Skip to main content
Industry News
· 7 min read

OpenAI Just Cut the Price of an Always-On AI Agent to $20 a Month

Flat-vector line-art illustration on a dark navy background: a glowing AI agent orb at a small computer terminal with task-checklist icons and a crescent moon marking overnight work, accented in electric blue

On November 19, 2026, OpenAI quietly cut the entry price of an always-on AI workforce by 80 percent. Thibault "Tibo" Sottiaux, OpenAI's ChatGPT engineering lead, posted a two-line update on X: OpenAI is starting to test dots — the proactive agents that keep working while you sleep — with a small number of ChatGPT Plus users at $20 a month, and rolling them out to all ChatGPT Business Standard users starting that same day. Until this week, dots required a ChatGPT Pro subscription from $100 a month or a Business Premium seat. The post pulled nearly ten thousand likes in a day. When an always-on AI employee costs the same as a streaming subscription, the question for a Temecula or Murrieta business stops being "is this real" and becomes "what do I safely hand a $20 agent." Here is what actually shipped, what OpenAI still has not decided, and the five moves that separate a cheap experiment from an expensive mess.

What OpenAI Actually Shipped

Dots are not chatbots. Each dot is a proactive agent with its own cloud computer that works on tasks in the background and makes progress even when you are not interacting with it. Dots reach over 4,000 apps through OpenAI's plugin ecosystem, can connect to your PC and control it on your behalf, and you can check in on any dot by opening its "computer" and watching what it is doing. They launched at DevDay in late September powered by GPT-6 Astra — we covered that launch in depth here — and can now also run on GPT-6.1 Terra. This week OpenAI also shipped iOS widgets, Shortcuts integration, iMessage and RCS support, and notification replies, which is the plumbing that lets a dot nudge you when a task finishes.

The price cut comes in two parts. The Plus test is exactly that — a "small number" of $20-a-month users, with no promise the access is permanent. The Business rollout is the real news for employers: every ChatGPT Business Standard seat gets dots, and a Standard seat runs $25 per user per month, or $20 billed annually, against the $125 Premium seat that was previously the business floor. One open question matters for budgeting: for Pro and Business Premium subscribers, dot conversations do not count toward regular ChatGPT usage limits as long as they are not complex tasks — but OpenAI has not said whether that generosity extends to the cheap seats. Assume the meter is running until told otherwise.

Why This Matters Locally

  • The always-on agent market just went mass-market. Five weeks ago your options were a $100 OpenAI dot or a $125 Premium seat. Now it is $20 — the same week Google made its AI agents free to call your business and Perplexity pushed its own always-on teammates. Every major lab has decided the subscription floor for an AI employee is about the price of a pizza night.
  • Your competition's excuse just expired. "Too expensive to try" stopped being true this week. A Murrieta plumber can put a dot on invoice follow-ups and review drafts before they send — the same workflow an enterprise team pays ten times more for. The differentiator is no longer access; it is knowing what to delegate.
  • This is the worst possible week to hand an agent your bank login. The same month OpenAI made agents cheap, the banks made them dangerous: TechCrunch reported this week that Citi, PNC, Citizens, JPMorgan, and Bank of America have all updated terms to say that anything an AI agent does in your account is on you — even if the agent gets hacked. A dot with your banking credentials is not a shortcut; it is an unbounded liability with your name on it. This is the credential discipline we laid out when ChatGPT plugins started touching bank passwords, now with bank-sized teeth.
  • Review-consequential-work is OpenAI's own rule, not ours. At launch OpenAI said plainly that dots "can still make mistakes, so always review consequential work." A $20 seat does not make the mistakes cheaper — an agent that mis-sends an invoice or emails the wrong client does the same damage on any plan.
  • Insurance is starting to price this in. As agents spread to cheap seats, agent liability coverage is becoming a real line item. If a dot touches money, orders, or customer data, ask what your policy says about AI-initiated actions before, not after.

The Five-Move Playbook

  • Run the $20 experiment on low-stakes work first. Research, competitor price checks, drafting follow-up emails, summarizing a week of inquiries. Prove the dot's judgment on work where a mistake costs nothing before you let it near anything that sends, pays, or publishes.
  • Give the dot its own accounts, never your passwords. Create dedicated logins for any tool the agent touches. Credential-sharing is what the banks just made a liability trap, and it is invisible to you when it goes wrong — the agent looks exactly like you to every system it logs into.
  • Put approval gates on anything consequential. Invoices, outgoing emails, orders, posts: the dot drafts, you press send. That one habit deletes most of the tail risk of an always-on agent.
  • Watch the usage meter for a month before you build on it. OpenAI has not said whether Plus and Business Standard dot usage counts against plan limits. Check your usage panel after real workloads, and do not move a business-critical workflow onto a seat whose limits you have not seen.
  • Remember what a dot is not: your customer's first call. A general agent inside your ChatGPT account helps you run the back office. It does not answer your phone at 7pm, quote a job, or book the appointment — the customer-facing side is a different build, and it is where the revenue actually lives.

The Bottom Line

The price of "always-on" just hit the same shelf as a family Netflix plan, and OpenAI put it inside the exact $20-and-$25 subscriptions small teams already pay for. Used right, that is a free back-office hire: a dot that chases invoices, watches reviews, and drafts the follow-ups you never get to, while you approve the ones that matter. Used wrong — your bank password pasted into a plugin, no approval gates — it is a liability the banks have already disclaimed in writing. Experiment cheap, isolate credentials, gate the consequential work. And when you are ready for the agents that face your customers — the phone that answers itself, the site that books the job — that is what we build. PepeWebTech builds AI employees from $397/mo and websites from $1,497 one-time — free setup, no contracts, cancel anytime.

Sources

  • X — Thibault "Tibo" Sottiaux, ChatGPT engineering lead, OpenAI (November 19, 2026) — the primary announcement: "we're starting to test dots with a small number of ChatGPT Plus users… We're also bringing dots to all ChatGPT Business Standard users. Rollout will start today." Post drew ~9.8K likes and 1.5M views within a day.
  • Engadget — Steve Dent (November 20, 2026) — dots reach over 4,000 apps via the plugin ecosystem, work autonomously, can connect to and control your PC, progress viewable via a dot's "computer"; originally GPT-6 Astra, now also GPT-6.1 Terra; new iOS widgets, Shortcuts, iMessage and RCS support, and iOS notification replies; dot conversations don't count toward regular usage limits for Pro/Business Premium unless complex, with no word yet for Plus/Standard.
  • The Verge — Jess Weatherbed (November 20, 2026) — a "small" number of $20/month Plus users get the always-on agents; dots were previously restricted to Pro tier subscriptions starting at $100; rolling out to all ChatGPT Business Standard users.
  • OpenAI Help Center — ChatGPT Business overview — Standard seats cost $25 per user per month, or $20 per user per month billed annually; Premium seats $125 monthly or $100 annually; seat pricing and workspace structure.
  • TechCrunch (November 18, 2026) — Citi, PNC, Citizens, JPMorgan, and Bank of America have updated terms so customers who grant AI agents account access are responsible for everything the agents do — including damage from hacks of the agent's leaked credentials; Anthropic's own finance tool is billed "read-only."