Estimated read time: 7 minutes
For three years, “AI at work” mostly meant a chat box you pasted things into. On July 9, OpenAI decided that wasn’t ambitious enough and launched ChatGPT Work—an agent that doesn’t just answer questions but actually goes and does the task: clicking through your apps, reading your files, and building the finished spreadsheet while you do something else. It’s the clearest sign yet that the AI industry has stopped selling advice and started selling labor. For a small business owner, the question isn’t whether that’s impressive. It’s whether it’s useful to you this week—and what it will cost you.
Table of Contents
What OpenAI Actually Launched
ChatGPT Work went live on July 9, alongside the full rollout of OpenAI’s GPT-5.6 model family. It’s an autonomous agent with OpenAI’s Codex engine built in. Instead of waiting for you to prompt every step, it reads data from your connected apps, runs multi-step workflows on its own, and works in the background across web, mobile, and desktop. You can kick off a task on your phone during a commute, have it run on your Mac at the office, and check the result from anywhere. The rollout order tells you who OpenAI is chasing first: Pro, Enterprise, and Edu users got it at launch, with Plus and Business users following over the days after.
What ChatGPT Work Does That Regular ChatGPT Doesn’t
The difference is agency. Regular ChatGPT is a conversation. ChatGPT Work is a coworker you delegate to. It can tap into your local files and apps directly, or spin up an embedded browser to use websites and online tools the way a person would. Plugins let you connect external systems—your CRM, your accounting tool, your project tracker. And a new “Sites” beta lets it build interactive reports, live dashboards, and small web apps as deliverables, not just documents.
The practical version: ask it to “pull last quarter’s numbers from the finance folder, build a dashboard, and draft the summary email,” and it will attempt the whole chain rather than handing you step one and waiting. That’s the leap—from tool you operate to worker you assign. If you want the fuller picture of how these systems work and where they still stumble, our rundown of AI agents breaks it down.
The GPT-5.6 Engine Under the Hood
ChatGPT Work runs on GPT-5.6, which OpenAI shipped in three flavors. Sol is the heavyweight—the one you want for complex reasoning. Luna is tuned for speed. Terra sits in the middle for everyday work. Model pricing starts at roughly $1 per million tokens, but the full Codex agent requires at least a Plus plan, with higher usage caps on the $100 and $200 Pro tiers. For a small business, the takeaway is blunt: the “smart” version and the “cheap” version are now different products, and picking the wrong one either burns money or underdelivers. We broke down which GPT-5.6 tier is worth paying for if you want to match the model to your budget.
What It Means for a Small Business
Here’s where it gets real. The pitch OpenAI is making—and Anthropic, with Claude’s Cowork layer, and Microsoft with Copilot—is that a three-person company can now punch like a ten-person one. The busywork that used to justify your first operations hire (compiling reports, reconciling data across tools, drafting the same five documents every week) is exactly what an agent is built to absorb. If it works as advertised, the leverage is obvious: you spend less on the tasks that don’t need a human and more on the ones that do. That’s a genuine shift, not a gimmick.
But “if it works as advertised” is carrying a lot of weight in that sentence.
The Catch: Credits, Costs, and Supervision
Two catches. First, money. Agent workflows don’t bill like a flat subscription—they consume tokens, and for workspace agents, credits, with OpenAI’s free ride on that billing having ended July 6. An agent that browses, reads files, and runs multi-step tasks burns far more tokens than a quick chat. The bill is variable by design, which is great for OpenAI and nerve-wracking for anyone on a fixed budget. Model your costs before you turn agents loose on recurring jobs.
Second, supervision. An agent that acts on your behalf can act wrongly on your behalf: email the wrong client, overwrite the wrong file, misread a spreadsheet and build a confident, incorrect dashboard on top of it. Autonomy cuts both ways. For anything that touches customers, money, or data you can’t easily un-break, you still want a human reading the output before it ships. “Set it and forget it” is a marketing phrase, not an operations plan.
Should You Actually Use It Yet?
Short answer: try it, scope it, don’t bet the business on it. Start with low-stakes, high-repetition tasks—the weekly report nobody wants to build, the data cleanup, the first draft of routine docs. Keep a human in the loop on anything customer- or money-facing. Watch your token and credit usage for the first month the way you’d watch a new hire’s expense account. If the leverage is real for your specific workflows, expand. If it’s mostly impressive demos that need babysitting, you’ve lost a few evenings, not your operation.
The bigger picture is hard to miss: the industry has officially moved from “assistant that talks” to “agent that acts,” and OpenAI, Anthropic, and Microsoft are all sprinting the same direction. Small businesses that learn to delegate to these tools now—carefully—will have a head start on the ones who wait until it’s idiot-proof. It won’t be idiot-proof for a while.
FAQ
What is ChatGPT Work?
An autonomous AI agent from OpenAI, launched July 9, 2026, that completes multi-step tasks across your apps, files, and the web—rather than just answering questions in a chat window.
How is it different from regular ChatGPT?
Regular ChatGPT responds to prompts. ChatGPT Work takes actions: reading your files, using an embedded browser, running workflows in the background, and producing finished deliverables like reports and dashboards.
What does ChatGPT Work cost?
The full agent requires at least a ChatGPT Plus plan, with higher usage caps on the $100 and $200 Pro tiers. It runs on GPT-5.6 (priced from about $1 per million tokens), and workspace-agent usage consumes credits that began billing on July 6, 2026.
Is it safe to let it run on its own?
For low-stakes, repetitive tasks, largely yes. For anything touching customers, money, or hard-to-reverse data, keep a human reviewing its work—agents can act wrongly as confidently as they act correctly.
Should a small business adopt it now?
Test it on low-risk, repetitive work, monitor your token and credit spend closely, and expand only if the time savings prove real for your specific workflows.
Related Coverage
- AI Agent Showdown — how autonomous agents actually work, and where they still fall short.
- GPT-5.6 Tiers for Small Business — Sol vs Terra vs Luna, and which one is worth paying for.
- AI Small Business Hiring Guide — deciding what to automate versus what to actually hire for.
Faceted Media Magazine covers business, AI, and entrepreneurship for the people building what’s next.