
When Not to Build a Multi-Agent System
A multi-agent system buys you exactly one thing, context isolation, and most teams reach for one long before that trade-off is worth making.
Field notes · 241 articles
Practical notes on AI, business software, and the decisions behind a production build.
AI, software, and business

A multi-agent system buys you exactly one thing, context isolation, and most teams reach for one long before that trade-off is worth making.

Slopsquatting is when an attacker registers the package name an AI coding tool keeps hallucinating, then waits for someone to install it. Here's how it works and how to defend against it.

A documented AI agent sandbox escape in July 2026 is forcing a hard question in AI agent security: what does your agent actually have the power to reach?

The 2026-07-28 MCP revision drops sessions, the initialize handshake, and server-initiated requests. What breaks, and how to migrate.

GitHub's native stacked PRs, in practice: how the gh stack workflow works, how to size a stack, and when stacking is the wrong tool.

agent-for-insurance is an open-source drafting aid that will not state a coverage conclusion without citing your policy's own text. Here's how it works, and the two parsing bugs that taught us why that rule has to be enforced in code, not prose.

A practical comparison of AI agent memory frameworks Mem0, Zep, and Letta, covering the LongMemEval benchmark and which one actually fits your constraints.

Why an AI coding tools stack beats a single tool: what job Cursor, Claude Code, and GitHub Copilot each do, and how to decide which one gets a given task.

We open-sourced an AI agent that reconciles a bank statement against a ledger, computes MACRS depreciation, and refuses to give tax advice. Here's what it actually does, and the double-counting bug fuzzing caught that reading never would have.

AI coding agents in large codebases fail because the context that makes a change correct rarely lives in the file they're editing, and repository intelligence, instruction files, commit history, and subsystem scoping are how you fix that.

An independent comparison of the three agentic models that launched this month. Real pricing, production failure modes, cost per completed task, and the routing table we actually run, not a benchmark aggregation.

A step-by-step tutorial for bulk AI document extraction with the Claude Opus 5 Batch API: structured outputs with JSON schemas, 50% batch pricing, the 300K output beta, and the real cost of processing 40,000 vendor contracts.

How to use the Claude Opus 5 effort parameter (low, medium, high, xhigh, max) to cut Claude API costs. Real per-request cost numbers, working Python code, and why we retired our Haiku/Sonnet/Opus routing layer.

A hands-on guide to building autonomous AI agents on the Claude Opus 5 API: the agent loop in Python, mid-conversation system messages, prompt caching costs, and what an overnight dependency-upgrade run actually cost.

Claude Opus 5's safety classifiers can decline requests with stop_reason: refusal, even on legitimate security work. How to handle refusals in the Claude API, use the new fallbacks parameter, and design pipelines that degrade gracefully.

An honest Grok 4.5 review after a week in a production agent stack. Grok 4.5 API pricing ($2/$6 per MTok), OpenAI-compatible integration code, reasoning effort settings, and a Grok 4.5 vs Claude Opus 5 routing comparison.

A hands-on Grok Voice API tutorial: WebSocket setup, turn detection, and function calling for live order lookups in a support agent.

A practical guide to building a multilingual AI customer support voice agent on Grok Voice: the new 26-voice roster, language hints, key terms for code-switched speech, pronunciation maps, and custom brand voices.

Build an AI phone answering service on Grok Voice: SIP integration, native G.711 telephony audio, VAD tuning for phone calls, and the real per-call cost of an after-hours AI answering agent versus a human service.

A hands-on Meta Model API guide: Muse Spark 1.1 pricing at $1.25/$4.25 per million tokens, OpenAI- and Anthropic-compatible endpoints, working Python setup code, and an honest look at what the public preview is missing.