AI & Automation Briefing - October 26, 2026
Apollo Economist Finds Zero Financial Payoff from AI Outside the Tech Sector
Torsten Slok, chief economist at Apollo Global Management, published a note arguing that AI has produced no measurable bottom-line benefit for large U.S. corporations outside the technology industry. Slok analyzed what he calls the S&P 493, the S&P 500 minus Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla, and found profit margins and revenue per employee have remained flat with no visible productivity improvement. He expressed doubt that non-tech companies will be able to convert cheaper AI tools into higher revenue or margins, writing that business outcomes depend on judgment, accountability, trust, and culture. Those are areas where he argues AI falls short. His findings align with recent independent studies that have similarly failed to detect significant productivity gains from AI adoption.
RSIHub Opens a Shared Infrastructure for Building and Comparing Self-Improving Agents
Researchers have released RSIHub, an open-source framework that standardizes how agent-evolution methods are built, tested, and extended. The framework lets developers compose reusable operators that modify agent prompts, skills, tools, and orchestration logic, while keeping evaluation isolated from mutable components and maintaining traceable candidate histories. The team used RSIHub to implement four existing evolution methods, AHE, Hyperagents, A-Evolve, and GEPA, on a single shared infrastructure and benchmarked them against MiniSWE and Codex on Terminal Bench 2 and the τ³ Banking dataset. The modular, recipe-based design is architecturally similar to how n8n-style workflow tools chain reusable nodes, and signals where composable, self-optimizing agent pipelines are heading for production automation use cases.
EvalResearchBench Tests Whether AI Agents Can Design Their Own Evaluations for Self-Improvement
A new benchmark called EvalResearchBench examines whether AI agents can autonomously construct the evaluation criteria needed to drive recursive self-improvement. The research finds that automated evaluators currently fall short of the agreement levels required to reliably assess agent outputs, and they run into compounding problems around budget constraints and score dominance. For anyone building autonomous QA loops or self-correcting pipelines, this is a concrete data point on where automated evaluation breaks down today. The findings are a useful counterweight to assumptions that AI agents can close the feedback loop on their own without human checkpoints.