Newsletter archive
Applied Signals Weekly: 2026-05-24
May 18, 2026 to May 24, 2026
In this week’s edition
This week’s edition surfaces where agentic systems face the sharpest pressure points: reliability under coordination, evaluation discipline, and the practical mechanics that keep multi-agent behavior stable over time.
- Building a self-improving agent on a context graph of human disagreement — You can build a measurably better agent from data you already have, without retraining a thing. The data is what your experienced humans do when they correct the AI. Capture...
- MagenticLite, MagenticBrain, Fara1.5: An agentic experience optimized for small… — MagenticLite is an agentic system for small models that works across the browser and local file system in a single workflow. It combines specialized models and orchestration to support efficient agentic performance on…
- How to build LLM-as-a-Judge evaluators that hold up in production — Learn how to design, calibrate, and run LLM-as-a-judge evaluators with fixed labels, human agreement checks, trace context, and Phoenix Evals.
- Coding agent tracing and evaluation: An open source tool to improve AI coding… — Announcing coding harness tracing for observing, evaluating, and improving coding agent workflows across Claude Code, Cursor, Codex, GitHub Copilot, and Gemini CLI.
- Fast-tracking genetic leads to reverse cellular aging — Biologists use Co-Scientist to find novel factors that successfully rejuvenate human cells.
- OpenAI and Dell partner to bring Codex to hybrid and on-premise enterprise… — OpenAI and Dell partner to bring Codex to hybrid and on-premise environments, helping enterprises deploy AI coding agents securely across data and workflows.
- What we learned testing 7 models under the same agent harness — Model swaps look like configuration changes, but they behave more like product migrations. A new model may be cheaper, faster, easier to get capacity for, or stronger on public benchmarks....
Deep dive
In the real world, the hardest part of scaling isn’t raw capability—it’s keeping failure modes observable and fixable as systems become more decentralized. When evaluation loops, instrumentation, and state-management drift faster than fixes, teams accumulate “invisible” debt that shows up only during deployment.
Worth watching
- planning and tool-use often showed up together—watch whether that pairing moves from discussion into repeatable engineering patterns.
- frameworks and tool-use often showed up together—watch whether that pairing moves from discussion into repeatable engineering patterns.
- evaluation and frameworks often showed up together—watch whether that pairing moves from discussion into repeatable engineering patterns.
Read the full analysis
- HealthCraft, χ-Bench, and REFLECT quantify reliability ceilings (weekly trend)
- May 24, 2026 Phoenix Evals LLM-as-a-Judge production evaluator design
- May 23, 2026 WorldKV KV-cache retrieval, coding-agent tracing, SocialReasoning-Bench outcomes
- May 22, 2026 AgentCo-Op workflows, HealthCraft V8 safety-gated tool use
- May 21, 2026 What Do Agents Communicate, SLIM β, and Mem-π guidance policies
- May 20, 2026 CASPIAN cascade attribution, STAR-PólyaMath supervision, REFLECT judge reliability
- May 19, 2026 Skill library scaling, GRASP cost control, and semantic collapse
- May 18, 2026 SDOF routing, TeamTR fine-tuning, SkillSmith compilation, AgentStop energy
Subscribe
Get Applied Signals weekly newsletter in your inbox