Your mission
Your mission
Your mission
Papers, future state, newsletters, and architectures for advanced learners.
Best for: Researchers, advanced learners, and people tracking the frontier.
Research
Why this is useful: LLM agents are shaped not only by their language models but also by the runtime harness that mediates observation, tool use, action execution, feedback interpretation, and trajectory control. Life-Harness improves frozen LLM agents without changing model weights by converting recurring interaction failures into reusable interventions across seven deterministic environments.
Research
Why this is useful: Accompanies the survey "Code as Agent Harness: Toward Executable, Verifiable, and Stateful Agent Systems." Studies the emerging role of code in agentic AI: code is no longer only a generated artifact but increasingly serves as an executable, inspectable, and stateful harness through which agents reason, act, model environments, receive feedback, and coordinate.
Research
Why this is useful: Introduces context engineering as a formal discipline that transcends simple prompt design. The survey covers systematic optimisation of information payloads for LLMs across the full inference pipeline.
Research
Why this is useful: Language-Induced Priors (LIP) uses LLMs to translate natural-language descriptions into probabilistic priors that guide learning in cold-start settings. The framework identifies which historical systems are most relevant based on semantic context, then automatically adapts the new model using EM. Tested on descriptive, predictive, and prescriptive (RL) models.
Research
Why this is useful: Treats context strategies like recommendation items: inputs are users, strategies are items, and task accuracy is the interaction signal. NCCE models the observation that different inputs benefit from different forms of guidance explicitly, rather than optimising a single best prompt that averages away instance-level differences.