FROM THE TRENCHES Issue №309

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.

Tencent Hy4 preview routes a million-token codebase through coding-agent tools and verification gates AI & Agents

Tencent Hy4 Preview Review: Is the 1M-Context Coding Model Worth a Pilot?

A buyer-focused Hy4 preview review covering coding benchmarks, WorkBuddy, API cost, self-hosting hardware, known limits and a two-week evaluation plan.

Kevin Riedl
Read
Three teammates shaping one AI-built application in a shared real-time workspace Leadership & Teams

Google Play with Putty: Is Multiplayer Vibe Coding Ready for Teams?

Google Play with Putty moves AI app building from a solo prompt into a shared room. Here is where that helps teams, which controls remain unknown and how to run a production-minded pilot.

Kevin Riedl
Read
PageLM document learning platform: self-hosting, licensing and data-flow review AI & Agents

PageLM Review: Self-Hosting and Commercial Use

A source-checked PageLM review for teams: what it generates, where data goes, what commercial use requires and how to test a training pilot.

Kevin Riedl
Read
Shielded sign-in form passing credentials directly to ChatGPT's remote cloud browser AI & Agents

ChatGPT Can Now Log In Without Seeing Your Password

ChatGPT Work can sign in to supported websites without exposing credentials to the model. Here is what the session can still access and how to pilot it safely.

Kevin Riedl
Read
M6 Mac mini, M5 Pro Mac mini and M5 Mac Studio arranged by local AI memory capacity AI & Agents

M6 Mac mini vs M5 Mac Studio for Local AI: Buyer’s Guide

Choose between the M6 Mac mini, M5 Pro Mac mini, M5 Max Mac Studio and M5 Ultra Mac Studio using memory headroom, bandwidth, concurrency and cost per accepted AI task.

Kevin Riedl
Read
AI agent harness architecture connecting context, policy, tools, verification and observability around an LLM AI & Agents

AI Agent Harness, Explained: The Reliability Layer Around an LLM

A visual architecture, failure map and buyer checklist for the context, policy, tools, verification and observability that make an AI agent dependable.

Kevin Riedl
Read
FreeToken routes MoE experts between consumer GPU VRAM, CPU and system RAM AI & Agents

FreeToken AI Review: Run Frontier MoE Models on Consumer GPUs?

FreeToken turns GPU, CPU and RAM into one MoE inference system. Check its hardware requirements, author-reported benchmarks, setup, limits and production fit.

Kevin Riedl
Read
Darkbloom AI routes encrypted inference between an application, a confidential coordinator and verified Apple Silicon Macs AI & Agents

Darkbloom AI Review: Private Inference on Idle Macs

A buyer-focused review of Darkbloom’s privacy boundary, OpenAI-compatible API, public-alpha pricing, Mac provider economics and production risks.

Kevin Riedl
Read
CRA incident timeline with 24-hour, 72-hour, and final-report checkpoints Business & Regulation

Cyber Resilience Act Reporting: The 24/72-Hour Operational Playbook

Operational CRA reporting: 24/72-hour deadlines, distinct final reports for exploited vulnerabilities and severe incidents, ownership and submission evidence.

Kevin Riedl
Read
Inbox, without the noise

Get the next field note

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