What is Claude Cowork?

Claude Cowork is Anthropic's autonomous AI agent platform for enterprise knowledge work. It runs multi-step tasks in the background on Anthropic's cloud infrastructure, even when no user device is online. Tasks can be scheduled across any cadence, cross-device sessions continue seamlessly, and the agent asks for human approval before irreversible actions. Available on Pro ($20/month), Max 5x ($100/month), Max 20x ($200/month), Team, and Enterprise plans at claude.com.

What is ChatGPT Work?

ChatGPT Work is OpenAI's enterprise AI workspace agent, launched July 9, 2026, and powered by GPT-5.6. It researches and synthesizes information across connected apps and files, creates documents, slides, sheets, and Sites, and runs scheduled or triggered tasks without a live session. Workspace agents can be created, tested, shared, and run inside ChatGPT and Slack for Business and Enterprise accounts.

How does Claude Cowork differ from ChatGPT Work?

Both platforms execute multi-step agentic tasks in the background, but they differ in their primary integration surface, pricing model, and enterprise control layer. Claude Cowork emphasizes persistent cross-device continuity and detailed audit logging with run-by-run compliance records. ChatGPT Work emphasizes Slack-native workspace agents that plug into existing channel workflows. Both require human approval before consequential actions.

Which enterprises should prioritize Claude Cowork versus ChatGPT Work?

Teams in compliance-sensitive environments, or those already standardized on Claude for sensitive workflows, benefit most from Cowork's granular audit trails and role-based spend controls. Teams running Slack-heavy operations and already on ChatGPT Enterprise get the most from ChatGPT Work's native Slack agents. Most organizations will pilot both and route task types based on cost-per-output and their existing tool stack.

What are the security risks of background AI agents?

The primary risks are prompt injection via external data and unauthorized scope creep across connected systems. Both Anthropic and OpenAI address these through sandboxed browsing, permission scoping on connectors, and human-in-the-loop gates before consequential actions. Enterprise admins on both platforms can restrict which connectors each agent accesses and set spending caps to limit blast radius. Neither platform has fully solved adversarial data in context, which remains an active area of research.

Claude Cowork vs ChatGPT Work: Enterprise AI Workspace Race

On July 9, 2026, two of the most consequential enterprise AI products in recent memory launched on the same day. Anthropic rolled out Claude Cowork to web and mobile; OpenAI released ChatGPT Work alongside GPT-5.6. Both represent the same fundamental bet: the next frontier in enterprise AI is not better chat responses, but AI that ships finished work while you sleep.

For enterprise teams evaluating AI platforms right now, the simultaneous release changes the calculus. This is no longer a question of which model scores higher on benchmarks. It is a question of which autonomous work layer your organization will build processes around for the next three to five years.

What Launched This Week

Claude Cowork: Background Agents Across Every Device

Anthropic announced on July 7, 2026 that Claude Cowork was expanding from desktop to web (claude.ai), iOS, and Android. The upgrade introduced cloud execution: tasks now run on Anthropic’s infrastructure, decoupled from the device that started them.

The practical shift is significant. A Monday morning client brief scheduled for 6 a.m. runs before the user wakes up. When Claude reaches a decision only a human can make, it routes a notification to the user’s phone. When it finishes, the deliverable is waiting. Shutting a laptop no longer stops the agent.

Enterprise controls launched alongside the product expansion: role-based access controls, per-team spend caps, admin analytics, and full audit logs that record every file touchpoint for compliance review. Human-in-the-loop rules force explicit approval before Cowork sends emails or deletes records. TechCrunch noted that more than 90% of Claude Cowork usage is already not software development. Business operations accounts for 33.4% of sessions; content creation and copywriting follows at 16.4%. The product is a general-purpose enterprise work platform by usage, not a coding tool that expanded.

Pricing spans Pro ($20/month), Max 5x ($100/month), Max 20x ($200/month), Team ($20 to $125 per seat depending on tier), and custom Enterprise contracts. Anthropic extended doubled usage limits through August 5, 2026 to encourage early adoption at scale.

ChatGPT Work: Workspace Agents That Ship Documents

OpenAI launched ChatGPT Work on July 9, 2026, powered by GPT-5.6 Terra for Business subscribers and the higher-capability Sol model for Pro and Enterprise users.

Bloomberg described ChatGPT Work as an agent for longer, more involved tasks that can research and analyze information, work across connected apps and files, and create finished documents, spreadsheets, presentations, reports, and Sites. Scheduled tasks run once, repeat on a cadence, trigger on conditions, or monitor for changes and then act.

The Slack integration is the enterprise unlock. Workspace agents can be created, tested, shared, and run inside ChatGPT and in Slack for Business and Enterprise workspaces. For organizations where Slack is the primary coordination layer, agents can be woven directly into existing channels without adding a new interface. Token-based pricing for workspace agent runs began July 6, billed on input, cached input, and output tokens. Business plan pricing sits at $20 per seat per month; enterprise contracts average approximately $60 per user per month with a 150-seat minimum on annual terms.

Side-by-Side: Claude Cowork vs ChatGPT Work

Dimension	Claude Cowork	ChatGPT Work
Launch date	July 7-9, 2026	July 9, 2026
Underlying model	Claude family	GPT-5.6 (Sol / Terra / Luna)
Background execution	Cloud-hosted, device-free	Scheduled tasks and triggers
Primary interface	claude.ai, iOS, Android	ChatGPT web plus Slack
Enterprise controls	RBAC, spend caps, audit logs	Workspace admin, token billing
Human-in-the-loop	Required before consequential actions	Pauses for confirmation
Entry price	$20/month (Pro plan)	$20/seat/month (Business)
Rollout order	Max plan first, then others	Pro and Enterprise first, then Business
Usage breakdown	33% ops, 16% content, less than 10% code	Not yet disclosed

Why the Same-Week Launch Signals a Market Shift

The simultaneous release is not a coincidence in the way overlapping product cycles often are. Both companies signaled for months that autonomous, persistent enterprise agents were coming. The compressed timeline reflects a deliberate race for the enterprise workflow layer.

For enterprise buyers, competition between Anthropic and OpenAI on the workspace means feature velocity will be high, pricing will stay competitive, and integrations with existing tools will expand quickly. The risk is premature lock-in: the agent that handles client prep, deal research, and weekly ops summaries becomes a workflow dependency faster than most procurement cycles anticipate.

The enterprise AI adoption gap documented earlier this year was largely attributed to a lack of production-grade infrastructure, unclear compliance paths, and thin enterprise controls. Both Claude Cowork and ChatGPT Work directly address those objections with RBAC, spend caps, and structured audit trails. The residual risk is prompt injection: both platforms connect agents to live files and external sources, and sandboxed browsing remains an incomplete solution when agents operate across internal knowledge bases at scale.

Claude Sonnet 5, released in July, raised the capability floor for what agentic models can reliably execute. Cowork and ChatGPT Work are the product layers that put that capability in front of knowledge workers who have no interest in configuring an API.

What This Means for GTM and Ops Teams

Both platforms launched in the same 48-hour window because both companies reached the same conclusion: the revenue unlock in enterprise AI is not better answers, it is better execution.

For GTM teams, the clearest near-term patterns across both platforms are pre-meeting research (builds briefing docs from CRM notes, email threads, and news before calls), competitive monitoring (scheduled tasks check competitor pages and flag updates), and proposal drafts (synthesizes requirements from intake sessions, prior decks, and legal templates into a structured first draft).

For ops teams, the highest-leverage patterns are exception routing (an agent monitors a process, escalates anomalies, and continues without a live session), report generation (weekly and monthly summaries pulled from connected tools and formatted automatically), and structured onboarding sequences (task chains that run across systems on a new hire’s first week without manual shepherding).

These are not theoretical. Claude Cowork’s usage distribution, already 33.4% business operations and 16.4% content creation at scale, reflects exactly these patterns across enterprise and team subscribers today.

As AI-native enterprises have started to separate from AI-aware incumbents, the organizations building production processes around agents, rather than treating AI as a prompt layer on top of manual workflows, are the ones compounding advantage. Cowork and ChatGPT Work make the production-ready agent layer accessible to organizations that cannot build it themselves.

Anthropic and OpenAI have both decided that the enterprise workspace is their next large market. The week of July 7, 2026 is when that bet became a product.