Something shifted in the last few weeks. If your feeds have been full of people announcing they're canceling ChatGPT and switching to Claude, you're not imagining it.
Here's what happened, and more importantly, what it means for your agency.
The moment that changed everything
The surge didn't come from a single product launch. It came from two back-to-back events that put Anthropic, the company behind Claude, in front of a mainstream audience for the first time.
In early February, Anthropic ran a Super Bowl ad that directly took aim at OpenAI. Claude jumped from around number 42 in the App Store to the top 10 overnight and stayed there.
Then in late February, Anthropic refused a Pentagon contract that would have allowed Claude to be used for mass domestic surveillance and autonomous weapons. The Department of Defense responded by labeling Anthropic a supply chain risk. Hours later, OpenAI signed its own deal with the Pentagon. The backlash was immediate. ChatGPT uninstalls spiked 295% in a single day. Claude hit number one in the US App Store. A movement called QuitGPT reportedly saw over 1.5 million people cancel their ChatGPT subscriptions in a short period.
To be precise about what this means: ChatGPT still has a significant lead in total users. Claude hasn't overtaken it. But the momentum shift is real, daily sign-ups are at record highs, and Claude is landing in front of people who had never heard of Anthropic a month ago.
What's actually new with Claude
The cultural moment is one thing. The product reality is another, and it's worth understanding both.
In early February, Anthropic released Opus 4.6, currently the most capable model in the Claude family. As of late February it holds the longest autonomous task-completion time horizon of any AI model independently measured, meaning it can run complex, multi-step tasks for longer without losing coherence or going off the rails. That's a meaningful benchmark for anyone thinking about agents.
But the more relevant development for agency owners isn't the model. It's what Anthropic has built around it.
What Cowork actually is, and why it matters
Cowork is Anthropic's answer to a real problem: Claude Code, their most powerful agentic tool, required comfort with a terminal and a developer's workflow. Most people running agencies aren't developers. Cowork brings those same capabilities into a graphical interface on your desktop, without any coding required.
The simplest way to think about it: rather than responding to your prompts one at a time, Claude in Cowork takes on multi-step tasks and executes them on your behalf. You describe an outcome, step away, and come back to finished work.
Here's what that actually looks like in practice.
Connectors. Cowork connects to your external tools through something called MCP, which stands for Model Context Protocol. Think of MCP as a universal connector standard that allows Claude to not just read information from your tools, but take actions inside them. Right now that includes Gmail, Google Drive, Slack, Notion, Google Calendar, HubSpot, and GitHub, with the connector directory now sitting at 38 and growing. Claude can review your inbox, update a project in your project management system, respond to Slack messages, or check your calendar, all from a single conversation, by typing or dictating.
Skills. Skills are reusable instruction documents you build inside Cowork. These are essentially your SOPs and style guides in a format Claude can load and follow. A skill might be your email response style, your client reporting format, or a checklist for how you onboard new accounts. You build it once and Claude follows it every time. You can ask Claude to help you build the skill, which means the barrier to getting started is low.
Plugins. Once you have skills, you can bundle them together with connectors and specific commands into plugins. Each plugin is configured for a particular function, think of it as a purpose-built assistant for a specific job. Cowork includes a built-in tool called Plugin Create that walks you through building custom plugins from scratch, no technical knowledge required.
Scheduled tasks. This is where Cowork starts to feel genuinely agentic. You can schedule Claude to run tasks automatically: check email at 7 a.m. and send a Slack summary, pull weekly performance data and format it into a report, flag any client deliverables due in the next 48 hours. One practical note: scheduled tasks only run while your computer is awake and the desktop app is open, so this isn't fully autonomous background execution yet.
Workflows. At the top of the stack are workflows, automated sequences that chain multiple skills and plugins together into a repeatable process. This is where Cowork starts to look less like a tool and more like a junior operator running parts of your business on a schedule.
Cowork is currently available on Mac and Windows for paid Claude users. It is not available on mobile or web.
What this means for your agency right now
The honest framing: Cowork is early. It's powerful in narrow, well-defined workflows. It's not yet reliable enough to hand off anything client-facing without review.
But that's exactly the point made in last week's issue. The agents worth deploying right now are the ones that operate internally, where failure is recoverable and output is easy to check. Cowork is built for precisely that use case.
The practical move this week: if you're not already on a paid Claude plan, Pro starts at $20 a month. Once you're in, open Cowork and point it at one internal process. Not a client deliverable. Something like summarizing your weekly emails, organizing a folder, or pulling together a briefing before a sales call. Run it. See where it breaks. That experience is worth more than any amount of reading about what agents can theoretically do.