The decision in one paragraph. All three of these agentic work products do roughly the same thing, take a goal and complete a multi-step task autonomously, and for most small businesses the right choice is dictated less by which is best in the abstract and more by which ecosystem you already live in. If you run Microsoft 365, Copilot Cowork is the natural fit. If your team already works in ChatGPT, Codex is the path of least resistance. If you value the strongest autonomous reasoning and you work across your own files and apps, Claude Cowork is the one to weigh seriously. The differences that matter for a small team are about fit, not about which demo looked most impressive.
The bigger point is that the category itself is the story. The move from AI that answers to AI that does is the defining shift of 2026, and the convergence of all three giants on it in a single month means this is becoming standard equipment, not a frontier experiment. Learning to delegate real work to an agent now, on whichever of these fits you, is the skill that compounds.
Copilot Cowork, ChatGPT Codex, and Claude Cowork all do autonomous multi-step work, and all three shipped within three weeks in June 2026. Pick by your existing stack: Microsoft 365 shops use Copilot Cowork, ChatGPT-based teams use Codex, and teams wanting the strongest autonomous reasoning across their own files use Claude Cowork. The category, not the specific product, is the thing to start learning now.
The shift from assistant to agent
To choose well, understand what these products have in common, because it is the genuinely new thing. For three years, AI assistants answered questions and drafted text one prompt at a time, with you driving every step. These new products are agents. You give them a goal, and they plan a sequence of steps, work across multiple applications and files, execute the steps, check the results, and come back with a finished deliverable rather than a suggestion.
Microsoft describes Copilot Cowork as doing longer, multi-step tasks that span several apps. Anthropic describes Claude Cowork as taking a goal and working on your computer, local files, and applications to return a finished deliverable. OpenAI describes Codex as agentic work that plans and executes across tools. The language is nearly identical because the capability is the same: autonomous completion of work that used to require a human to chain the steps together.
For a small business, this is the more valuable mode of AI, because your bottleneck is rarely getting an answer to one question. It is getting a multi-step piece of work done when no one has the time to do it. An agent that takes the monthly report from raw data to finished document, or the campaign from brief to assets, addresses that bottleneck directly in a way that a question-answering assistant never could. That is why the whole industry pivoted here at once, and why it is worth your attention regardless of which product you pick.
Microsoft Copilot Cowork
Copilot Cowork reached general availability on June 16, 2026, and its defining advantage is where it lives: inside Microsoft 365, alongside the Word, Excel, Outlook, Teams, and SharePoint that a huge share of small businesses already use every day. It does longer, multi-step tasks that span several apps, which in a Microsoft shop means it can move across your documents, spreadsheets, email, and files as one connected workflow.
The strength is integration and proximity. If your business runs on Microsoft 365, Copilot Cowork works inside the tools your team already knows, with access to the documents and data already there, and under the Microsoft security and compliance posture you already rely on. There is no new platform to adopt and no data to move. For a Microsoft-stack business, that proximity is worth a great deal, because the integration is the thing that determines whether an agent gets adopted or abandoned, and Copilot Cowork starts already integrated.
The consideration is cost and fit. Copilot for business sits on top of Microsoft 365 licensing, and the promotional pricing that ran near 18 dollars per user per month locks in higher afterward, so the all-in cost per seat is a real number to check against the value. And Copilot Cowork is only the obvious choice if you genuinely live in Microsoft 365. If your team's real work happens elsewhere, the integration advantage that makes it compelling largely disappears, and you are paying for proximity to apps you do not use.
OpenAI ChatGPT Codex
OpenAI brought Codex into the ChatGPT app on June 2, 2026, with six role-specific business plugins covering sales, data analytics, creative production, product design, and finance functions, all running on a capable agentic model. Its defining advantage is reach and familiarity: it lives inside ChatGPT, the AI tool the largest number of people already know how to use.
The strength is the low barrier and the breadth of the plugins. If your team already uses ChatGPT, Codex is the path of least resistance to agentic work, because it is in a product they already open daily, with no new tool to learn. The data analytics plugin, which answers questions with your data and builds reports, and the creative production plugin, which turns a brief into marketing assets, are the two most broadly useful for a small business, and they extend agentic work to analysts and marketers who are not developers. It is available across ChatGPT plans, so you can begin on the plan you already pay for.
The consideration is the consumption cost and the newness. Agentic work burns far more usage than chat, so the meter runs faster than teams expect, and the business plugins are new as of June 2026, which means rough edges and uneven reliability are normal. Codex is the natural pick for ChatGPT-centric teams who want to extend into agentic work without changing tools, and a less obvious one for businesses whose center of gravity is Microsoft 365 or who have standardised on Claude.
Anthropic Claude Cowork
Claude Cowork is Anthropic's agentic product for knowledge work: you give it a goal and Claude works on your computer, local files, and applications to return a finished deliverable. Its defining advantage is the underlying model's strength at exactly the kind of careful, multi-step reasoning that autonomous work demands, paired with Anthropic's strong record on lower hallucination rates.
The strength is reasoning quality and working across your own materials. For tasks that require the agent to hold a complex goal, work through it carefully, and handle your actual files and applications rather than a walled-garden environment, Claude Cowork is built for that, and the model's accuracy advantage matters most precisely when an agent is running many steps unsupervised, because an error early in an autonomous chain propagates. For a small business whose agentic work involves analysis, long documents, or anything where a hallucination would be costly, the accuracy floor is a real reason to weigh Claude Cowork seriously.
The consideration is ecosystem fit and polish for non-technical users. Anthropic is strong on model capability and has been catching up on the surrounding integrations and the consumer-grade ease that Microsoft and OpenAI offer. If your team is not already in the Claude ecosystem, adopting Claude Cowork is a more deliberate choice than defaulting to the agent already inside the tools you use. It is the right pick when reasoning quality and accuracy on your own files are your priorities, and a less automatic one when convenience and existing-stack proximity matter most.
How to choose between them
For most small businesses, the honest truth is that all three are good enough that the deciding factor is not raw capability but fit, and fit comes down to three questions you can answer quickly.
First, where does your team already work? If the answer is Microsoft 365, that points hard at Copilot Cowork. If it is ChatGPT, that points at Codex. If your team has standardised on Claude, that points at Claude Cowork. The agent that lives in the tools you already use has a large, often decisive, adoption advantage, because the integration work is mostly already done and your team does not have to learn a new home.
Second, how sensitive is your work to accuracy? If your agentic tasks involve analysis, long documents, contracts, or anything where a confident wrong answer is costly, the reasoning and accuracy advantages weigh toward Claude Cowork even against an ecosystem default. For lower-stakes operational and creative work, that advantage matters less and convenience wins.
Third, what is your real all-in cost? Each of these has a different pricing shape, Microsoft's per-seat licensing, OpenAI's plan-plus-consumption, Anthropic's plan-plus-consumption, and the agentic usage underneath all of them burns faster than chat. The product that is cheapest on the headline may not be cheapest in practice once you account for how much agentic work your team will actually run. Check the all-in number, not the sticker.
The answer by the stack you already run
Condensed to the decision most small businesses will actually make, based on the single biggest factor, the stack you already run.
If you are a Microsoft 365 business, start with Copilot Cowork. It is already inside the tools your team uses, under your existing security posture, with access to your existing documents, and that proximity is worth more than marginal capability differences for most operational work. Check the per-seat cost against the value, but make this your default starting point.
If your team already lives in ChatGPT, start with Codex. It extends a tool your team already knows into agentic work with no new platform to learn, and the data analytics and creative production plugins cover the most common small-business needs. Begin on your current plan and watch the consumption.
If accuracy is paramount or you already use Claude, start with Claude Cowork. When your agentic work is analysis-heavy or a wrong answer is expensive, the reasoning and accuracy advantages justify the more deliberate adoption, and if you are already in the Claude ecosystem the fit is natural. And if you are starting from a blank slate with no entrenched stack, weigh accuracy needs against convenience and pick the one whose strengths match your most valuable use case, rather than agonising over leaderboard differences that will not show up in your actual work.
Getting value without wasting a month
Whichever you pick, the way to get value is the same, and it is the same discipline that separates winning AI deployments from failed ones generally. Choose one high-value, multi-step task that currently eats real time, hand it to the agent, integrate it into your actual workflow, and measure the result for a month before expanding.
Do not try to agentify your whole business in week one. The teams that get value pick a single recurring task, a monthly report, a campaign workflow, a data analysis you do repeatedly, and get the agent genuinely good at that one thing before adding a second. The teams that waste a month try everything at once, never get any single workflow working well, and conclude the technology does not work when really the approach did not.
And keep a human in the loop on anything that leaves the building or carries risk. These agents do real work autonomously, which means an error can propagate through several steps before you see it. The agent does the work; you own the result and you check it before it reaches a customer or a decision. That division is what lets you capture the speed of autonomous work without inheriting the risk of unsupervised mistakes, and it holds regardless of which of the three products you chose.
The verdict for small teams
The verdict is that the choice matters less than the move. Copilot Cowork, ChatGPT Codex, and Claude Cowork are all credible agentic work products, and for most small businesses the right one is simply the one that fits the stack you already run, with a tilt toward Claude Cowork when accuracy is paramount and toward Copilot Cowork or Codex when ecosystem convenience dominates. You will not go badly wrong with any of them if you pick on fit and use it with discipline.
What you would go wrong on is sitting it out. The convergence of all three giants on autonomous work agents in a single month of 2026 is the clearest signal available that this is where business software is heading, and the businesses that learn to delegate real multi-step work to an agent now will have a year of practice when it becomes the baseline expectation. Pick the one that fits, start with a single task, measure honestly, and keep your hand on the wheel. The specific logo on the agent matters far less than the fact that you started learning to work with one while your competitors were still deciding whether to.
Sources
- Anthropic — Claude Cowork, agentic AI for knowledge work
- Microsoft 365 — What's New in Microsoft 365 Copilot, June 2026 (Copilot Cowork GA)
- Microsoft 365 Blog — Introducing Microsoft 365 Business with Copilot: the new standard for small business
- 9to5Mac — OpenAI putting Codex inside ChatGPT, 6 new business plugins now available
- DigitalApplied — OpenAI Puts Codex in ChatGPT With 6 Business Plugins
- MindStudio — Code with Claude 2026: 5 New Agent Features Anthropic Just Shipped
- Strategic Micro Systems — Microsoft 365 Copilot Business: What Changes on July 1, 2026
- InfoQ — Anthropic's Code with Claude Announces Managed Agents and Proactive Workflows