A new industry-standard benchmark reveals that top AI models from OpenAI and Anthropic remain incapable of performing complex, long-horizon professional tasks despite massive capital investment.
The promise of “job-ready” AI agents has met a sobering reality check with the release of the Agents’ Last Exam (ALE). Developed by Berkeley RDI and 300 industry experts, the benchmark evaluates AI systems on 1,500 economically valuable tasks across 55 sub-industries. The results indicate that the Algorithmic State is not yet ready to replace human professional expertise. This benchmark measures a new capability frontier, focusing on sustained work that spans 40 industry subdomains, a massive expansion over previous standards like Terminal-Bench.
Testing of frontier systems including OpenAI’s GPT-5.5, Anthropic’s Claude Fable 5, and Composer 2.5 revealed a significant performance ceiling. On the most difficult tier—tasks requiring sustained reasoning and deep domain knowledge—every mainstream agent recorded a 0% success rate. Even on broader evaluations, the best-performing configurations only managed a 25.2% pass rate. This is a stark contrast to traditional benchmarks where these models often reach 82%. ALE tasks are derived from real projects previously completed by human experts, requiring agents to operate via GUI and CLI environments just as a human professional would.
Data capitalism’s financial toll is also coming into focus. Anthropic’s Fable 5 recorded an average per-task API cost of $15.70, significantly higher than the $3.80 required by GPT-5.5 and the $1.33 for Composer 2.5. These cost disparities, combined with reliability issues that forced Fable 5 to roll back to older Opus 4.8 backbones in 35% of runs, suggest that AI infrastructure remains volatile. Safety and routing behaviors are materially dragging down performance, yet the industry continues to push these systems for enterprise deployment.
This volatility is evidenced by Anthropic’s recent decision to pause token-based billing for its Claude Agent SDK to prevent cost overruns. As AI companies pivot toward the energy sector to secure scarce electricity, the physical and financial costs of maintaining these digital agents collide with the limitations of their actual utility. The AI boom is driving companies across the economy into the energy business as electricity emerges as a scarce commodity, even as oil prices hit three-month lows following a U.S.-Iran ceasefire.
While the Berkeley RDI report highlights current failures, the corporate sector continues to pour billions into the frontier. Prometheus, an industrial AI startup led by Vik Bajaj and Jeff Bezos, recently secured a $12 billion Series B round at a $41 billion valuation to develop an “artificial general engineer.” Simultaneously, Visa and OpenAI have announced a partnership to integrate AI agents into payment networks, a move that raises fresh surveillance and sovereignty concerns as automated systems gain the power to execute financial transactions.
Strategic moves are also being made in the infrastructure layer. SpaceX recently acquired the AI coding platform Cursor for $60 billion, while Google DeepMind released DiffusionGemma, a 26B-parameter model that uses diffusion techniques to refine text blocks in parallel. As the industry prepares for the 2026 Agentic AI Summit in August, the ALE benchmark serves as a necessary corrective to Big Tech’s marketing. For citizens concerned with digital sovereignty, the message is clear: the current generation of AI agents can simulate work, but they cannot yet master the complex workflows that define human labor.
Lisa Grant serves as a Staff Writer for Just Right News, where she spearheads the publication’s coverage of Technology, Data Capitalism, and Surveillance. With a focus on the encroaching influence of Big Tech on the American way of life, Grant brings a critical, liberty-minded perspective to the most complex digital issues of the modern era. Her reporting is defined by a deep-seated skepticism of centralized power and a commitment to protecting the privacy and autonomy of the individual against the rising tide of what she calls the “Algorithmic State.”
Grant’s unique insight into the tech industry is rooted in her upbringing in Palo Alto, California. Growing up in the epicenter of Silicon Valley, she witnessed firsthand the transformation of the technology sector from a hub of scrappy, freedom-loving innovators into a landscape dominated by monolithic corporations. This proximity to the birth of the digital revolution provided her with an insider’s understanding of the culture and motivations driving the industry. For Grant, the shift toward data capitalism—where personal information is harvested as a primary commodity—is not just a market evolution, but a fundamental challenge to traditional American values of property rights and personal privacy. She saw the “garage startup” ethos replaced by a culture of data-mining and social engineering, a transition that informs her vigilant reporting today.
Now based in Seattle, Washington, Grant operates from another of the nation’s primary technological frontiers. Her location in the Pacific Northwest allows her to observe the real-world consequences of the tech industry’s expansion, from the implementation of invasive surveillance technologies in urban centers to the growing partnership between corporate entities and municipal governance. By reporting from the ground in Seattle, she bridges the gap between the abstract world of coding and the tangible impact it has on citizens’ daily lives, often highlighting how local policies serve as a testing ground for broader national surveillance initiatives.
At the heart of her work for Just Right News is her acclaimed feature series, “The Algorithmic State.” Through this series, Grant explores the ways in which automated systems and artificial intelligence are increasingly used to bypass traditional legislative processes and social norms. She argues that the reliance on opaque algorithms to manage society threatens to erode the transparency and accountability essential to a free republic. Her work meticulously documents how data-driven governance can lead to a “soft” surveillance state that penalizes traditional viewpoints and rewards digital conformity.
Grant’s reporting is a vital resource for readers who are wary of the “nanny state” and the unchecked power of digital gatekeepers. She views the defense of the digital frontier as the next great battle for constitutional conservatives. By exposing the mechanisms of data capitalism and the quiet expansion of surveillance networks, she empowers her audience to reclaim their digital sovereignty. In an era where information is often weaponized by those in power, Lisa Grant remains a steadfast advocate for the truth, ensuring that the principles of liberty and individual agency are not lost in the transition to an increasingly digital world.