Blog
Articles on AI agents, LLM engineering, and applied machine learning — prompt design, RAG, evaluation, and building production AI features.
142 articles
Twenty sources and one unattended Claude Code agent on what really moves agent results: the harness. Harness-only gains, context rot, self-grading bias, and the self-tests a harness needs.
Read More
How to write a CLAUDE.md file in six steps: run /init, prune, write commands and hard rules, split into rules and skills, then test it with /context.
Read More
The Backstage MCP actions backend read closely: what tools it exposes by default, why the default endpoint shows everything, and the filters and auth that fence it.
Read More
Stack Overflow's 2026 survey says 26.2% use AI agents at work and only 6.6% trust AI for important decisions. What the numbers allow, and a review policy.
Read More
METR's February 2026 update reports minus 18 and minus 4 percent task-time changes, both with intervals that cross zero. What it says, what it cannot, and how to measure your own team.
Read More
How Perplexity's open-source SPACE runs agent tasks in Firecracker microVMs: warm pools, suspend and wake, an E2B-compatible API and red-team results.
Read More
Run Claude Code or Codex unattended in a microVM with sbx, then move it to Docker's cloud with one command. What transfers, what does not, and what it costs.
Read More
kagent Kubernetes tutorial: run AI agents on a laptop kind cluster, wire RemoteMCPServer tools, keep sessions, and choose between 0.10 and the 1.0 alpha.
Read More
Kubernetes Agent Sandbox CRD explained: Sandbox, SandboxTemplate, SandboxClaim and WarmPool for isolated stateful AI agents, with gVisor and tuning traps.
Read More
Kubernetes AI conformance explained: what KARs demand, the v1.35 to v1.37 checklist diff, the test suite's silent skips and what a badge does not prove.
Read More
Two tools, search and execute, replace thousands of per-endpoint MCP tools. What Cloudflare measured, how the sandbox works, and what it does not protect.
Read More
Elastic nightshift AI SRE has four engines: Context, Detection, Investigation, Remediation. Why Investigation alone on existing alerts is the safest start.
Read More
MCP connects an agent to tools; A2A connects agents to each other. Discovery, state, long-running tasks, auth and governance compared in one table.
Read More
Agents loop, spawn subagents and grow their context, so per-call pricing math underestimates them. Measure cost per completed task, then control it with max_turns, subagent caps, tool search, compaction and model routing.
Read More
An agent can return the right answer after calling the wrong tool, with the wrong arguments, in the wrong order. Here is how to grade the trajectory itself, from production traces to promptfoo assertions against a NestJS-hosted agent.
Read More
Connectors became apps, apps became plugins, and the Apps SDK URL now redirects to the plugins docs. Here is what a 2026 ChatGPT plugin actually is, and how to build one for an ERP order-status lookup.
Read More
Claude Managed Agents and the OpenAI Agents API both run the agent loop for you. A side-by-side of objects, sandboxes, custom tools, cron scheduling, ZDR, residency and pricing for teams retiring their own loop.
Read More
A hands-on Google ADK 2.0 tutorial: build a procurement approval graph with routes and human approval, persist sessions across restarts, choose between graph, dynamic and collaborative workflows, test it, and migrate from 1.x.
Read More
LangChain 1.0 replaced create_react_agent with create_agent and a middleware list. This guide migrates an old agent, then adds PII redaction, tool-call budgets, a closed-period guard, retries with model fallback and human approval, and explains why their order matters.
Read More
The same three-step invoice workflow built in LangGraph and in CrewAI Flows, compared on control model, checkpointing, resume semantics, human approval, observability and learning curve.
Read More
A hands-on Mastra 1.x tutorial for TypeScript teams: a Zod tool scoped by RequestContext, an agent with structured output, a purchase-order workflow that suspends for a manager, quick-check evals in Vitest, Studio, and how it compares with the Vercel AI SDK and the OpenAI Agents SDK.
Read More
Microsoft Agent Framework 1.0 replaces Semantic Kernel and AutoGen for new agent work. Here is how the concepts map, what the C# and Python code looks like before and after, and what the BUILD 2026 harness and the pre-release CodeAct package change.
Read More
Agent Builder goes away on 30 November 2026 and Evals turns read-only a month earlier. An inventory, export and parity-test plan for moving each workflow to the Agents SDK or a Workspace Agent.
Read More
A hands-on Agents API walkthrough: create a session in an OpenAI-hosted sandbox, stream turns, steer, handle webhooks, clean up, and check residency first.
Read More