Microsoft researchers introduced Coding-Agent Skill Distillation (CASD), a method that writes improved system prompts from a static corpus of saved agent logs in a single pass for approximately $1.60,…
Mara Chain, a refinement procedure that retains and iteratively refines rejected candidate configurations instead of discarding them, outperformed GEPA, ACE, and SkillOpt-Lite by up to 20.5% in relati…
DSPy 3.4.0 has been released with PythonInterpreter improvements, faster GEPA optimization, and MCP v2 compatibility, according to the Stanford NLP project's release notes. The Python framework, which…
Deepfates released Imp, a full port of the DSPy language-model programming framework to the Elixir/BEAM ecosystem, letting developers declare typed model calls, agent loops and retrieval that run unde…
GEPA, a prompt-optimization method in which a model reads an agent's full trace — reasoning, tool calls and errors — and rewrites the prompt, reportedly doubled the gains of GRPO after a single round …
A newly released paper applies Sutton's Bitter Lesson to data agents, arguing that as large language models improve, general coding agents will absorb the hand-engineered scaffolding that human-design…
A developer released Shrewd, a project that distills labels generated by frontier LLMs into small, single-purpose classifiers that run efficiently on-device, and published results on public datasets u…
Sutro released jev-align, an experimental open-source CLI that calibrates TypeSafe's Jev AI Functions by having users label uncertain examples and then applying GEPA to improve the function. The tool …
A new benchmark of 500 single-error clinical note pairs shows LLM judges detect added or altered content with paired discrimination scores of 0.79-0.94 but fail on omissions, scoring 0.50-0.63, accord…
RL environments are the essential data for training AI agents, according to a recent perspective shared by an industry commentator. The author argues that just as curated datasets enabled deep learnin…
Agentic Context Engineering (ACE) improves LLM systems by evolving their context instead of their weights, using three agents—Generator, Reflector, and Curator—to maintain a playbook through small, de…
DynaContext, a framework introduced in arXiv:2608.22014v1, combines offline-optimized extraction with inference-time contextual adaptation and validation-gated self-improvement, boosting field-level F…
Carnegie Mellon University PhD student Michael Isaac, during an internship at cmpnd, implemented Flex, a new DSPy module that lets models rewrite program code, not just instructions, to optimize perfo…
Surya and Cameron Franz trained a language model to generate editable p5.brush JavaScript sketches that render into watercolour-style images, using reinforcement learning with a judge model comparing …
A new arXiv paper (2608.10694v1) proposes a cost-aware cross-tier transfer method for evolutionary optimization of LLM prompts and agentic programs, reducing search cost by 5.6-14x (up to 25-54x with …
DSPy creator Omar Khattab's toolkit and its GEPA optimizer are the targets of a polemic by an unnamed developer who argues that treating LLM workflows as modular programs is 'backwards, rigid, and the…
Sidekick's continual learning loop compresses production experience into model weights, delivering higher quality than frontier models while cutting costs by 96% in Shopify's GraphQL agent. The proces…
Researchers introduced FLARE (Few-shot Learning-based Adaptive Reflective Engine), a framework that uses reflective mechanisms and few-shot reference examples to optimize prompts for large language mo…
Researchers introduce GEPA (Genetic-Pareto), a prompt optimizer that uses natural language reflection to learn from trial and error, outperforming reinforcement learning methods like GRPO by 6% on ave…
DeepMind's AlphaEvolve broke a 56-year-old record by multiplying 4x4 matrices in 48 scalar multiplications, while recovering 0.7% of Google's worldwide compute and cutting Gemini's training time by 1%…