FROM THE TRENCHES Issue №354

Practical decisions for AI and software

Technical guides to choose models, verify software and plan delivery.

Latest articles

New comparisons, practical guides and engineering lessons.

Two source generations connected by one atomic active pointer for AI coding agents AI & Agents

Atomic Multi-File Edits for AI Coding Agents: The Semaprax Lesson

Why sequential file writes can expose a broken half-update, and how immutable generations plus one active pointer create a safer publication boundary.

A phone waveform passes through the Alma voice LLM into a production evaluation scorecard AI & Agents

Phonely Alma Review: Is the Voice LLM Ready for Production?

Alma claims 182 ms TTFT, a 206 ms P99 and $0.55 per blended million tokens. See what the benchmark proves, what it omits and how to run a buyer-side voice-agent pilot.

Layered Utopia knowledge records represent preserved history AI & Agents

Utopia Review: Temporal Knowledge Graph for Enterprise

Evaluate Utopia for enterprise knowledge: two-clock history, source provenance, self-hosting limits and a practical pilot plan before you invest.

Frontier AI model writing a reusable manual that guides a cheaper model through repeated agent tasks AI & Agents

AI Agent Knowledge Transfer: Use Frontier Models Once, Then Scale Cheaper

Agno’s ARC-AGI-3 result shows how explicit, verified manuals can move discovery cost out of repeated agent runs without retraining a model.

Claude Code devices routed through one quota-aware claude-rotate proxy AI & Agents

claude-rotate: One Proxy for Multiple Claude Max Accounts

A review of claude-rotate’s quota tracking, Anthropic’s rules for subscription credentials, and supported ways to handle Claude Code limits.

NVIDIA PAIR routes local AI agent requests across RTX, Mac and an AMD Strix Halo node AI & Agents

NVIDIA PAIR Review: Local AI Routing and the AMD Gap

PAIR routes independent agent requests across local computers without pooling GPUs. See how it works, where Strix Halo falls short and what an AMD port must prove.

Feynman research agents connect papers, code, citations and verified findings in a terminal workflow AI & Agents

Feynman Review: Is This Open-Source AI Research Agent Ready for Teams?

A buyer-focused review of Feynman’s multi-agent research, citation checks, local models, GPU replication, costs, risks and pilot fit.

Obscura Rust browser evaluation with benchmark and Chrome fallback lanes AI & Agents

Obscura Browser Review: Claims, Limits and Production Fit

Obscura promises 30 MB memory, native rendering and stealth without Chromium. See what the evidence proves, where compatibility ends and how to pilot it safely.

Four-part Claude Code design system with reference, instructions, rules and examples AI & Agents

Claude Code Design System: 4 Parts for On-Brand UI

Stop re-explaining your brand in every prompt. Build a reusable Claude Code design system with three Markdown files, an examples directory and a review loop that teams can govern.

Inbox, without the noise

Get the next field note

One concise email when we publish. No tracking pixels, and no inbox filler.

Facing one of these decisions right now? Talk it through with us →