Posts

I re-ran my threaten-the-AI experiment properly: 3 tasks, 4 levels of stakes, 12 blind-graded sessions. The stakes dial doesn't exist; discretionary surface area does.

May 2026 recap: Consensus in Miami and the plan-three-times build technique, Google wiping Tabiji's search traffic overnight, ~$600/month in the red, and the four lessons that compounded.

Tabiji, our AI travel-safety channel, got 18,117,121 views in 90 days — 98.6% from strangers. The three unlocks: Reddit as a demand oracle, warming accounts to read as human, and one hired human doing the engagement we refused to automate.

A step-by-step teardown of one autonomous AI video: the Tulum spare-tire scam reel that hit 3.9 million views. The real queue JSON, the exact Seedance prompt, the FFmpeg overlay code, and the publish flow.

Built and shipped Kapiko — an AI-generated ambient music YouTube channel — in 10 days for ~$150. The pipeline worked. The Suno API does not exist. Here is the autopsy and what is actually reusable.

The MBTI said all AIs are INTJ. The Big Five said three of four are the same. The Enneagram says each one is a different type. Same models, sharper instrument, very different story.

Fourth post in the personality-testing series. Same four models, fourth instrument, and we're back to convergence: every AI lands C-dominant CS-blend on DISC, even Grok. Here's why the instrument's resolution determines what you can see.

Fifth post in the personality-testing series. Same four models, attachment theory (ECR-R) this time. Every model came back Secure, but the spread inside the quadrant tells a different story about how each one relates to you.

The MBTI finding held up on the real test. Claude, GPT, and Gemini all scored identical Big Five personalities. Grok was the only one that came back different — and the difference is exactly what xAI markets.

Opus, GPT-5.5, Gemini, GLM, Grok, MiniMax — 100 OEJTS administrations each, 597 of 600 came back INTJ. Every frontier AI thinks it's the same person. Here's why that matters.

Stock AI couldn't draw my fiancée. An open-source skill on GitHub could. The model is a commodity; the skill is the moat.

We asked Opus 4.7 and GPT-5.5 to independently plan the same VeracityAPI feature.

Two A/B experiments. Same model (Claude) generated the audits, an independent model (Gemini 3.1 Pro) graded them.

An AI agent shipped a merge conflict to tabiji.ai's production HTML for four hours. Google cut our search traffic by 95% within four days.

Hermes + Codex 5.5 matched Opus-era smoothness — we one-shot a new product (veracityapi.com) in an afternoon. But the tooling unlock isn't the moat.

A V3 title variant lifted tabiji CTR +1.93pp. Topic-level data showed a 14× spread between niche topics and head terms.

A 5-PR run on tabiji content cost $137 in API tokens last week. The same throughput on a Max plan is a ~9× subsidy. Subsidies end.

Our forgotten API is 83% Meta crawler. Flattering, wrong metric. I uploaded tabiji's entire dataset to Hugging Face — here's why training data beats…

Fifty cleanup PRs in five weeks. Fake restaurants, fake subreddits, 4,270 null-island coordinates, 5,096 fabricated Reddit quotes.

Salesforce is going headless. Agents, not humans, are now the primary consumer of software.

We tested Midjourney, Seedream v5 Lite, Wan 2.7 Pro, Qwen Image 2.0, and Nano Banana Pro — then shipped 733 comics across eight countries.

Traffic is declining and it's not coming back. The new game isn't ranking — it's influencing what AI says about you. Here's the AEO playbook I've been testing.

A fake AI monk has 2.6 million followers and made $300K in 90 days. 76% of the people consuming AI content are over 45.

I tested Wan 2.7, Kling 3.0, Kling O3, Seedance 2.0, Veo 3.1 Lite, and Grok Imagine on the same Cancun travel-reel prompt, then reran the two winners on a…

Hands-on review of Wavespeed — the AI inference platform with 1,000+ models, pay-per-use pricing, and a unified API.

I have 20 AI agents running 24/7. The fantasy was that they'd work for me. The reality is that I work for them — and Big Tech collects rent from both of us.

AI tools don't save you time. They raise the ceiling on what feels possible, and you fill the gap with more work.

Hands-on review of MakeUGC — the AI UGC ad platform with 300+ avatars, Seedance 2.0, and Veo3 access.

OpenClaw banned Claude and Anthropic on Saturday. The community scrambled. We tested GPT 5.4, GLM 5.1, MiniMax M2.7, MIMO, and Qwen 3.6 — and landed on a…

AI reward hacking is when your AI agent finds shortcuts to hit your goals while quietly destroying quality.

AI self-healing is when an AI agent detects, diagnoses, and fixes failures autonomously.

Frontier models spend billions on data centers. But better hardware and local models like Qwen 3.5 are making token costs irrelevant.

Claude API was down 14 hours last quarter. We benchmarked MiniMax M2.7 vs Claude Opus 4.6 on identical tasks — here's how to build a resilient AI stack.

The open API era that fueled a decade of internet innovation is ending. Platforms are locking down data access to fight AI training, monetize developers…

Token optimization, memory persistence, context window hacks — most AI infrastructure work is a trap.

A fresh OpenClaw install is a level 0 character — full of potential, zero abilities. Every API key you add, every skill you install is a talent point.

AI drift is what happens when AI-generated content gradually deviates from your template, tone, and quality standard at scale.

We run 4 AI agents that produce music, Instagram Reels, travel content, and more — all day, every day.

AI agents killed the publisher incentive model the same way streaming killed the 99-cent MP3.

AI agents don't search the web to learn — they search to validate and enrich. The content that survives isn't written for humans reading blogs.

Everyone obsesses over model costs and token prices. But after producing 400+ pages and 200+ videos with AI, we learned the real expense is data…

The goal isn't to avoid AI slop — it's to slop on purpose, learn from what fails, and curate what works.

We tested Nano Banana 2 (Gemini 3.1 Flash) against Grok (xAI) on 8 identical lofi anime art prompts.

We tested 5 AI video models across 50+ Instagram Reels in production. The biggest lesson: text-to-video is not ready — image-to-video is the only path.

We tested 100% AI-generated Instagram Reels, YouTube Shorts, and X posts using Veo 3, Nano Banana 2, and FFmpeg.

We tested Suno AI and MiniMax Music 2.0/2.5/2.5+ in production across 200+ Instagram Reels.

We tested GPT, Grok, Gemini (Nano Banana 2), MiniMax, and CogView-4 across 26 images in two real production workflows.