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

Recent analyses confirm that AI systems have achieved near-complete automation in core AI engineering tasks, while research processes remain less automated but are rapidly advancing toward similar levels of capability. This shift is poised to reshape AI development timelines and strategies, with significant implications for the industry.

According to Thorsten Meyer’s synthesis of Jack Clark’s recent work, six benchmarks measuring AI’s proficiency in core AI research and engineering skills are nearing saturation. For example, the CORE-Bench, which tests AI’s ability to reproduce research papers, reached 95.5% in December 2025, with some experts claiming it is effectively ‘solved.’ Similarly, the MLE-Bench, assessing performance on Kaggle competitions, has also shown significant progress, reaching 64.4% in February 2026, with the leaderboard paused to develop fairer evaluation methods as models outgrow existing benchmarks.

Clark’s analysis suggests that the bottleneck in AI research—reproducing experiments, generating optimized code, and designing kernels—is rapidly diminishing. These capabilities are transitioning from experimental to production-grade, with recent advances demonstrated through automated GPU kernel design and PyTorch-to-CUDA conversions. The evidence indicates that engineering tasks are increasingly fully automated, while research tasks are on a similar trajectory but remain less fully automated, leaving some residual work.

Implications for AI Development and Industry Strategy

This rapid progress in automating AI engineering tasks could drastically shorten AI development cycles, reduce costs, and democratize access to advanced AI tools. As engineering becomes automated, the focus may shift toward higher-level research and innovation, which are less mature in automation. The potential for AI to also automate research processes suggests a future where human involvement is minimized, accelerating breakthroughs and possibly transforming the landscape of AI R&D.

Key Benchmarks and Progress in AI Capabilities

Recent assessments, including Clark’s review of six core benchmarks, show that AI systems are nearing or have achieved saturation levels in tasks critical to AI research and engineering. The CORE-Bench, which tests research reproduction, improved 4.4× over fifteen months, with some claiming it is now ‘solved.’ Similarly, the Kaggle-based MLE-Bench has moved from 16.9% to 64.4%, indicating AI’s growing proficiency in competitive ML tasks. These benchmarks reflect a broader pattern of rapid capability growth across multiple domains, driven by advances in model architecture, training techniques, and specialized tools.

The development of automated kernel design, GPU optimization, and code translation further exemplifies the trend toward automation of engineering tasks. Experts note that these capabilities are now transitioning from experimental to production-ready, suggesting a structural shift in how AI systems are developed and deployed.

“The evidence indicates that engineering tasks are increasingly fully automated, while research tasks are on a similar trajectory but remain less fully automated.”

— Thorsten Meyer

Unconfirmed Aspects of Research Automation Speed

While evidence suggests research automation is accelerating, it is not yet clear how much of AI research—beyond engineering tasks—can be fully automated. The structural question Clark leaves open remains unresolved: whether research itself is just scaled engineering, which could mean faster automation than currently observed. Additionally, the long-term implications of automation on creativity and innovation are still uncertain.

Next Steps for AI Automation and Research Integration

In the coming months, researchers and industry leaders will focus on developing more comprehensive benchmarks for research automation and testing AI’s capabilities in higher-level scientific tasks. Monitoring the progression of AI in automating research processes will be critical, as will efforts to understand how automation impacts innovation cycles. Industry strategies may shift toward integrating increasingly autonomous AI systems into the core of R&D workflows, potentially transforming the pace of AI development.

Key Questions

How close are AI systems to fully automating AI research?

Current evidence suggests that AI has nearly automated core engineering tasks and is rapidly advancing in automating research processes, but full automation of all research aspects remains unconfirmed and is still under development.

What are the main benchmarks indicating progress?

The CORE-Bench for research reproduction and the MLE-Bench for Kaggle competition performance are key indicators, both nearing saturation and demonstrating rapid improvements.

What does this mean for human researchers?

Automation of engineering tasks will likely reduce manual effort and costs, shifting human focus toward higher-level research and creative problem-solving, though some research aspects may still require human insight.

Are there risks associated with AI fully automating research?

Potential risks include reduced oversight, loss of scientific diversity, and challenges in maintaining quality control, but these are areas under active discussion and regulation development.

Source: ThorstenMeyerAI.com