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.

Graphify codebase knowledge graph connecting code, data, infrastructure and documentation AI & Agents

Graphify Review 2026: Is a Codebase Knowledge Graph Worth It?

Buyer review of Graphify vs search and RAG: architecture, privacy boundaries, benchmark limits, real adoption cost and a two-week pilot scorecard.

Kevin Riedl
Read
Bonsai 27B 1-bit local AI model compressed from 54 GB to a phone-class footprint AI & Agents

Bonsai 27B Review: Can a 27B LLM Really Run on a Phone?

Verified buyer's review of 1-bit vs ternary, real memory, uneven benchmark losses, phone and WebGPU speed, use cases and pilot gates.

Kevin Riedl
Read
Soofi S European sovereign LLM procurement review AI & Agents

Soofi S: Is Germany's Sovereign LLM Ready for Business?

Germany's 31.6B sparse model is strong on German and code, but the current checkpoint is a selected-partner closed beta with no final license. Benchmarks, openness, infrastructure, alternatives and the pilot checklist EU buyers need.

Kevin Riedl
Read
AI pilot kill-or-scale scorecard with 12 business, quality, reliability and adoption metrics AI & Agents

AI Pilot Kill-or-Scale Scorecard: 12 Metrics to Check After 30 Days

A decision-grade scorecard with formulas, hard gates and one worked example across baseline cost, successful actions, straight-through completion, correction time, failures, latency, adoption, auditability, data readiness and payback.

Kevin Riedl
Read
T3MP3ST autonomous AI red-team harness reviewed against OWASP APTS AI & Agents

T3MP3ST Review 2026: Can It Replace a Penetration Test?

A buyer-focused review separating benchmarked single-agent capabilities from the unproven swarm, mapped against OWASP APTS with a safe pilot decision for CTOs.

Kevin Riedl
Read
Vibe coder and junior developer taking different paths through an AI-assisted engineering system Leadership & Teams

Are Vibe Coders the New Junior Developers?

The traditional ticket-taking junior role is under pressure, but the junior developer is not extinct. Current labour data, the difference between prompting and engineering, and a hiring model for AI-native entry-level talent.

Kevin Riedl
Read
External QA benchmark for software defects found in the first 30 days Delivery & QA

External QA Benchmark: What We Find in the First 30 Days

A research-backed defect benchmark for permissions, regressions, browser/device combinations, data integrity, AI failures and production escapes, and why no honest universal bug-count median exists.

Christof Jori
Read
Programming languages converging into runtime, memory and architecture decisions Delivery & QA

Programming Languages Matter Less. Software Engineering Matters More.

AI makes syntax, boilerplate and cross-language translation cheap. It does not erase runtime, memory, concurrency, security or maintenance trade-offs. What CTOs and engineers should optimise for now.

Kevin Riedl
Read
Colibri streams GLM-5.2 experts from NVMe through RAM and VRAM on consumer hardware AI & Agents

Colibri Runs GLM-5.2 on Consumer Hardware. Here Is the Catch.

Colibri now supports nine engine families and Brio closed-set scoring. Updated hardware requirements, the 372 GB versus 429 GB download discrepancy, real benchmark limits, GLM-5.3 support and a reproducible local API setup.

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.