Research & Deep Dives
Research papers, reports, scenarios, and technical findings translated into practical implications for builders and operators.
Open Research & Deep DivesA survey argues that agent reliability depends as much on the execution harness as on the model and gives builders a vocabulary for the infrastructure surrounding agents. It organizes execution, context, state, tools, recovery, and evaluation into one reliability model.
AgingBench argues that deployed agents can degrade as their memory state changes, so reliability needs lifespan testing rather than only day-one benchmarks. Its evaluation tests how accumulated state and environmental change affect performance over time.
Language Models Need Sleep argues that long-context systems may require offline consolidation time, not merely larger memory, to reason over context they can no longer attend to. The analogy motivates architectures that compress experience between active reasoning cycles.
A legal history argues that AI authorship is not a new copyright problem but the latest form of mediated authorship. Earlier disputes over photography and mechanical creation reveal recurring legal tests for control and originality.
Polar shows how to train coding agents through the harnesses they already use instead of rebuilding those environments for reinforcement learning. The approach preserves realistic tools and feedback while making training operationally tractable.
Epicure shows how ingredient embeddings can become navigable tools for cooking, menu design, and food AI. The system turns flavor relationships into controllable dimensions for substitution, exploration, menu planning, and creative composition rather than returning another opaque ranking.
SkillOpt treats reusable agent skills as operating assets that can be trained, validated, and reused. It shows how procedural instructions can be optimized against tasks, checked for regressions, and distributed with evidence.
Constraint decay explains why AI coding agents struggle when backend work carries real architectural and database rules. The paper traces how requirements fade across long trajectories and proposes stronger constraints, memory, and verification.