📊 Full opportunity report: Engineering Is Automated. Research Is the Residual. on ThorstenMeyerAI.com — validation score, market gap, and execution plan.

Recent advances in AI capabilities demonstrate that AI systems can now automate most core engineering tasks involved in AI development, while the automation of AI research itself remains uncertain.

Multiple benchmarks, including CORE-Bench and MLE-Bench, show AI systems reaching near-saturation levels in core engineering skills such as reproducing research and competing in Kaggle competitions. For example, CORE-Bench, which measures research reproduction, improved from 21.5% to 95.5% over fifteen months, with some authors declaring it ‘solved.’ Similarly, AI performance on Kaggle competitions has risen to a level where it matches mid-tier human practitioners, reaching 64.4% of competitions at a bronze-medal level by February 2026.

These developments suggest that the engineering side of AI—installing dependencies, running experiments, optimizing kernels—can now be largely automated, reducing the need for human intervention. However, the progress in automating AI research—the creative and theoretical aspects—remains less certain, with ongoing debates about whether research itself is just scaled-up engineering or involves distinct, harder-to-automate processes.

Implications for AI Development and Research Automation

The automation of AI engineering tasks signifies a potential shift in AI development, reducing costs and timeframes while raising questions about the future role of human researchers. If research processes also become automatable, it could accelerate AI innovation but also challenge existing institutional and academic structures. This shift may lead to faster deployment of AI systems and a reevaluation of research methodologies, but the residual uncertainty around research automation means the full impact remains uncertain.

Recent Benchmarks and Progress in AI Engineering Capabilities

Over the past year, multiple independent benchmarks have demonstrated rapid progress in AI’s technical skills relevant to AI development. CORE-Bench, measuring research reproduction, improved from 21.5% to 95.5%; MLE-Bench, assessing Kaggle competition performance, rose from 16.9% to 64.4%. These trajectories suggest that AI systems are reaching or surpassing human-level proficiency in core engineering tasks. The progress aligns with broader patterns of saturation and measurement limits across different skill domains, indicating that much of the engineering process can now be automated.

Despite these advances, the automation of research—such as hypothesis generation, theoretical development, and experimental design—remains less certain. Clark’s analysis leaves open whether research is fundamentally different from engineering at scale, which influences the future of AI development strategies.

“The pattern across multiple benchmarks indicates that AI is approaching or has reached saturation in core engineering skills, suggesting that much of the engineering process can be automated.”

— Thorsten Meyer

Unclear Extent of Research Automation

While engineering tasks have shown near-complete automation, the automation of AI research—such as hypothesis formulation, theoretical innovation, and experimental design—remains uncertain. It is not yet confirmed whether research is reducible to scaled engineering or involves fundamentally different cognitive processes that resist automation.

Next Steps in AI Capability Development

Further benchmarking and research are needed to determine whether research automation is achievable at scale. Expect continued progress in engineering capabilities, with potential breakthroughs in automating research processes within the next 32 months. Monitoring institutional responses and new research publications will be critical to assess the trajectory.

Key Questions

How close are AI systems to fully automating AI engineering?

Recent benchmarks indicate AI can automate most core engineering tasks, with performance approaching or surpassing human levels in some areas. However, complete automation of all engineering aspects is still developing.

Can AI fully automate AI research?

It is currently unclear. While engineering tasks are nearing automation, the creative and theoretical aspects of research remain uncertain, and ongoing debate questions whether they can be fully automated.

What does this mean for human researchers?

If engineering becomes fully automated, human researchers may focus more on high-level hypothesis and strategic planning, but the full impact depends on whether research itself can be automated.

When might we see breakthroughs in automating research?

Based on current trajectories, significant advances could occur within the next 32 months, but the timeline depends on ongoing developments and institutional responses.

Source: ThorstenMeyerAI.com