The short version: On July 9, 2026, Anthropic and OpenAI both pushed their biggest updates of the summer into the world on the same day. Anthropic expanded Claude Cowork to mobile and web. OpenAI launched a new agent called ChatGPT Work, built on its fresh GPT-5.6 model. Both do the same basic thing: you hand over a goal, not a prompt, and the AI plans the steps, uses your files and apps, and comes back with finished work. For small business owners, this is the moment AI automation for small business stops meaning “a chatbot that answers questions” and starts meaning “a system that can be assigned a task and checked on later.”
What actually happened this week
Anthropic’s rollout began July 7 and continued through the week: Claude Cowork, previously limited mostly to desktop, is now available on web and mobile too, starting with Claude Max subscribers. A Cowork session now lives in the cloud rather than on one machine, so a task started at a desk keeps running after the laptop closes, and progress can be checked from a phone. Anthropic also doubled Cowork’s usage limits through August 5 to mark the launch.
Two days later, OpenAI answered with ChatGPT Work, an agent inside ChatGPT powered by the newly released GPT-5.6 model. According to OpenAI’s own announcement, the agent “can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work,” producing documents, spreadsheets, presentations, and even simple web apps. It shipped first to Pro, Enterprise, and Edu accounts, with Plus and Business following within days.
Neither company called this a coincidence outright, but the timing is hard to read any other way. Anthropic has been building toward this with Claude’s small-business toolkit and Tag memory feature since May. OpenAI’s own three-tier GPT-5.6 pricing, previewed just days earlier, was clearly built with an agent like this in mind: the heavier “Sol” model handles the reasoning-intensive parts of a multi-hour task, while cheaper variants pick up simpler steps.
What does an AI “work” agent actually do differently?
The distinction that matters is between answering and finishing. A chatbot answers a question in one exchange. An agent like Cowork or ChatGPT Work takes a goal such as “turn this week’s Slack updates into a client-ready status report” and breaks it into steps on its own: pulling messages, drafting language, formatting the document, and flagging anything it wasn’t sure about. It can run for hours, work while a device is offline, and pick back up where it left off.
Both companies are also leaning into scheduled, recurring runs rather than one-off requests. ChatGPT Work can watch for new messages in Teams or Slack and independently turn them into updated docs or slides. Cowork supports daily, weekly, or monthly scheduled tasks tied to a connected data source. That is the practical shift: less “ask AI a question,” more “assign AI a standing job.”
Is this actually automation, or just a better chatbot with extra steps?
It is a fair question, and the honest answer is: it depends on how well-scoped the job is. Multi-step AI agents are still prone to drift on loosely defined tasks, and a business that hands over an ambiguous goal (“handle my inbox”) is more likely to get a mess than a report. Where these tools earn their keep is on narrow, repeatable, well-documented jobs, the same kind of work a business would otherwise assign to a template or an SOP. Turning a weekly sales call log into a formatted summary is a good fit. Making judgment calls about a customer refund is not, at least not without a human checking the output first.
That caveat is also the reason this counts as AI automation for small business rather than AI replacing anyone on the team. The agent produces a draft; a person still decides what goes out the door. Owners who have already tried structured, low-risk pilots with tools like these, as covered in our practical guide to AI for small business, tend to get further than owners who hand over open-ended authority on day one.
What should a small business actually do about this?
Three moves make sense this week, not a full rebuild of how the business runs:
- Pick one recurring report or document, not a whole workflow. A weekly status update, a monthly expense summary, a recurring client recap. Something with a predictable shape and low stakes if the first draft needs edits.
- Connect the real data source before judging the output. Both agents are built to pull from files, Slack, Teams, or a CRM. A test run with no connected data will look weaker than the tool actually is.
- Watch the usage cost, not just the sticker price. Multi-hour agent runs burn more compute than a single chat reply. Our coverage of ChatGPT’s workspace agent credit pricing is worth a look before scheduling recurring runs on a metered plan.
Neither Cowork nor ChatGPT Work is fully rolled out yet: Cowork’s mobile and web access is expanding plan by plan, and ChatGPT Work is still working through Plus and Business accounts. For most owners, the right move this week is a single small pilot, not a full switch.
Frequently Asked Questions
What is Claude Cowork?
Claude Cowork is Anthropic’s agent for multi-step work: it plans tasks, uses connected apps and files, and produces finished documents, spreadsheets, or slides. As of this week it runs on desktop, web, and mobile, with sessions saved to the cloud so work continues across devices.
What is ChatGPT Work?
ChatGPT Work is OpenAI’s new agent, launched July 9, 2026, and powered by the GPT-5.6 model. It takes a stated goal, gathers information from connected apps and files, and independently completes multi-hour projects like reports, presentations, or simple websites.
Do I need both Claude Cowork and ChatGPT Work for my business?
No. Most small businesses already lean toward one assistant for day-to-day work. The practical move is to try one agent on one recurring task and see whether the output needs light editing or a full rewrite before deciding whether to expand.
Will an AI work agent replace my staff?
That is not what these tools are built for, and treating them that way tends to backfire. They are strongest at producing a first draft of recurring, well-defined work, freeing a person to review and finish it rather than start from a blank page. The judgment calls, client relationships, and final sign-off still belong to your team.
If you have tried Claude Cowork or ChatGPT Work on a real task this week, what did it get right, and where did you still have to step in?
