Human expertise becomes AI’s biggest bottleneck

By AI, Created 1:51 PM UTC, June 01, 2026, /AGP/ – Business Reporter published an article from Prolific’s Jasmehr Bhatia arguing that AI progress is increasingly limited by the quality of human feedback, not compute or cost. The piece says organizations now need domain experts who can judge both outputs and reasoning, making expert calibration central to safer, more reliable model evaluation.

Why it matters: - AI systems are getting more complex, but evaluation is not keeping up. - The main constraint on safer, more reliable AI is shifting from computing power and cost to the quality of human feedback. - Poor evaluation can slow model improvement and make it harder to know whether training is actually working.

What happened: - Business Reporter published an article featuring Jasmehr Bhatia of Prolific on June 2, 2026. - The article focuses on the growing challenge of evaluating advanced AI systems with human experts. - The industry problem is being described as the “instruction gap,” a mismatch between what engineers need and what experts can provide without enough operational context.

The details: - Earlier AI systems often depended on repetitive labeling tasks. - Current models require domain experts to judge whether an output is correct and whether the reasoning behind it is sound. - A cardiologist reviewing a diagnostic AI system needs to know whether the tool is meant for emergency triage or routine screening. - A securities lawyer assessing a contract review agent needs the regulatory and commercial context around the task. - The article says organizations should treat expert selection like hiring. - The article also recommends long-term relationships with contributors, calibration and training, and continuous feedback loops inside AI development workflows. - Prolific says its platform gives over 35,000 AI developers, researchers and organizations access to quality human data from real participants. - Prolific says a new study launches on its platform every 2 minutes. - Business Reporter publishes content across online hubs including Bloomberg, Fortune, USA Today, Wired, Independent, Die Welt, Business Insider Germany and Le Figaro. - Business Reporter also hosts conferences, debates, breakfast meetings and exclusive summits. - More information is available in the company announcement.

Between the lines: - The article suggests AI progress now depends as much on human workflow design as on model architecture. - That raises the value of subject-matter experts who can work with AI teams, not just answer isolated prompts. - The “instruction gap” points to a broader operational problem: expertise is only useful if the task is framed clearly enough for experts to evaluate it consistently.

What’s next: - Organizations are likely to invest more in expert onboarding, calibration and feedback systems. - AI teams may need tighter task definitions and better context-sharing if they want reliable evaluation at scale. - The article frames human feedback as a continuing bottleneck, so better evaluation processes may become a competitive advantage.