Agentic AI excitement runs ahead of results

Investor excitement for agentic artificial intelligence is outpacing the technology's capacity to generate substantial revenue and productivity gains, according to a BCA Research report. Analyst Matt Dyson cautioned that maintaining positive market sentiment through 2027 will necessitate either higher AI revenues or a groundbreaking application like personal assistants.

BCA Research identified a persistent disparity between AI adoption and quantifiable financial returns. A 2025 MIT study found that 95% of corporate AI trials didn't yield deployment with positive returns within six months. A 2026 McKinsey survey showed that only 6% of respondents claimed their AI adoption contributed at least 5% to operating earnings.

BCA Research estimated that AI systems aided by human oversight could execute tasks around 75% cheaper than U.S. employees—but these cost savings frequently overlook what the firm terms the "AI validation tax," referring to the extra time employees spend scrutinizing AI-generated work. The report also queried whether the surge in AI token consumption truly mirrors productive utilization.

Sophisticated reasoning models consume four to 15 times as many tokens, providing an explanation for the surge in demand without a corresponding rise in useful output. Agentic AI has also sparked apprehensions in traditional software companies. Anthropic's Claude Cowork plugins' introduction led to a 15% drop in the MSCI U.S. IT Services index in January, impacting firms like Salesforce, Intuit, and Workday.

BCA Research deemed this market reaction as overly exaggerated, asserting that smaller businesses would likely remain loyal to established software platforms. The firm anticipates companies will require at least a year to adjust workflows around AI capabilities. Dyson also warned that merely 20% of freelance tasks executed by leading AI models currently satisfy reasonable client acceptance standards, underscoring the gap between technical performance and commercial usefulness.