The AI agent space has become crowded fast. Every major AI lab has announced agentic products, open-source frameworks multiply weekly, and startups are raising on agent-focused pitches. Here is a clear-eyed overview of who is building what, what the meaningful differences are, and where the category seems to be heading.
OpenAI: Operator and GPT Actions
OpenAI's Operator is a browser-based agent that can navigate websites, fill forms, and complete web tasks on your behalf. It runs in the cloud and communicates with websites on your behalf. The capability is real — it handles travel booking, form submission, and structured web workflows with reasonable reliability. The limitation is that it operates within the browser context and requires OpenAI infrastructure, raising the standard cloud privacy questions. GPT Actions (formerly plugins) allow GPT-4 to connect to external APIs, though with less autonomy than a full agent.
Anthropic: Claude Desktop and Claude Projects
Anthropic has taken a more cautious agentic approach, consistent with their safety-focused research culture. Claude Desktop (available through the Claude.ai subscription) gives Claude access to your local file system, making it genuinely useful for document analysis and local search. Claude Projects maintains context across conversations within a defined project. The computer-use capability (using the screen like a human) is technically impressive but positioned as a research preview rather than a production feature. Privacy-wise, all processing goes through Anthropic's cloud.
OpenClaw and Open-Source Frameworks
The open-source ecosystem includes several mature frameworks: LangGraph for building stateful multi-agent systems, CrewAI for multi-agent collaborative workflows, AutoGPT as one of the original autonomous agent experiments, and Open Interpreter for a code-execution-first approach. These frameworks are powerful and highly customisable — but require developer expertise to deploy. They are infrastructure for building agents, not finished products for end users. See how Skales compares to OpenClaw specifically.
Microsoft: Copilot and Copilot Studio
Microsoft has integrated AI deeply into its Office suite via Copilot, with agentic features in Copilot Studio for building custom enterprise agents. The Copilot integration in Word, Excel, and Outlook is genuinely useful for Microsoft 365 users. The enterprise angle is clear — Copilot is priced and positioned for organisations, not individuals. Privacy is governed by Microsoft's enterprise data processing agreements, which are more robust than consumer equivalents but still cloud-based.
Google: Project Mariner and Gemini Agents
Google's Project Mariner is a research prototype that can navigate the web using Chrome, similar in concept to OpenAI's Operator. Gemini-based agentic features are being integrated into Google Workspace (Gmail, Calendar, Docs). The capability is strong, the integration with Google services is natural, and the privacy implications are exactly what you would expect from Google — your data contributes to improving their products.
Local-First Desktop Agents
A smaller but distinct category is desktop agents designed for privacy-first, local operation. Skales is one of the few finished products in this space — a native desktop app that runs its AI processing locally (via Ollama or other local providers), stores all data on your machine, and provides agentic capabilities across email, calendar, files, browser, and more without requiring cloud infrastructure.
Where the Category Is Heading
Three trends seem durable. First: computer-use capabilities (agents that can see and interact with any interface) will mature from research preview to production feature within 12-18 months. Second: local model quality will continue closing the gap with frontier cloud models, making privacy-first local agents increasingly competitive on capability. Third: the enterprise market will consolidate around a small number of cloud providers (Microsoft, Google, OpenAI), while the individual/privacy-conscious market remains underserved and interesting for independent products.
For a deeper comparison of Skales against specific alternatives: vs. ChatGPT and vs. OpenClaw.