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.

Geometric code core and four tool nodes inside layered runtime frames on a dark Wavect header AI & Agents

OpenAI Agents API Review: Migration, Costs and Data Controls

What the managed Codex harness replaces, what stays your responsibility, and how to test the launch numbers before migrating a real workflow.

Large source files go to a worker model while only focused summaries return to Claude; reasoning stays separate AI & Agents

Spotify shunt Review: Setup, Savings and Limits

What Spotify’s 90% claim actually measures, what Portal setup requires, and how to test file-read delegation without moving the cost into rework.

Three AI coworker profiles connect to a policy and audit gateway before using browser, file and MCP tools; conversation storage is shown separately AI & Agents

OpenBot Review: Self-Hosted AI Coworkers, Costs & Controls

A source-reviewed guide to CopilotKit OpenBot: where its action gateway helps, which defaults need changing, and when self-hosting beats buying a managed assistant.

A phone and laptop split one large language model across browser tabs while WebRTC carries activations between them AI & Agents

SwarmLLM Review 2026: Browser P2P LLM Inference Across Phones and Laptops

A buyer-focused review of SwarmLLM: browser-native 27B inference across phones and laptops, WebGPU/WebRTC architecture, benchmark limits, peer trust, and when to pilot it.

A background coding agent receives a prompt, restores a prepared development snapshot, works inside an isolated sandbox, verifies the change and opens a pull request AI & Agents

Ramp Inspect Architecture 2026: Background Coding Agents at Scale

How Ramp Inspect uses full-stack Modal sandboxes, filesystem snapshots, coordination and verification, plus the controls a regulated engineering team needs before scaling background coding agents.

AI agent connecting through a standardized interface to programmable lab and factory hardware AI & Agents

Model Hardware Standard: Enterprise Guide to Physical AI

Evaluate Anthropic’s MHS for lab and factory automation: architecture, MHS vs MCP, safety boundaries, pilot economics and a practical adoption checklist.

A voice AI phone platform connected to SIP, business APIs, and a build versus buy decision matrix AI & Agents

Fonio AI Review 2026: Pricing, API, GDPR & Build vs Buy

A buyer-focused Fonio AI review with current phone pricing, SIP and API integration, EU recording duties, operational limits, and a practical build-versus-buy scorecard.

Several coding-agent session timelines converge into one shared team context store AI & Agents

Mosaic (YC S26) Review: Shared Memory for Team AI Agents

Mosaic centralizes coding-agent sessions across tools and teammates. See how shared session memory differs from Claude Code Agent Teams, agent memory, Git, and a real team knowledge layer.

A coding agent narrows a repository through a deterministic ranked graph before opening source files AI & Agents

Ripwire Review 2026: AI Repo Context Without Embeddings?

A buyer-focused review of Red Hat ET’s Ripwire: deterministic repository context, ranked call graphs, token claims, CLI vs MCP, benchmark limits, and when it beats grep, Graft or Graphify.

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 →