourself.
This is the same logic behind the AI rollup wave, where operators buy boring service firms and run them on agents. I wrote up Greg's $5 trillion AI rollups thesis if you want to see where vertical AI plus operational businesses is heading.
The framework: trigger, decision, action, feedback
Here's the mental model Greg uses to think about building in this world, and it's the most useful 60 seconds of the episode.
Every agentic transaction has four parts:
- Trigger: something happens. An invoice comes due, inventory runs low, it's tax time.
- Decision: ChatGPT figures out what should happen next.
- Action: an app actually does the thing. Files it, submits it, fixes it.
- Feedback: was it approved? Rejected? Did it work?
The platform owns the decision. That's the part you will never own, so stop fighting for it.
Greg's bet: the strongest businesses own the trigger, the action, and the feedback. If your product knows when the invoice is due, executes the fix, and confirms the result, the platform needs you on both ends of its decision.
→ Own the ends of the loop. Rent the middle.
This is also where boring operational work becomes a moat. The founders I interview who win with automation don't sell "AI". They sell a completed loop: lead comes in, gets qualified, gets followed up, gets closed. Agencies like TwiLead already productize exactly that kind of loop as business automations for companies that don't want to build it in-house. The agentic era is that same playbook with a much bigger router on top.
Two businesses Greg is begging someone to build
He threw out two specific ideas in the episode. Both are real companies waiting for an owner.
1. The operational work API. Millions of Dots requests will be digital. But plenty are structured operational jobs: file this permit, clear this shipment through customs. Someone needs to build the vetted network of permit expediters and customs brokers that an agent can hand the last mile to, and take a transaction fee on every job. All the software in the world still can't walk into a county office.
2. Analytics for agent discovery. Right now thousands of teams are asking the same question: how do I get my plugin surfaced inside Dots? And Muse? And Google's Spark, and Instinct, and Grockbot? Someone will build the layer that generates realistic test prompts, measures which plugin gets selected, tracks activation share by intent, and flags metadata problems that cost you invocations.
Every platform shift mints this company. The web era gave us Google Analytics, Mixpanel, and Amplitude. The AI search era gave us Profound, which Greg notes raised at a valuation somewhere in the billion dollar range. The personal agent era's version doesn't exist yet.
(If you build either of these, my DMs are open. I want that episode.)
What about GPT 6.1 Soul and the rest?
Yes, the model news was big too. GPT 6.1 Soul is a major upgrade. Ultrafast is exactly what it sounds like, way faster at the cost of more tokens. Private intelligence and Codex Security Cloud target enterprises scared to hand over their data. Spaces bolts agent-native slides and docs onto the platform.
All fine, and all expected.
But Greg's point, and mine, is that model upgrades don't change your business. Distribution changes your business. Dots, Sign in with ChatGPT, the Decisions API, and the Agents API are distribution and cost-structure changes, and those four are what actually moved for a solo founder this week.
Sell workflows, not tokens. The moats left standing: proprietary information, a specialist workflow, access to outside systems, auditability, a network of human suppliers, and a measurable business result.
None of those got commoditized on that stage. They got more valuable.
FAQ
What was announced at OpenAI DevDay 2026?
Over 20 launches. The four Greg Isenberg flags as founder-relevant: Dots (a personal agent platform reaching ChatGPT's 1.2 billion weekly active users), Sign in with ChatGPT (users bring their own token allowance to your app), the Decisions API (real-time classification and routing, in limited preview), and the Agents API with computer use. GPT 6.1 Soul, Ultrafast, private intelligence, Codex Security Cloud, and Spaces shipped too.
What is Dots in ChatGPT?
Dots is OpenAI's personal agent platform, its answer to Meta's Muse and Instinct. Users hand ChatGPT a job, and it selects a third-party plugin to complete it. For builders, it's a new distribution channel where the model routes customer intent directly to your mini app, before the customer ever searches for your name.
How does Sign in with ChatGPT help small SaaS founders?
It removes inference costs from your side of the table. Users log in with their ChatGPT account and spend their own tokens inside your app, so you can offer a free core product and charge for advanced features. Niche tools that couldn't justify a $30 a month subscription became viable overnight.
Is it too late to build on OpenAI's platform?
The opposite. Plugins are relaunching now, the Decisions API is still in limited preview, and nobody owns agent discovery analytics yet. Greg's framework: own the trigger, the action, and the feedback loop in one niche you understand better than a generalist. The platform decides; you execute.
I interview bootstrapped founders every week about exactly this: catching a platform shift early and turning it into revenue.
Real numbers, real playbooks, from people at $100K to $10M a year. No hype cycles, just what's working.
