AI is capable of helping humanity solve our most important challenges. AI systems can accelerate scientific discovery, transform education, and protect humanity from future pandemics. This isn't hyperbole - we're seeing early evidence of these capabilities today.

Yet our current approaches to building AI systems, particularly AI agents that can act with autonomy, severely limit this potential. By constraining agents through hardcoded workflows, rigid frameworks, and human knowledge, we reduce transformative technology to mere automation. It is a recurring fallacy that our field has fallen into throughout the history of AI.

Our principles

1. Agents must be active learners. Knowledge cannot be static, pre-defined, or constrained by what humans think agents need to know. True intelligence requires the ability to identify gaps in understanding and autonomously acquire new knowledge.

2. Agents must be flexible and autonomous. Engineering specific workflows into agents locks them into today's capabilities, while letting agents develop their own approaches allows them to grow with model improvements.

3. Agents must be collaborative partners. We assert that meaningful work requires fluid collaboration between humans and agents, where agents can independently recognize when they need human input, actively seek it, and adapt their work based on human verification and feedback.

Our rejection of the status quo

- We reject the premise that constrained workflows and human knowledge make them better.

- We reject building agents with constraints that make them obsolete as models advance.

- We reject the choice between uncontrolled autonomy and restrictive workflows.

Our vision

We envision a future where AI agents are not merely better automation software but collaborators in solving humanity's most important challenges. This requires a fundamental shift in how we think about agents:

- Agents that actively learn and grow, unconstrained by predefined knowledge.

- Agents that scale with the increasing capabilities of foundation models.

- Agents that collaborate meaningfully with humans, balancing autonomy with verification.

Introducing the agent workspace

We are building Superagent as the agent workspace – where humans and agents come together to solve meaningful problems. The workspace enables agents to:

- Learn and expand their knowledge autonomously.

- Plan and execute their own work.

- Partner with humans through feedback and verification.

The era of rigid workflows must end. The future lies in flexible, autonomous agents that can truly learn and adapt. We invite those who share this vision to join us in building it.