Every case study, system breakdown, and field note, newest first and filterable by topic. No theory, no demos. Only systems that run in production and what it took to get them there.
This is the complete archive: all 819 posts across 65 topics, newest first, filterable by category. It holds the case studies, the model and tool comparisons, and the field notes from systems I built and run. For the curated view, where the writing sits alongside the free Claude Code skills, start at the Library instead.
The archive leans in four directions: AI Search Visibility (139), AI Agent Architecture (73), AI News (48), and AI Business Ideas (46). Those four account for most of what I publish, because they are where most of the client questions land.
Three places to start. The AI search visibility guide is the hub for the largest cluster and links out to every spoke in it. Grok Imagine vs Midjourney is the image model comparison, rebuilt against live leaderboard data rather than left to go stale. OutreachIQ is the longest running system breakdown here, from first prototype through to a public repo.
Jack Roberts demonstrates an abliterated API, local Ollama models and an always-on VPS agent. Follow the real data and hardware trade-offs, with an official-docs setup and all chapters.
World of AI shows striking Qwen 4 preview demos. Separate Alibaba's announcement from anonymous checkpoints and leaks, then follow a verified local Qwen setup you can use today.
Tech With Tim explains local inference and demonstrates LM Studio, Ollama, Docker Model Runner and Python. Use this beginner guide to choose a route, try a small model and understand the real privacy and hardware limits.
David Ondrej explains abliterated AI models and three ways to run them. Learn what weight edits change, how his demo differs from V4.1, and follow a source-checked guide to downloading DeepSeek V4.1 Flash 1017.
xCreate runs DeepSeek V4.1 Flash and Qwen3.8 Flash Next on a 64 GB M1 Max using Inferencer streaming. Compare speed, memory and disk across three Macs, plus what the demo does not prove.
Nick Saraev explains why companies test Chinese AI models. We check the Coinbase, Airbnb and OpenRouter evidence, compare hosted and self-hosted options, and offer a practical model-selection test.
NetworkChuck tests a sponsored WiCi One engineering sample with DeepSeek V4.1 Flash. Here is how GPU sharing and SSD-backed inference work, what the speed numbers mean, and what buyers should check.
Matthew Berman reads Google's Gemini 4 Carbon launch. Compare independent and Google-reported scores, introductory versus standard prices, safety claims and the limited rollout.
David Ondrej tests Kimi K3.1 Pro with Blender and FreeCAD, then asks what its speed and price mean for agent loops. Separate his demo, Moonshot benchmarks, cyber controls and the influencer allegation.
Matthew Berman tests DeepSeek V4.1 Flash 1017 on 3D, coding and vision. Here is what the demos show, what DeepSeek confirms, and the realistic web, API and local setup routes.
NetworkChuck tests TypeSafe Jev on spam comments, model routing, a live teleprompter and browser automation. Learn what the decision model actually does, what the surrounding code must do, and how to evaluate it fairly.
Pat Simmons compares Kimi K3.1 Pro with Opus 5.5, Fable 5.1, GPT-6 Astra and DeepSeek V4.1 Flash on a billiards game, website clone and launch video. See the prompts, published costs and key limitations.
Duncan Rogoff explains GPT-6.1 Terra. Check OpenAI's computer-use benchmarks, the $1/$5 API rates versus Claude Opus 5.5, actual access routes, and the safeguards a first agent test needs.
Prompt Engineering gave Kimi K3.1 Pro and Claude Opus 5.5 the same 13 long-horizon tasks. Compare research, 3D, Blender, front-end quality, runtime and the real $45.91 vs $83.46 bill.
World of AI tests Kimi K3.1 Pro on web design, browser games, 3D scenes, Blender, research and computer use. See what the demos show, where Opus still leads, and the verified API price difference.
Matthew Berman tests Kimi K3.1 Pro on a Rubik's Cube, WebGL phone, Blender knot, motion graphic, game and website. Compare visible results with Kimi's release claims, independent scores, prices and weight availability.
Pat Simmons gives Gemini 4 Argon, Opus 5.5, Fable 5.1, and GPT-6 Astra the same four creative builds. Compare the live results, blind rankings, time, and API-equivalent cost without mistaking them for a general benchmark.
Paul J Lipsky tests Gemini 4 Argon on games, a wearable-controlled prototype, code bugs, research, a scary story, and video editing. Here is what worked, what needed help, and where access differs.
Matt Wolfe tests Gemini 4 Argon on coding, interactive physics, scheduling, and a 3D print. Compare the visible results with Google's launch claims, current access, API prices, and stalled Tasks.
Matthew Berman tests Meta Muse Wonder and Wonder Full on a Rubik's Cube, a game, and 3D scenes. Compare the actual demos, early quota trouble, official access and prices, and independent benchmark context.
RandomAI surveys alleged Fable 5.5 animations, games, and physics demos. Here is what the clips show, why a knowledge-cutoff test cannot prove model identity, and how to compare the work fairly.
Samson compares Opus 5.5 with GPT-6 Astra, Sol, and Luna through games, interactive 3D, animation, and agent workflows. Explore all 77 linked examples with creator credits and original sources.
Matthew Berman tests eight Jev use cases across browser cleanup, email priority, semantic search, dynamic UI, video clipping, color palettes, and emoji selection. This guide separates the model decision from the software around it.
Peter stress-tests Claude Opus 5.5 and GPT-6 Sol on ambitious 3D and animated HTML builds. The results reveal a quality ceiling model and an efficient iteration model, not a simple winner.
Every post here is about a system that actually shipped. Book a free call and let's talk about what could ship for you.
Book Free 30-min CallAdd your name and email to unlock recording.
Your recording is emailed to me and kept on this site's server for 12 months, then deleted. Privacy policy.
Thanks for your voice message. I listen to every one myself and I will get back to you within one business day.
Explore João's services and public work, or arrange a free 30-minute consultation.
Choose a time with Cal.com. Your booking status is shown there. No calendar details are shared with this assistant.
Open booking in a new tabWhen enabled, each question is processed by Cloudflare AI and checked for abuse with Turnstile. No chat history is stored by this app. Please leave out private information.
Prepared answers · Public information