We write
What we keepfinding out.
Working notes on 3D, generative AI and development. With the sources in plain sight.
-
How to take an AI prototype to production without rebuilding everything
Review requirements, code, dependencies, security, licensing, tests and maintenance before turning an AI-generated prototype into a production system.
-
Is WebGPU supported in Chrome, Safari and Firefox?
See where WebGPU works in Chrome, Safari and Firefox, why an adapter may still be unavailable and what to test before removing a WebGL viewer.
-
AGENTS.md and Skills: documenting a project for AI agents
Learn what belongs in AGENTS.md, what to move into Skills and how to test project instructions without accumulating conflicting or outdated rules.
-
Astra in Blender and Unreal: what the 3D workflow demonstrates
Understand Astra's Blender and Unreal example: scripts, scene transfer and visual review. Plan a bounded test of automation in your own 3D pipeline.
-
Blender 5.2 LTS: planning a studio migration
Plan a Blender 5.2 LTS migration by testing scenes, scripts, materials and exports before moving production. Keep a verified route back.
-
Blender and AI: project policy, connectors and asset rights
Separate Blender's AI policy from experimental connectors and asset licensing. Practical considerations for studios introducing automation.
-
Zero-click searches and AI: adapting your website content
Read zero-click and AI-summary research in context. Diagnose your own search traffic and improve useful content without relying on guaranteed GEO claims.
-
Product 3D models: preparing assets for multiple channels
Prepare product models for Shopify, Google, Amazon and Safari. Check format support, eligibility and destination-specific tests before reusing an asset.
-
AGI and AI benchmarks: what the results actually tell you
Learn how to interpret AI benchmarks, check the conditions behind a score and build a practical evaluation before adopting a model for your business.
-
The Fable 5 suspension: planning for an AI service interruption
What the Fable 5 suspension teaches about continuity: preserve task state, test alternative models and decide when a manual process is enough.
-
AI 3D generation: what quads, 4K and UVs actually mean
Understand quad topology, 4K resolution and UV quality in AI-generated 3D models. Learn what to inspect before choosing a generator for production.
-
AI 3D generators in Spain: how to review the licence
Check territory, commercial use, reference image rights, plan conditions and client delivery before using an AI 3D generator for a project in Spain.
-
GPT-6 Astra: what to check before using it in a project
Read the GPT-6 Astra specifications and assess tools, context, permissions and real task performance before integrating the model into a project.
-
The Hugging Face incident: security lessons for AI workflows
Lessons from the July 2026 Hugging Face incident: handling external files, limiting credentials, isolating execution and preparing an incident response.
-
AI and copyright: what to document before delivering creative work
Separate authorship, commercial permission and third-party rights in AI-assisted work. Learn what to document and why Spanish and US rules differ.
-
Spain’s AI bill: current status and proposed penalties
The status of Spain’s AI bill on 20 September 2026. Understand the proposed penalties and distinguish draft national rules from EU AI Act obligations.
-
Article 50: how to mark and disclose AI-generated content
Understand technical marking, public disclosures and C2PA credentials. What to check when delivering AI-generated images, video or text in the EU.
-
What is MCP? Connecting an AI agent to your tools
Understand Model Context Protocol and the version, permission and security checks needed before connecting an AI agent to your business systems.
-
What is an agent harness, and why does it affect results?
A model is only part of an AI agent. Learn how context, tools, permissions, state and error handling affect results and how to compare integrations.
-
How much does an AI agent cost? Estimating the full workflow
Understand AI agent costs across development, platforms, usage, review and maintenance. Compare proposals against the same task and acceptance criteria.
-
Safari 27 and the model element: what it adds to web 3D
Explore Safari 27’s model element on iPhone, iPad and Mac. Plan format checks, accessible controls and a compatible fallback for your product pages.
-
Sora API shutdown: how to prepare a replacement workflow
The Sora API is scheduled to close on 24 September 2026. Learn what to preserve and how to evaluate a video alternative by quality, licensing and delivery.
-
Spec-driven development: when it helps AI-assisted coding
Understand spec-driven development, how Kiro and Spec Kit structure it, and what to measure when testing whether specifications improve AI-assisted coding.
-
Gaussian splatting for a web showroom: what to evaluate
When to use gaussian splatting for a real space on the web. Assess capture quality, formats, navigation and mobile performance before choosing a viewer.
-
Generative video in EXR: how to assess it for compositing
Learn how to inspect generative video delivered as OpenEXR: colour space, bit depth, alpha, compression and tests before bringing it into a VFX composite.
-
Web compatibility in 2026: how to decide if you need a polyfill
Use Baseline and Interop to check support across Chrome, Safari and Firefox and decide when a web feature still needs a fallback or polyfill.
-
How to choose a platform for a virtual or immersive showroom
Compare 3D websites, WebXR and headset apps before building a virtual showroom. Choose according to audience, devices, maintenance and reach.
-
Online store accessibility in Spain: what to check
Understand accessibility requirements for online stores in Spain, the microenterprise exemption and practical checks across the complete checkout journey.
-
AI agents: how to decide what you can automate
Assess an AI agent before automating a process: repeated tests, limited permissions, human oversight and metrics for errors and review effort.
-
ComfyUI security: reviewing nodes and protecting your work
A practical ComfyUI security guide for studios: review custom nodes, restrict network access and test updates without exposing client files or credentials.
-
3D configurators: when they help and how to measure it
Decide whether your store needs a 3D configurator, assess case studies and plan a pilot measuring sales, returns and mobile performance.
-
Hunyuan3D in the EU: what its licence allows
Understand Hunyuan3D community licence restrictions in the EU, how they address generated outputs and what to check before delivering a 3D asset.
-
AI-generated 3D models: how to check production readiness
Review AI-generated 3D models for shape, topology, UVs, materials and performance. Decide what needs fixing before an asset enters production.
-
3D art outsourcing: plan the brief, review and handover
Plan a 3D art outsourcing project with clear requirements, a representative pilot and acceptance checks, including your team's coordination effort.
-
How many polygons should a 3D model have?
Set a useful polygon budget for your platform, camera and scene. Consider frame time, materials, textures and measurable asset acceptance criteria.
-
EU AI Act in 2026: deadlines and obligations to check
Review the EU AI Act timetable after the July 2026 amendment: high-risk systems, transparency and practical questions for creative and digital projects.
-
Choosing a web stack you can maintain
Choose a web stack with support deadlines, upgrades and dependencies in mind. Check Node.js and PHP dates and the published Next.js support policy.
-
Mobile 3D viewers: limits, formats and performance checks
Prepare a 3D model for mobile: separate limits from recommendations, validate GLB and USDZ, and test loading, interaction and augmented reality.