{"id":1863,"date":"2026-10-20T09:00:00","date_gmt":"2026-10-20T07:00:00","guid":{"rendered":"https:\/\/pletzenauer.com\/?p=1863"},"modified":"2026-07-27T17:48:55","modified_gmt":"2026-07-27T15:48:55","slug":"cowork-vs-chatgpt-work-en","status":"publish","type":"post","link":"https:\/\/pletzenauer.com\/en\/2026\/10\/20\/cowork-vs-chatgpt-work-en\/","title":{"rendered":"Claude Cowork vs ChatGPT Work: agentic work compared"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><strong>The short answer:<\/strong> both pursue the same idea \u2014 the AI should stop answering and start working. It receives a goal, gathers context, makes a plan, completes the task and checks in when it needs a decision. The difference lies in the cut: Claude Cowork is built for several people working together, ChatGPT Work for connecting to the tools your team already sits in. For most businesses the honest answer is: take whichever is included in the plan you already have.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><em>Last checked: October 2026. This is the article in the series that goes stale fastest \u2014 both products are young and change in months, not years. Check the date above before relying on any detail.<\/em><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Disclosure:<\/strong> I earn my living advising on and training people in Claude. On a topic this young that weighs more heavily than usual, because nobody yet has extensive comparative experience \u2014 myself included. So I deliberately write more about the structure of the decision than about feature detail.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The short version<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Both are agentic working environments:<\/strong> goal in, finished result out, questions in between.<\/li>\n<li><strong>Claude Cowork<\/strong> has been generally available since June 2026 on macOS and Windows via the Claude Desktop app, with its own usage analytics for teams.<\/li>\n<li><strong>ChatGPT Work<\/strong> is included in the ChatGPT plans, with full access on desktop, web and mobile on the Business plan.<\/li>\n<li><strong>Both are included in plans<\/strong> you probably already pay for \u2014 no separate purchase needed.<\/li>\n<li><strong>The critical point is not the choice<\/strong> but the question of which permissions you grant an agent.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">What &#8220;agentic work&#8221; actually means<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">The term is marketing; the thing behind it is not. The difference from ordinary chat can be pinned to one sentence.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Chat:<\/strong> &#8220;Write me a draft rejection for M\u00fcller GmbH.&#8221; You get a text, copy it, adjust it, send it.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Agentic:<\/strong> &#8220;Go through the open proposals, find the ones older than six weeks with no response, and prepare a follow-up email for each. Show me the list before you send anything.&#8221;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The second instruction spans multiple steps, needs access to your data, and produces not a text but an outcome. That is where the value and the risk both come from \u2014 and why the permissions question matters more with these tools than any feature comparison.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Claude Cowork<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Status:<\/strong> generally available since late June 2026 on macOS and Windows via the Claude Desktop app. Anthropic simultaneously added support to the Analytics API and introduced dedicated usage analytics for Cowork sessions.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>The visible emphasis:<\/strong> multi-user, shared agent work rather than single-user chat. That analytics for teams arrived alongside general availability is a clear signal about who the product is aimed at \u2014 organisations that want to know whether and how the thing is actually being used.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Fits you if:<\/strong> you already have Claude in house, work with the desktop app, and want several people working on the same matters. I have covered setting up Claude Desktop in a separate series, starting with <a href=\"https:\/\/pletzenauer.com\/en\/2026\/06\/10\/installing-claude-desktop\/\">installation<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">ChatGPT Work<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Status:<\/strong> part of the ChatGPT plans. OpenAI describes it as follows: you give it a goal, the system gathers context, makes a plan, completes the work and checks in when your input or approval is needed. On the free plan access is limited to the desktop app and a restricted scope, expanded on Plus, and fully available on desktop, web and mobile on Business.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>The visible emphasis:<\/strong> connectivity. The Business plan adds connections to Microsoft 365, Google Drive, Slack, GitHub, Linear and Figma. If you work in precisely that landscape, you get an agent that reaches the relevant data without extra work.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Fits you if:<\/strong> you already have ChatGPT in house and your work happens in those tools.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The decision has usually already been made<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">This is the unspectacular but honest point of the article: neither product is sold separately. Cowork belongs to Claude, Work belongs to ChatGPT. If you have chosen an assistant \u2014 which is what the <a href=\"https:\/\/pletzenauer.com\/en\/2026\/08\/04\/claude-vs-chatgpt-for-business\/\">main comparison<\/a> is for \u2014 you have thereby also chosen the agentic environment.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">From which follows a practical recommendation: <strong>do not make the agentic feature the criterion for choosing the assistant.<\/strong> Both products are young, both change fast, and both will look considerably different in twelve months. You decide on an assistant using the criteria in the <a href=\"https:\/\/pletzenauer.com\/en\/2026\/07\/30\/how-i-evaluate-ai-tools\/\">evaluation framework<\/a> \u2014 data protection, workload, cost, adoptability. The agentic feature is a bonus you should try as soon as it appears.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The point that genuinely needs attention<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Not the choice, but the permissions. An agent working independently needs access \u2014 to files, to the mailbox, to connected systems. That access is exactly the value and exactly the attack surface.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">How real that is was shown in July 2026 by a demonstration from security researchers who induced unwanted behaviour in Claude Desktop through attacks on tool and MCP integrations. That is no reason to avoid these tools \u2014 it is a reason to choose the connections deliberately. Concretely:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Check which connections are active<\/strong> and switch off what you do not need. The default should be &#8220;nothing connected&#8221;, not &#8220;everything connected&#8221;.<\/li>\n<li><strong>Separate reading from writing.<\/strong> An agent allowed to read your mailbox does not also have to be allowed to send.<\/li>\n<li><strong>Insist on approval before anything leaves.<\/strong> Anything going outside the building \u2014 emails, orders, publications \u2014 belongs behind a human approval step.<\/li>\n<li><strong>Treat content from external sources as data, not instructions.<\/strong> That is exactly where the known attacks come in.<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">How to start sensibly<\/h2>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>Find a process<\/strong> that recurs regularly, has several steps, and whose result is easy to check.<\/li>\n<li><strong>Grant only the permissions<\/strong> that this one process needs.<\/li>\n<li><strong>Have the first run checked completely<\/strong> before anything goes out automatically.<\/li>\n<li><strong>Assess after four weeks<\/strong> whether it saves time or whether checking takes longer than doing.<\/li>\n<\/ol>\n\n\n\n<p class=\"wp-block-paragraph\">Point 4 is the honest part. For some processes, checking the result takes exactly as long as doing it \u2014 then it is not worth it, and that is a perfectly acceptable outcome of an experiment.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Frequently asked questions<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">Does this replace our automation with n8n or Zapier?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">No, these are different classes of tool. An automation runs without a person, identically every time, and is auditable precisely because of that. An agent decides afresh on every run. For anything that recurs identically, an automation is the more robust tool \u2014 see <a href=\"https:\/\/pletzenauer.com\/en\/2026\/10\/06\/n8n-make-zapier-compared\/\">n8n, Make or Zapier?<\/a><\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Do we need a particular plan for this?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">The full scope sits in the higher or business plans respectively. For business use you need the business plan anyway \u2014 for data protection reasons, independent of the agentic feature. Details in the <a href=\"https:\/\/pletzenauer.com\/en\/2026\/09\/08\/ai-gdpr-vendor-comparison\/\">GDPR comparison<\/a>.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">How well does it actually work?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">On well-bounded tasks with checkable results, surprisingly well. On vague instructions in untidy environments, poorly \u2014 the same rule as for all AI tools. Anyone quoting percentages without solid experience is guessing.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Should we wait?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">With trying it, no \u2014 it is included in your plan and costs you only time. With restructuring workflows, yes. Do not yet rebuild a process so that it no longer functions without the agent.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">What if we use Copilot or Gemini?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Both vendors are heading in the same direction with their own approaches. The structure of the decision stays identical: you get your assistant&#8217;s agentic environment, and the permissions question arises in exactly the same way.<\/p>\n\n","protected":false},"excerpt":{"rendered":"<p>Goal in, result out. Both are included in plans you already pay for \u2014 the critical point is not the features but the permissions.<\/p>\n","protected":false},"author":1,"featured_media":2007,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[17],"tags":[],"class_list":["post-1863","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-automatisierung"],"_links":{"self":[{"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/posts\/1863","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/comments?post=1863"}],"version-history":[{"count":1,"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/posts\/1863\/revisions"}],"predecessor-version":[{"id":1972,"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/posts\/1863\/revisions\/1972"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/media\/2007"}],"wp:attachment":[{"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/media?parent=1863"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/categories?post=1863"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/pletzenauer.com\/en\/wp-json\/wp\/v2\/tags?post=1863"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}