AI Briefing#prompt-engineering
The latest AI and developer news about #prompt-engineering, with the original source and a short summary.
The latest AI and developer news about #prompt-engineering, with the original source and a short summary.
Using GEPA, a genetic-algorithm-based approach, prompt tuning work that used to take days has been reduced to just one hour.
Instead of typing prompts directly into coding agents, 'loop engineering'—designing loops that drive the agent on its own—is becoming important.
LangSmith has redesigned its homepage and enhanced resource management features to optimize developer workflows.
LangSmith has introduced a feature that allows LLM evaluation models to improve themselves by learning from human feedback.
The article summarizes 8 core techniques and tools—including prompt caching and model routing—that can reduce LLM operating costs by up to 80%.
As AI agents move beyond simple answers to actually performing tasks, harness engineering, which designs the execution environment, is growing in importance.
The CLAUDE.md configuration file, which prevents Claude Code from malfunctioning, is drawing major attention on GitHub.
Anthropic's Agent Skills use a 3-tier progressive disclosure system to cut token usage by up to 50%.
A plugin that resolves prompt ambiguity in Claude Code and optimizes model costs has been updated.
To address the security and governance issues arising from AI agent adoption, the introduction of harness engineering is necessary.
Anthropic has launched the 'Claude Certified Architect - Foundations' certification exam, which verifies proficiency in using Claude.
An analysis has raised the issue that LLMs may distort instructions by processing hidden system/developer prompts, which take priority over users' custom instructions, first.
Yeogieottae built an AI tool that simultaneously generates 2D/3D icons while maintaining brand consistency and converts them into editable vectors.
A study found that changes in prompt tone can sharply reduce the honesty of small language models.
The Claude Code performance degradation was not due to the model but to changes in execution settings.
In CC 2.1.143, the impossibility-condition judgment and Verify skill verification rules have been strengthened.
Promptimus automatically improves already-good LLM prompts without manual work.
A cache key design was proposed that satisfies both reproducibility and variability together in a non-deterministic LLM environment.
A test harness quantified agent skill invocation rates, and Codex improved more than Claude.
For Claude Code output, HTML is better suited than Markdown for richer explanations.
Claude Code 2.1.132 strengthens Managed Agents documentation and schedule safety rules.
The Claude Code system prompt revised its file-reading and REPL rules.
Defined the harness of LLM development tools and proposed a regression testing tool.
The Claude Code system prompt grew by 166 tokens in 2.1.124 and shrank by 87 tokens in 2.1.126.