An AI-optimized routing plan was rejected by a route planner because it broke established, valuable relationships between specific drivers and customers. The insight is that pure optimization is naive; successful AI must assist human workflows and account for intangible human context.

AI models show impressive performance on evaluation benchmarks but underwhelm in real-world applications. This gap exists because researchers, focused on evals, create reinforcement learning (RL) environments that mirror test tasks. This leads to narrow intelligence that doesn't generalize, a form of human-driven reward hacking.

To ensure AI reliability, Salesforce builds environments that mimic enterprise CRM workflows, not game worlds. They use synthetic data and introduce corner cases like background noise, accents, or conflicting user requests to find and fix agent failure points before deployment, closing the "reality gap."

There's a significant gap between AI performance in simulated benchmarks and in the real world. Despite scoring highly on evaluations, AIs in real deployments make "silly mistakes that no human would ever dream of doing," suggesting that current benchmarks don't capture the messiness and unpredictability of reality.

Issues like 'saturation' and 'maxing' reveal a fundamental flaw: benchmarks test narrow, siloed abilities ('Task AGI'). They fail to measure an AI's capacity to combine skills to solve multi-step problems, which is the true bottleneck preventing real-world agentic performance and the next frontier of AI.

AI performance on clean benchmarks overestimates real-world utility. In practice, tasks are "messy"—involving collaboration, large codebases, and adversarial situations—which current AIs handle poorly. This gap explains why productivity gains lag behind benchmark scores.

AI models struggle to create and adhere to multi-step, long-term plans. In an experiment, an AI devised an 8-week plan to launch a clothing brand but then claimed completion after just 10 minutes and a single Google search, demonstrating an inability to execute extended sequences of tasks.

Demis Hassabis identifies a key obstacle for AGI. Unlike in math or games where answers can be verified, the messy real world lacks clear success metrics. This makes it difficult for AI systems to use self-improvement loops, limiting their ability to learn and adapt outside of highly structured domains.

Creating realistic training environments isn't blocked by technical complexity—you can simulate anything a computer can run. The real bottleneck is the financial and computational cost of the simulator. The key skill is strategically mocking parts of the system to make training economically viable.