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.

Parcel candidates connected to structured evidence in a Wavect illustration Delivery & QA

Mireye Land Search API: MCP, Pricing and Parcel Screening

Search vs Enrich vs free Fetch: a bounded U.S. parcel workflow with current MCP setup, a $1.40 cost example and checks for sources, missing data and parcel identity.

Trace branches pass a sampling gate beside a clock in an original Wavect illustration Delivery & QA

LangSmith Trace Retention: Sampling, Evals and Cost

Separate trace sampling, evaluation coverage and retention upgrades. Current rules, a worked cost example, spend-cap behavior and dataset lifecycle checks.

Parallel changes joining a repository branch beside a separate verification path Delivery & QA

git-ai Review: Concurrent Agent Merges and Verification

What gitai checks before an AI conflict resolution lands, what graft-verify checks afterward, and how to evaluate concurrent agent merges without overstating the benchmarks.

PHP2Shell security shield with paths to supported PHP or a staged migration to another runtime Delivery & QA

PHP2Shell: Upgrade Legacy PHP or Migrate to Node.js?

Stuck on PHP 7.4 or 8.1? Compare verified backports, a supported PHP upgrade and a staged Node.js migration, with practical checks before production.

A shield representing software delivery with explicit quality assurance Delivery & QA

Software Agencies Austria: A 2027 Development and QA Guide

Compare Austrian software agencies for your 2027 release. Check published development and QA scope, acceptance evidence, operating costs and handover.

Connected layers representing a specialist team working within an enterprise delivery programme Delivery & QA

Enterprise Software: Adding a Specialist Alongside Your System Integrator

Bring a specialist into an existing enterprise programme. A worked brief clarifies the contribution, interfaces, decisions, acceptance and handover.

Layered harness frames around a fixed model, illustrating recursive improvement rather than model retraining Business & Regulation

Google Harness²: Setup, Benchmarks and the Real Cost

Better agent results without retraining the model? Read what Google Harness² changes, what its benchmarks prove, and why a safe pilot needs a cost and acceptance boundary.

Separate GPU capacity and software verification layers, with a shared fleet below protected release gates. Delivery & QA

NVIDIA Run:ai Open Source: GPU Fleet vs Software Factory

GPU orchestration just became easier to inspect and adopt. Compare Run:ai modules, Community Product terms and the missing task-to-patch acceptance layer.

Stacked document shapes illustrating an Ultramarine 1 Flash document-agent evaluation Delivery & QA

Ultramarine 1 Flash for document agents: pricing, API limits and a migration test

The 60x claim gets attention. Document fidelity, API compatibility and human review decide whether switching actually pays.

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 →