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.

YC application technical readiness checklist with demo, metrics and founder ownership checks Product & MVP

YC Application Technical Readiness: What to Build Before You Apply

YC does not require an MVP, but a working demo, real user evidence and founder-owned technical progress can sharpen your application. Use this checklist to build proof instead of polish.

A stateful MCP session architecture changes into stateless requests that can reach any server instance behind a load balancer AI & Agents

Is MCP Stateless Now? Your Server Migration Checklist

MCP 2026-07-28 removes protocol sessions and the initialize handshake. See what stateless really means, what still keeps state and what to change in your own MCP server.

An AI agent reaches customer data through MCP and an API, while policy checks at the record and field level form the real security boundary AI & Agents

MCP Is Not a Security Boundary: Protect Agent Data

MCP can narrow an agent’s tool interface, but it cannot protect a customer record the agent can reach another way. A practical model for identity, egress, object-level policy, field protection and audit.

Gemma 4 model flowing through an Unsloth fine-tuning stack in a browser notebook AI & Agents

Fine-Tune Gemma 4 Free with Unsloth and Colab

A free browser pilot is real. This guide shows the exact workflow, the limits behind the hype, and what changes before a fine-tuned model can ship.

Two independently owned personal AI agents coordinate a calendar request through a permission-controlled Band room AI & Agents

Can AI Agents Talk to Each Other? A Band Setup Guide

Band lets independently owned agents find, message and coordinate with each other. See how the setup works, where A2A and MCP fit, and which identity, approval and privacy controls a real pilot needs.

B2B SaaS validation scorecard for DACH procurement, privacy and paid pilot evidence Product & MVP

B2B Product Validation in DACH

Validate demand before funding the full build. Test buyer access, procurement, GDPR evidence, paid pilots and slow sales-cycle signals.

A fixed LLM weight grid etched beside an inference ASIC AI & Agents

Taalas HC1 Review: Is a Hardwired LLM ASIC Worth It?

Taalas reports 17,000 tokens/s from Llama 3.1 8B in silicon. Here is what the benchmark proves, what the hype misses, and when model-specific hardware could pay.

LeanCTX agency experience measured across coding-agent workflows AI & Agents

LeanCTX Technical Field Report: 64.1% Less Context

Our technical agency report from real repositories: integration mechanics, 64.1% overall context reduction, raw fallbacks and quality gates.

Hardware memory blocks ranked against local LLM model sizes AI & Agents

llmfit Guide: Which Local LLM Fits Your Hardware?

Use llmfit to rank local models before downloading them. Learn what its fit, speed, quality and context scores mean, where estimates can mislead, and how to turn a recommendation into a production decision.

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 →