RESEARCH · 11
Research& field notes.
An Autonomous AI Agent in Digital Production — Architecture, Cases, and Lessons
A case study of a single autonomous AI agent operating as a producer-coordinator inside a digital production studio over March–May 2026. Three concurrent projects, 3000+ messages in the largest, 95 hiring candidates processed, one critical mistake, the architectural decisions that worked, and the ones that didn't.
Hyperfocus Engineering
A prompt architecture for on-demand executive function in ADHD / neurodivergent operators. Six composable skills built from ten executive-function primitives, auto-triggered by behavioral signal detection rather than slash invocation.
Agentic Generation of Music-Processor Templates
Letting an LLM design and apply concrete multi-effects patches to a hardware guitar processor from plain-English prompts. A three-day hardware-in-the-loop study: no training, no dataset, ~20 musically coherent patches validated on metal.
A.O.C. — A Prompt Framework for Generative Image Models
Anchor · Optics · Chemistry. A structured prompt framework that reduces every visual decision a shot requires to three orthogonal axes corresponding to the three independently-controllable stages of real photographic production.
Deliberate Omission
When silence in a prompt is the feature. For multi-channel generation, saying more in text actively degrades output when the text and the other channels describe the same dimensions.
A Generative-AI Video Production Pipeline
From client brief to finished shot, stage by stage. The thesis: generative AI video at production quality is a constraint-propagation pipeline, not a generation model. Six stages, three consistency mechanisms, a dependency graph that lets the work parallelise.
The Lock Layer
Identity and brand preservation as a separable prompt architecture. The generative prompt describes this shot; the Lock Layer wraps it with an immutable structured constraint, replayed verbatim across every generation in a batch.
Sequential Consistency via Prompt Architecture
Holding coherence across multi-shot and multi-scene generative AI without training. Four composable prompt architectures — reference-inheritance, series-aware anti-repetition, neighbor-aware prompts, reference condensation — that together replace per-subject fine-tuning for one-off productions.
The Two-Stage Architect Pattern
Separating creative intent from technical generation in LLM pipelines. Different temperatures want different tasks; a single prompt asking for both gets neither.
Virtual Try-On for Eyewear
AI-driven virtual glasses try-on system combining LoRA fine-tuning, ControlNet inpainting, and latent space processing. Solved transparent lens rendering and eye identity preservation.
Neural Render — AI in 3D Pipeline
Blender add-on integrating AI models as transparent post-processing nodes in the render pipeline. Upscaling, style transfer, and creative transformations directly in the 3D workflow.