But coding was just the beginning. Advanced multi-agent tools are aimed at people who don’t need or want software. Desktop apps like Anthropic’s Claude Cowork (which the firm claims was built in just 10 days using Claude Code, rather than the months such projects would have taken), OpenAI’s Codex, and Perplexity’s Computer are all designed as general-purpose productivity tools for white-collar professionals. They let you assign predefined workflows to teams of agents that integrate into a wide range of computer-based office tasks, from managing inboxes and inventory to handling customer complaints.
And it’s not just office work. Multi-agent tools like Google DeepMind’s Co-Scientist allow researchers to coordinate teams of AI agents to search the literature, generate and test hypotheses, design experiments, and more.
Think of multi-agent systems as new assembly lines. Henry Ford’s innovation disrupted entire industries in the last century. In theory, networks of AI agents could do for white-collar knowledge what assembly lines did for manufacturing.
At least that’s the vision. Because this technology also comes with great risks. It’s no secret that LLMs can be unpredictable. It’s a problem when chatbots get stuck inside their screen. When they start interacting more with the real world, it can be disastrous. Are we ready for agents to abandon our ubiquitous digital infrastructure, from healthcare to finance, from social media to missile launchers?