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.

A comparison of time-based billing and accepted software results in Wavect's editorial design Business & Regulation

AI Agent Hours + Human Hours: Why Clients Should Push Back

AI makes delivery faster. Should it also create a second clock on your invoice? The client case for accepted results, clear budgets and accountable software delivery.

Two overlapping waveforms illustrate conversational timing in voice AI AI & Agents

Cue Turn Detection: Phone Numbers, Benchmarks and Limits

A pause is not a finished phone number. What Pype’s Cue v1 improves, what the benchmarks measure, and how to test correct, confirmed capture.

Wavect editorial illustration for RTK 1.0 token filtering and evaluation AI & Agents

RTK AI 1.2: Real Token Savings, Setup and Licensing

RTK 1.2.0 improves how partial reads are counted and preserves more nextest failure detail. Check the license split, agent setup and proxy routes before measuring savings.

Stacked computing layers illustrating a model distributed across devices Business & Regulation

Anthropic’s self-replicating AI warning: the local AI takeaway

A security warning with an unexpected local-AI lesson. Separate the three experiments, check the speed arithmetic and make ownership testable.

Wavect editorial header for Agent Mesh, with a shield representing controlled access to tools AI & Agents

Google SAM is now Agent Mesh: MCP tools across machines

Connect approved tools across machines. Understand the control plane, signed EU labels and the checks a useful first pilot should pass.

Layered turquoise geometry representing an Anatomy illustration assembled from editable parts AI & Agents

Anatomy for Claude Code: Interactive SVGs for Documentation

Turn a technical explanation into a figure readers can explore. Install Anatomy, choose HTML or React, and check the model before embedding it in your docs.

Layered document cards representing EdgeQuake's PostgreSQL knowledge store AI & Agents

EdgeQuake: PostgreSQL GraphRAG for MCP agent memory

Connect document relationships to your agents. A v0.40.12 review of PostgreSQL deployment, MCP access, local model roles and the LightRAG benchmark.

Three ternary weight values, minus one, zero and plus one, connected to a CPU AI & Agents

BitNet on CPU: 2B4T, real RAM needs and the 100B claim

Which BitNet model can you download, and what fits your CPU? Separate the 100B benchmark, 0.4 GB metric and 1.19 GB file, then test a real task.

Conflicting policy documents converge on an owned decision and a checked business rule Product & MVP

Enterprise software: settle the business rules before AI writes the code

A refund looks like one if statement until sales, finance and support disagree. Resolve the rule, its owner and its boundary cases before automating it.

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 →