Fork in the Code: Designing AI for Control or Independence?

Artificial intelligence systems are becoming more capable of handling tasks previously reserved for human workers, but there is increasing debate about how these systems should be designed and deployed. While much recent development has focused on models that act independently to complete tasks, there is also a strong case for systems that remain closely involved with human users in collaborative or communicative contexts.

Anthropic’s Claude Code, part of the Claude Opus 4.5 release, has gained widespread attention for its ability to handle complex tasks with little human input. In contrast, newly launched startup Humans& is focusing on building systems that maintain ongoing collaboration with users, particularly in group and workplace communication.

Claude Code’s viral success shows growing acceptance of autonomous AI models in real-world work, with users claiming drastic productivity improvements across technical and non-technical tasks.

Meanwhile, Humans& is positioning its product around active user engagement, contextual memory, and support within team environments.

Why It Matters: The way AI tools are being designed today will shape how people relate to software tomorrow. Whether systems stay responsive to human input or evolve toward independent execution will influence decisions about how work is organized, who remains involved, and what kinds of tools users come to expect in everyday environments.

• Conversation-Based Integration: Humans& is developing artificial intelligence that functions within communication platforms. Rather than executing commands in isolation, the system interacts with users through dialogue, asking for clarification and helping teams keep track of shared information. This interaction is intended to support coordination and reduce potential points of friction in group tasks. Through their integration, Humans& is focusing on tools that respond to human decision-making instead of bypassing it.

• Autonomous Execution in Claude Code: Claude Code is built so that artificial intelligence can carry out workflows independently once a clear objective is set. Many users have described turning over projects to Claude, then watching the system complete them with little additional input. The tool operates through a desktop interface and can access user files, browse the internet, and interact with various applications, often stringing together actions across different systems.

• Memory and Workflow Structure: Humans& places a strong emphasis on memory and continuity. Its AI tools are being trained to store information from past interactions and use it in future responses. This design allows the system to support long-term projects without needing repeated instructions. The AI acts more like a persistent collaborator than a one-time assistant. Claude Code, in contrast, treats most tasks as discrete problems to solve. Each prompt generates a new output based on the context provided in that moment. While powerful, this approach can lead to gaps in continuity when users want the system to follow a line of thinking or retain information over time.

• Environment and Access: Claude Code operates in a system that connects the AI to multiple digital environments. Users report using it to manage local files and automate tasks that cross between platforms, acting as an intermediary between tools to take action without requiring the user. Humans& is designing its product for environments where human communication already happens, such as team chats or project management tools. This difference affects how users experience the tool and how they need to adjust their routines to incorporate it.

• Impact on Work Design: Claude Code has led some users to postpone or cancel hiring plans, saying the system allows them to do the work of multiple employees. Others describe a sense of dislocation, realizing that tasks they once saw as specialized can now be completed by software. Humans&’s promotion of AI remaining as a part of the communication process is being supported by major investors, including Nvidia and Jeff Bezos, who contributed to the company’s $480 million seed round. This could lead to systems built around interaction and group coordination over full task delegation, supporting organizational models that maintain human oversight and judgment while still using artificial intelligence for assistance.

Go Deeper -> An A.I. Start-Up Says It Wants to Empower Workers, Not Replace Them – The New York Times

AI startup Humans& raises $480 million at $4.5 billion valuation in seed round – Reuters

Claude Is Taking the AI World by Storm, and Even Non-Nerds Are Blown Away – Wall Street Journal