From Code to Discovery: The Next Frontier of AI Agents in Research

In a recent social media post, AI researcher Omar Saray hinted at a significant shift in how artificial intelligence systems are being developed and deployed. While current advancements have largely focused on what he terms "agentic coding"—where AI agents assist with or autonomously handle programming tasks—Saray suggests we should "prepare yourselves for 'agentic research'" and promises more details in the coming weeks.

This brief but provocative statement points toward what could be the next major evolution in AI capabilities: moving from tools that help build software to systems that can actively participate in the scientific research process itself.

The Current Landscape: Agentic Coding

Before examining what "agentic research" might entail, it's important to understand the foundation upon which it's being built. Agentic coding refers to AI systems that can perform programming-related tasks with significant autonomy. These aren't just code-completion tools like GitHub Copilot, but rather systems that can understand complex requirements, break them down into components, write and test code, debug issues, and iterate toward solutions.

Recent developments in this space include:

• AI systems that can complete entire software projects from natural language descriptions
• Autonomous debugging agents that identify and fix errors without human intervention
• Code-review assistants that analyze security vulnerabilities and suggest improvements

These systems represent a substantial leap from earlier AI programming tools, moving from assistance to agency—hence the term "agentic."

What Could Agentic Research Look Like?

While Saray hasn't yet provided specific details about what he means by "agentic research," we can extrapolate from current trends and research directions in AI. Agentic research would likely involve AI systems that can autonomously or semi-autonomously conduct various aspects of the scientific research process.

Potential capabilities might include:

Literature Review and Hypothesis Generation

AI agents could systematically review existing research, identify gaps in knowledge, and generate novel hypotheses worth testing. Unlike current literature review tools that simply help researchers find papers, agentic research systems might synthesize information across disciplines to propose entirely new research directions.

Experimental Design and Execution

In fields where experiments can be automated or simulated, AI agents could design experimental protocols, control laboratory equipment, run simulations, and collect data with minimal human oversight. This could dramatically accelerate the pace of discovery in fields like materials science, drug discovery, and physics.

Data Analysis and Interpretation

Beyond just running statistical tests, agentic research systems might identify patterns humans would miss, propose alternative interpretations of results, and suggest follow-up experiments based on findings.

Paper Writing and Peer Review

While controversial, AI systems might eventually draft research papers, respond to reviewer comments, and even participate in the peer review process—though human oversight would likely remain essential for the foreseeable future.

The Implications of Agentic Research

The shift from agentic coding to agentic research represents more than just another technical improvement—it could fundamentally change how knowledge is created and validated.

Accelerating Scientific Discovery

If AI agents can work 24/7 without fatigue, systematically explore possibilities humans might overlook, and rapidly iterate through experimental designs, the pace of scientific discovery could increase dramatically. This could be particularly valuable in addressing urgent global challenges like climate change, pandemic preparedness, and sustainable energy.

Democratizing Research

Agentic research tools could lower barriers to entry for scientific investigation, allowing smaller institutions, independent researchers, and even citizen scientists to conduct sophisticated research that currently requires large teams and expensive infrastructure.

Changing Research Roles

Rather than replacing human researchers, agentic AI systems would likely transform their roles. Researchers might spend less time on routine tasks like literature reviews and data collection, and more time on creative thinking, interpreting unexpected results, and designing high-level research programs.

Ethical and Validation Challenges

Agentic research systems would raise significant questions about research integrity, reproducibility, and accountability. How do we validate findings from AI-conducted research? Who is responsible if an AI system makes an error that leads to false conclusions? These questions would need careful consideration as the technology develops.

The Road Ahead

Omar Saray's promise of more details in the coming weeks suggests we may soon see concrete demonstrations or frameworks for agentic research systems. Given his background and the rapid progress in AI capabilities, it's reasonable to expect initial implementations focused on specific, well-defined research tasks rather than fully autonomous scientific discovery.

Early applications might include:

• Automated literature review systems for specific domains
• AI assistants for experimental design in computational fields
• Data analysis agents that can suggest next steps based on results

As with agentic coding, the transition will likely be gradual, with systems taking on increasingly complex research tasks as their capabilities improve and researchers gain confidence in their outputs.

Conclusion

The move from agentic coding to agentic research represents a natural evolution in AI capabilities—from systems that help us build tools to systems that help us build knowledge. While the full implications won't be clear until we see specific implementations, the potential to accelerate scientific progress is significant.

As we await more details from Omar Saray and other researchers working in this space, the scientific community should begin considering how to integrate such systems responsibly, how to validate their outputs, and how to adapt research methodologies for a future where AI is not just a tool but an active participant in the discovery process.

The age of agentic research may be closer than we think, and it promises to reshape not just how we conduct science, but potentially what we're capable of discovering.

Source: Omar Saray (@omarsar0) on X/Twitter