FROM THE TRENCHES Issue №288

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.

Claude text watermark signals moving through an API provenance pipeline AI & Agents

Does Claude Watermark Text? The 2026 API Answer

Anthropic is adding invisible text watermarks, but current coverage is not universal. Here is what Claude API teams need to record, test, and disclose.

OpenKB compiles raw documents into a linked Markdown wiki before an AI agent queries it AI & Agents

OpenKB Review: Knowledge Compiler vs RAG

OpenKB turns documents into an evolving Markdown wiki instead of a vector index. Compare the evidence, security gaps, cost model and enterprise fit.

Unsloth Desktop linking local model training, inference, coding agents and secure deployment AI & Agents

Unsloth Desktop Review: A Private Local AI Workstation?

Unsloth Desktop unifies local training, inference and coding-agent endpoints. See what is ready, what beta still means and how to run a production-minded pilot.

NeMo Switchyard routing agent turns between capable and efficient models AI & Agents

NeMo Switchyard 0.2: Agent Model Routing Without Training?

A practical review of NVIDIA's open-source router, its new Rust-native paths, stage-routing logic and the evidence a team needs before a production pilot.

Office documents enter one local parser and leave as consistent Markdown for AI agents AI & Agents

Firecrawl AnyDoc Review: 14 Formats to Markdown

Firecrawl AnyDoc converts mixed Office files to Markdown locally. Review its benchmark, limits, Docling and MarkItDown alternatives, and production fit.

MCP Cloud and Manufact Cloud compared across hosting, deployment, governance, pricing and production fit AI & Agents

MCP Cloud vs Manufact Cloud: MCP Hosting Guide

Compare MCP Cloud and Manufact Cloud for remote MCP hosting, governance, Git-based deployment, pricing, security and production fit.

Muse Glimmer 30B local agent model running inside a consumer hardware memory envelope AI & Agents

Muse Glimmer 30B: Is Meta's Local Agent Model Production-Ready?

A buyer-focused guide to Muse Glimmer hardware, agent benchmarks, Apache 2.0 licensing and the pilot evidence needed before production.

Layered writing instructions turning generic AI output into direct, specific and testable copy AI & Agents

How to Make AI Writing Sound Human with Agent Skills

No AI Slop, i-have-adhd and book-to-skill solve different writing problems. Here is how to combine, test and govern them without erasing your voice.

Typed Python class connects state, methods, memory and validated agent loops in NVIDIA NOOA AI & Agents

NVIDIA NOOA Review: Are Object-Oriented Agents Production-Ready?

NVIDIA NOOA turns an AI agent into a typed Python object. Review its architecture, benchmark evidence, security boundary, alternatives and 30-day pilot fit.

Inbox, without the noise

Get the next field note

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