Field notes
Production AI, written plainly.
Notes on the unglamorous parts of AI systems: data contracts, runtimes, review loops, compute limits, and the engineering habits that keep them useful after the demo.
The Agentic Trap: Why AI Systems Need Fewer Paths and More Truthful State
The fastest way to make production AI fail is to let execution paths disagree.
The Runtime Boundary Is the Product: Why AI Workflows Fail Outside the Model
Whether the surrounding runtime can make the model usable, observable, recoverable, and repeatable.
The Critic Needs a Contract: Why AI Review Systems Need Profiles, Schemas, and Evidence
A useful AI reviewer needs versioned behavior, constrained output, evidence, and failure metadata.
From Human Taste to Training Data: Making High-Volume AI Output Reviewable
Human review only scales when the system can learn how to filter.
Scarce Compute Is a Product Constraint: Designing AI Workflows Around One GPU
A single GPU changes what can run together, what must wait, and what the system must admit.
Running Qwen3.6 27B INT4 on a 5090
A field note on local VLM critic runtime, quantization, startup, and structured output.
Benchmark Before You Optimize: Turning AI Generation Speed Into a Measurable Loop
AI generation speed becomes tractable when the team measures workflow phases.
Prompt Tokens Are Product Budget: Why AI Image Prompts Need Measurement, Not Vibes
Prompt length is the budget available for product decisions and constraints.
Manifests Beat Memory: Making AI Generation Runs Resumable and Auditable
Long-running AI work should depend on manifests the system can inspect, resume, and verify.
When Humans Can Move Files: Why AI Image Pipelines Need One Canonical Storage Model
The fix for ordinary storage failure is one clear source of truth.
The Trigger Token Is a Contract: Avoiding Silent LoRA/Base Model Mixups
A LoRA run is a contract between model, weights, trigger token, runtime, and review expectations.
When a LoRA Looks Worse: Debugging Training Quality Before Blaming the Model
Treat a bad LoRA output like an investigation, not a verdict.
Do Not Train on Chaos: Why LoRA Datasets Need Copy-Only Data Prep
LoRA quality begins before training starts, with data prep the team can still explain.
The Remote Review Trap: When a Portal Becomes a Second Product
Remote review should be a narrow surface over canonical data, not a second product.