The potential for AI agents
Rakuten, a major Japanese technology conglomerate, is actively integrating AI into its 70+ businesses. In the process, it has become a significant, practical testing ground of high-performance Large Language Models (LLMs), particularly for coding and agentic workflows.
Recently, it asked Opus 4.6, the newest Claude model from Anthropic, to manage and disposition issues across a 50-person team. In a single day, it autonomously closed 13 issues and routed 12 more to the right team members, a real-world example of agentic AI handling complex work.
This is just one example of the potential impact when AI does the actual work.
On January 12, Anthropic launched Claude Cowork, a desktop agent for knowledge workers. Within days, it released plugins for specific skills, showing the world more practical examples of agents performing actual work. This included a legal plugin that does contract reviews, checks vendor agreements, generates briefings and more.
Later in the month, OpenClaw hit the news. This is an open-source, always-on agent that can communicate with external systems, complete tasks, and repeat the process if needed. It has thousands of skills (researching, preparing briefings, cleaning inboxes) and its capabilities are expanding. Success still depends on quality instructions, but when they're good, OpenClaw can demonstrate remarkable problem-solving. Here are a few examples:
- The "auto-negotiated" car purchase: Someone linked OpenClaw to his email and financial accounts and instructed it to find a specific car model and negotiate. OpenClaw then reportedly scoured listings and initiated email contact with dealers, negotiated back and forth on price, and secured a $4,200 discount all while the human was in meetings.
- The restaurant booking: In another instance, OpenClaw was asked to book a table at a popular restaurant that didn't have a direct website booking link. The agent first tried to use OpenTable but found no availability so it downloaded text-to-speech software, synthesized a voice, called the restaurant, and spoke to a human host. It secured the reservation and then messaged the user on WhatsApp with the confirmation details and an invite.
Then, in February, OpenAI launched Frontier, an enterprise platform for deploying and managing AI agents across business systems. It’s designed to get different agents working as teammates with shared context, processes, feedback loops, and clear permissions and boundaries.
Different approaches
Each of these 3 options—Claude Cowork, OpenClaw, OpenAI Frontier—demonstrates a different agentic approach:
- Cowork is bottom-up: It operates on the desktop, providing a well-designed UI that lets teams leverage the intelligence of Claude models to delegate work to agents.
- Frontier is top-down: A platform that sits across the enterprise, bringing together different systems and data, making it easier for companies to build, deploy, and manage agents.
- OpenClaw is uncompromising: A self-hosted, self-directed team of agents built for specific work without platform constraints. Of the three, it's probably the furthest from enterprise readiness, but it signals where things are heading with agents operating like workers, not tools.
Together, they all point to a clear trend: People want more than conversational AI. They want AI that acts by coordinating systems, agents, and even people to get work done.
Confronting inevitability
AI will slowly begin to demonstrate the skills and problem-solving capabilities of knowledge workers, able to handle multi-step tasks and achieve objectives without constant instruction. To confront this seemingly inevitable future, companies must continue to:
- Invest in AI fluency. People must learn how to manage AI, to understand when and how to delegate, supervise, and extract value from it, and to lead a hybrid human-AI workforce.
- Experiment. Test and enable new interactions, operating across a continuum from human-only work, workflow automation, AI augmentation to autonomous agents.
- Build the governance to move fast safely. Put in place governance structures, including risk frameworks, change advisory processes, and supervision standards. These will allow companies to accelerate AI transformation while minimizing risk.
The pace of change reinforces the need to stay vigilant. Organizations must invest in people and support leaders through education, tools and resources. These steps will help surface the right opportunities to utilize AI and, in the long term, help to manage a workforce where some of the work is being performed by AI agents and managed with new agentic tools.
