Testing & Evals for Claude Agents: Gate Releases Safely
Build an eval loop for Claude agents: outcome and trajectory scoring, calibrated LLM judges, and no-regression gates that block bad releases.
6 min read
Agentic AI & LLMs
Agentic AI & LLMs archive page 40 of 631
Agentic AI & LLMs
Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
Latest articles
Hardening Claude Agents: Sandboxing & Prompt Injection
Security hardening for Claude agents: sandboxing, least privilege, secrets handling, and prompt-injection defense for tool-using agentic systems.
7 min read
Sandboxing an LLM Agent That Reads Your Source Code
Least privilege, sandboxing, secret hygiene, and prompt-injection defense for a Claude source-code security agent. Harden the agent that reads your code.
8 min read
Security Hardening Claude Agents: Sandboxing & Least Privilege
Secure Claude agent orchestration with sandboxed tools, least-privilege scopes, server-side secrets, and layered prompt-injection defenses.
7 min read
Cut Claude Agent Token Cost: Caching & Batching
Make Claude agents cheaper and faster with prompt caching, batching, context trimming, and model routing. Practical token economics for production agents.
7 min read
Cutting Token Cost in an LLM Code Security Scanner
Use prompt caching, batching, and diff-scoped context to keep a Claude-powered source-code security agent fast and cheap without losing coverage.
8 min read
Cutting Token Cost in Claude Agents: Caching & Batching
Keep Claude agent orchestration cheap and fast with prompt caching, batching, model routing, and lean context that cut token bills without losing quality.
7 min read
Debugging a Claude Code Security Agent: Failure Modes
Fix loops, hallucinated CWE IDs, and wrong tool calls in LLM source-code security agents. A practical Claude Code debugging field guide for engineers.
9 min read
Debugging Claude Agent Orchestration: Loops & Bad Tool Calls
Catch loops, wrong tool calls, and hallucinated arguments in Claude agent orchestration with tracing, loop detection, and schema-validated tool boundaries.
8 min read