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.

Connected-product API architecture with device, user, policy, and export layers Delivery & QA

EU Data Act: Connected-Product API Checklist

Prepare the Article 3 access-by-design duty for products placed on the market after 12 September 2026, while the Data Act already applies generally.

Kevin Riedl
Read
Ox Alpha revealed as Z.ai GLM-5.3-Flash with updated access, privacy and cost checks AI & Agents

Ox Alpha Revealed as GLM-5.3-Flash: What Changed

Ox Alpha was revealed as Z.ai GLM-5.3-Flash. See what changed in access, pricing, privacy, model identity and production planning.

Kevin Riedl
Read
Rust compiler graph connecting persistent declaration identities to verified semantic patches AI & Agents

Why Agent Edits Need Semantic Identity: Building SEMAPRAX in Rust

A compiler engineering walkthrough of persistent IDs, checked HIR, deterministic graphs and replayable patches in the experimental SEMAPRAX language.

Kevin Riedl
Read
LangChain Deep Agents connects planning, subagents, files, approvals, sandboxes and production traces AI & Agents

LangChain Deep Agents Review: Is the Agent Harness Ready for Production?

A buyer-focused Deep Agents review covering architecture, v0.7 changes, security boundaries, operating costs and a measurable production pilot.

Kevin Riedl
Read
Cursor Origin and GitHub repositories connected through a reversible code-hosting pilot Delivery & QA

Cursor Origin vs GitHub: Should Your Team Switch?

Cursor now hosts code, but Origin is still an early beta. Compare sync behavior, CI, governance, price and a low-risk pilot before moving your source of truth.

Kevin Riedl
Read
Four Pika Audio API models mapped to sound effects, speech, music and synchronized video soundtracks with cost controls AI & Agents

Pika Audio Models API Pricing: Is SFX Really 20x Cheaper?

Pika lists SFX at $0.0002 per second, plus three more audio models. See the real break-even, API limits, quality gates and production fit before you switch.

Kevin Riedl
Read
GPU workloads sharing a pooled accelerator fleet through a virtualization layer AI & Agents

Thunder Compute's $13M GPU Virtualization Bet: Enterprise Buyer's Guide

Thunder Compute says GPU pooling can recover stranded capacity. Compare network virtualization with MIG, vGPU and passthrough, then scope a measurable enterprise pilot.

Kevin Riedl
Read
OpenViking filesystem connecting agent memory, RAG resources and skills through viking paths AI & Agents

OpenViking Review: Filesystem Memory for AI Agents

How OpenViking uses a virtual filesystem and layered summaries for agent memory, with checked LoCoMo results, token-saving caveats and pilot criteria.

Kevin Riedl
Read
Continuous verifier scores ranking several AI agent trajectories AI & Agents

LLM-as-a-Verifier Explained: Architecture, Costs, and Production Fit

A technical buyer guide to fine-grained LLM verification: how logprob scoring and pivot tournaments work, what the benchmarks prove, and how to pilot it safely.

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.