Last week I was on a train, scrolling through my phone, when I told Claude to scan my inbox, pull out everything urgent, and prep a summary before I got to the office. By the time I sat down at my desk, it was done.
That is not a pitch. That is Tuesday morning.
This week, three things happened almost simultaneously. Anthropic launched computer use in Claude, which means the AI can now click, scroll, open apps, and fill forms on your Mac. OpenAI announced it is merging ChatGPT, its browser, and its coding tool into a single desktop app. And Google leaked a Mac app called Gemini Desktop Intelligence that watches your screen and pulls context from whatever you are working on.
Meanwhile, an open-source project called OpenClaw hit 250,000 GitHub stars, Jensen Huang called it the operating system for personal AI, and Tencent integrated it directly into WeChat for over a billion users in China.
The AI industry just made a collective decision: the future is not a chat window. It is an assistant that lives inside your computer and actually does things.
For recruitment, this changes everything. Not someday. Now.
What actually happened this week
Claude can now control your Mac. Through Cowork (Anthropic's desktop tool), Claude can navigate your browser, open applications, fill spreadsheets, manage files, and complete multi-step tasks across different programs. It works by taking screenshots of your desktop and executing mouse and keyboard actions based on what it sees.
The system is smart about how it operates. When a direct connector exists (Gmail, Calendar, Slack), it uses that first. When there is no connector, it falls back to browser control. Mouse-and-keyboard manipulation is the last resort. That layered approach matters because structured API calls are faster and less error-prone than visual screen navigation.
Dispatch lets you assign tasks remotely. You message Claude from your phone, and it completes the work on your Mac while you are away. Set up recurring tasks: a morning inbox briefing, a Friday metrics roundup from a spreadsheet, a daily check on a project pipeline. It runs in the background on your own computer.
OpenAI is consolidating hard. Fidji Simo, their head of applications, told staff the company was distracted by side quests and needed to focus on productivity. They killed Sora (their video generator), walked away from a billion-dollar Disney deal, and are now building one app to do everything. The reason? Claude overtook ChatGPT as the most-downloaded app in the US this month, and Anthropic captures roughly 70% of first-time enterprise AI spending.
Google is testing quietly. A standalone Gemini app for Mac, codenamed Janus, includes a feature called Desktop Intelligence that can read what is on your screen and pull data from open applications. It is not yet clear whether it can take actions, but the direction is obvious.
The OpenClaw lesson: the vision matters, the tool does not
Here is the thing about OpenClaw. It does not matter as a product. What matters is what it revealed: people want one AI assistant that connects to their tools and gets things done.
OpenClaw was built by a single developer using shortcuts that would make any enterprise security team faint. Personal data in plain text files. The AI connecting to tools on its own. Code execution with minimal guardrails. And 20% of the plugins on its marketplace were found to contain malicious code.
But the demand is real. People do not want five different AI tools open at once. OpenClaw might become the Linux of agentic AI, while Anthropic, OpenAI, Microsoft, and Apple build the commercial versions of that same world.
The question for recruiters is not should I install OpenClaw? It is: what does this shift mean for how I work?
The part nobody wants to talk about
Desktop agents have another problem that matters specifically in recruitment: compliance.
When a general-purpose agent fills out your ATS by controlling the screen (instead of through auditable API calls), traceability breaks down. You cannot easily log what it did or why. You cannot prove it did not introduce bias. And in recruitment, that is not a theoretical risk.
The EU AI Act classifies employment AI as High Risk. Colorado's new AI Act (effective June 2026) requires impact assessments for AI in hiring. Illinois already bans AI use that results in bias against protected classes. Courts are testing whether AI hiring tools carry anti-discrimination liability.
Anthropic itself warns: do not use Cowork for regulated workloads. That is worth taking seriously.
This is exactly why purpose-built automations are the better path for recruitment right now. When you build a sourcing workflow in n8n or connect Claude to your tools through MCP, every step is logged. You control what data goes where. You can explain the process to a regulator. Try doing that with a desktop agent that navigated your screen by taking screenshots.
What to do right now
Desktop agents will get better. In a year or two, Claude controlling your screen might be as reliable as a direct API call. But right now, the technology is early, fragile, and not built for the compliance requirements of recruitment.
What does work right now is learning to build your own automations. Not coding from scratch, but connecting the tools you already use through structured workflows that run reliably, log everything, and save you hours every week.
That is what I teach recruitment teams at Klikwork. Not how to let AI click around your screen and hope for the best. How to build sourcing pipelines, outreach sequences, data enrichment flows, and reporting dashboards that actually work. With tools like n8n, MCP connectors, and Claude.
The recruiters who will come out ahead are not the ones chasing every new AI launch. They are the ones who understand automation well enough to know which tools to use, when to use them, and where to keep a human in the loop.
The lobster hype will pass. The ability to build your own automations will not.
