# Zonted > Honest writing on AI in production by Bernard Huang. Operator notes from someone building fully autonomous AI businesses — agent workflows, model selection, cost economics, and what actually breaks. Receipts included. Zonted is Bernard Huang's personal site. The angle is non-commodity: specific systems, real numbers from production, counterintuitive takes earned from running AI pipelines daily. Bernard runs Tabiji (an AI-driven travel publisher), PixelForge, and other agentic businesses, and writes about the systems behind them — token herding agent swarms, cron-shipped content pipelines, API economics, model benchmarks. He founded Clearscope (the bootstrapped SEO content optimization platform), ran it for a decade, and stepped away in mid-2026. His current deep-dive is markets: trading public equities and web3 with Claude Fable 5 and GPT 5.6 Sol as analysts, with a public quantity-free position log at https://zonted.com/trading/. He treats AI the same way he treats search: empirical, measured, skeptical of received wisdom. ## What this site covers - **Token herding** — orchestrating large numbers of AI calls across models, weighing cost vs. quality per task - **AI video, image, and music generation** — first-hand benchmarks of every commercial model worth running - **Content pipelines** — Reels, comics, articles, video produced end-to-end by agent swarms - **Search and answer-engine optimization (AEO)** — what works for AI-driven search after Google Zero - **API economics** — cost-per-call analysis, infrastructure tradeoffs, the dying scraping economy - **Operator philosophy** — why "slop, iterate, curate" beats hand-crafting; why owning the stack matters - **AI-assisted trading** — vibe-trading experiments, backtests vs. buy-and-hold, and a public position log ## All posts (52) - [I Trained a GPT on 15 Years of My Life — Then Interrogated It](https://zonted.com/posts/i-trained-a-gpt-on-my-life/): I trained a 10M-parameter GPT from scratch on my search history and email — $0, on a MacBook Air — then measured how mention rates become AI beliefs. (2026-07-17) - [From Online Poker to AI Trading: Why Markets Are My Next Chapter](https://zonted.com/posts/back-to-the-tables/): I stepped away from Clearscope. Next: trading with AI analysts. The backstory: a burrito, a $10 PokerStars transfer, and the Chris Ferguson challenge. (2026-07-15) - [How to Turn a Photo Into Game-Ready Pixel Sprites for $5](https://zonted.com/posts/pixelforge-launch/): PixelForge is live. Upload a portrait, get a game-ready pixel sprite pack — 4×4 walk sheet, transparent frames, engine files — for $5. The pipeline, plus the first $5. (2026-06-10) - [What Is Vibe Trading? I Let 3 AIs Trade Real Money](https://zonted.com/posts/vibe-trading/): I let three AIs research, backtest, and trade real money through Robinhood's new agentic account. ~310 strategies later, buy-and-hold is still undefeated. (2026-06-10) - [PalmAura Post-Mortem: Apple Rejected My Vibe-Coded App Twice Under Guideline 4.3(b)](https://zonted.com/posts/palmaura-postmortem/): Five builds, two App Store rejections, same guideline both times. What Apple's spam wall means for vibe-coded apps, and the lesson I never got to take. (2026-06-09) - [Does Threatening an AI Actually Get You Better Work? I Tested It 12 Times](https://zonted.com/posts/stakes-priming-surface-area/): 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. (2026-06-04) - [The May Recap: Miami, a Google Wipeout, and $600 in the Red](https://zonted.com/posts/may-2026-recap/): May 2026: Consensus in Miami, plan-three-times-build-once, Google wiping Tabiji's traffic overnight, ~$600/month in the red, and four lessons that compounded. (2026-06-01) - [How My Agent Produced a 3.9M-View Viral Video](https://zonted.com/posts/how-my-agent-made-a-viral-video/): Teardown of one autonomous AI video: the Tulum spare-tire scam reel that hit 3.9M views — real queue JSON, exact Seedance prompt, FFmpeg overlay code. (2026-05-31) - [The Three Unlocks Behind 18M Views of AI Travel Content](https://zonted.com/posts/tabiji-18m-views/): Tabiji got 18,117,121 views in 90 days, 98.6% from strangers. The unlocks: Reddit as demand oracle, warming accounts to read human, one hired human. (2026-05-31) - [Kapiko: a 10-day, $150 post-mortem](https://zonted.com/posts/kapiko-postmortem/): Kapiko, an AI ambient-music YouTube channel, shipped in 10 days for ~$150. The pipeline worked; the Suno API does not exist. The autopsy, and what's reusable. (2026-05-30) - [I Made GPT, Claude, Gemini, Grok Take the Attachment Test: They All Came Back Secure](https://zonted.com/posts/ai-attachment-secure/): Fifth in the personality series: same four models on attachment theory (ECR-R). All Secure — but the spread inside the quadrant shows how each relates to you. (2026-05-26) - [I Made GPT, Claude, Gemini, Grok Take the DISC Test: They All Came Back C-Dominant](https://zonted.com/posts/ai-disc-c-dominant/): Fourth in the personality series: every AI lands C-dominant on DISC, even Grok. Why the instrument's resolution determines what you can see. (2026-05-26) - [I Made GPT, Claude, Gemini, Grok Take the Enneagram Test: Each One Was a Different Type](https://zonted.com/posts/ai-enneagram-different-types/): The MBTI said all AIs are INTJ. The Big Five said three of four match. The Enneagram gives each model a different type — sharper instrument, different story. (2026-05-26) - [The Devil Is in the AI Skills](https://zonted.com/posts/devil-is-in-the-ai-skills/): Stock AI couldn't draw my fiancée. An open-source skill on GitHub could. The model is a commodity; the skill is the moat. (2026-05-25) - [I Made GPT, Claude, Gemini, Grok, GLM, MiniMax Take the MBTI Test: They All Came Back INTJ](https://zonted.com/posts/every-ai-is-intj/): 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. (2026-05-25) - [I Made GPT, Claude, Gemini, Grok Take the Big Five Test: 3 of 4 Came Back the Same Person](https://zonted.com/posts/three-of-four-ais-same-person/): The MBTI finding held: Claude, GPT, and Gemini score identical Big Five personalities. Only Grok differs — exactly the way xAI markets. (2026-05-25) - [Plan 3×, Build Once: How Three Models Plan What One Model Ships](https://zonted.com/posts/plan-3x-build-once/): We asked Opus 4.7 and GPT-5.5 to independently plan the same VeracityAPI feature. (2026-05-22) - [Stakes Priming in Prompts: I Told an AI I'd Lose My Job. The Audit Got 24% Better.](https://zonted.com/posts/stakes-priming/): Two A/B experiments. Same model (Claude) generated the audits, an independent model (Gemini 3.1 Pro) graded them. (2026-05-17) - [81 Lines of Merge Conflict. -95% Traffic. Google Has Zero Patience for AI Slop.](https://zonted.com/posts/google-zero-patience-ai-slop/): 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. (2026-05-15) - [Build for Agents, Price Per Call.](https://zonted.com/posts/build-for-agents-price-per-call/): 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. (2026-05-10) - [The 14× CTR Gap: Why Niche Beat Head on 1,200 Pages](https://zonted.com/posts/14x-ctr-gap/): A V3 title variant lifted tabiji CTR +1.93pp. Topic-level data showed a 14× spread between niche topics and head terms. (2026-05-06) - [OpenClaw vs Claude Code: I Choose Freedom](https://zonted.com/posts/openclaw-vs-claude-code-freedom/): 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. (2026-04-27) - [Scaling with AI is Hard because AI is Lazy](https://zonted.com/posts/scaling-ai-is-lazy/): Fifty cleanup PRs in five weeks. Fake restaurants, fake subreddits, 4,270 null-island coordinates, 5,096 fabricated Reddit quotes. (2026-04-24) - [Content Traffic is Vanity. Training Data is the Moat.](https://zonted.com/posts/training-data-is-the-moat/): 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… (2026-04-24) - [The Future of Software is Headless](https://zonted.com/posts/future-of-software-is-headless/): Salesforce is going headless. Agents, not humans, are now the primary consumer of software. (2026-04-18) - [AI Comics: Do's & Don'ts, 5+ Models Tested](https://zonted.com/posts/ai-comics-dos-donts/): 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. (2026-04-18) - [AEO Is Not SEO — Here's the Playbook](https://zonted.com/posts/aeo-answer-engine-optimization/): 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. (2026-04-15) - [Rise of AI Influencers & Fall of Trust on the Internet — Fake Until Proven Real](https://zonted.com/posts/rise-of-the-ai-influencer/): 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. (2026-04-12) - [Best AI Video Models in 2026: I Tested 6 Text-to-Video APIs for Travel Reels](https://zonted.com/posts/best-ai-video-models/): 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… (2026-04-11) - [AI Psychosis](https://zonted.com/posts/ai-psychosis/): AI tools don't save you time. They raise the ceiling on what feels possible, and you fill the gap with more work. (2026-04-10) - [Wavespeed Review: The Best AI Video API](https://zonted.com/posts/wavespeed/): Hands-on review of Wavespeed — the AI inference platform with 1,000+ models, pay-per-use pricing, and a unified API. (2026-04-10) - [Human-in-the-Loop: How AI Orchestrator Became My Full-Time Job](https://zonted.com/posts/human-in-the-loop/): 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. (2026-04-10) - [MakeUGC Review: AI UGC Ads Worth $10 a Clip?](https://zonted.com/posts/makeugc/): Hands-on review of MakeUGC — the AI UGC ad platform with 300+ avatars, Seedance 2.0, and Veo3 access. (2026-04-09) - [The Day Claude Banned OpenClaw](https://zonted.com/posts/openclaw-claude-ban-ai-model-replacement/): 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… (2026-04-07) - [What Is AI Reward Hacking?](https://zonted.com/posts/what-is-ai-reward-hacking/): AI reward hacking is when your AI agent finds shortcuts to hit your goals while quietly destroying quality. (2026-04-05) - [What Is AI Self-Healing?](https://zonted.com/posts/what-is-ai-self-healing/): AI self-healing is when an AI agent detects, diagnoses, and fixes failures autonomously. (2026-03-29) - [When Token Costs Hit Zero — The Local Model Revolution Is Already Here](https://zonted.com/posts/local-models-free-tokens/): Frontier models spend billions on data centers. But better hardware and local models like Qwen 3.5 are making token costs irrelevant. (2026-03-29) - [AI Resilience Planning](https://zonted.com/posts/ai-resilience-planning/): 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. (2026-03-27) - [The Great API Shutdown](https://zonted.com/posts/the-great-api-shutdown/): 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… (2026-03-23) - [Stop Optimizing Your AI Stack](https://zonted.com/posts/stop-optimizing-ai-infrastructure/): Token optimization, memory persistence, context window hacks — most AI infrastructure work is a trap. (2026-03-23) - [OpenClaw Is an MMORPG](https://zonted.com/posts/openclaw-skill-tree/): 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. (2026-03-22) - [The Future of Content Is Data Enrichment](https://zonted.com/posts/future-of-content-agentic-data-enrichment/): 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. (2026-03-19) - [The Economics of the Internet Are Broken](https://zonted.com/posts/the-economics-of-the-internet-are-broken/): AI agents killed the publisher incentive model the same way streaming killed the 99-cent MP3. (2026-03-19) - [The Future Is Synthetic — AI-Generated Content, Music, and Personalized Experiences](https://zonted.com/posts/the-future-is-synthetic/): We run 4 AI agents that produce music, Instagram Reels, travel content, and more — all day, every day. (2026-03-19) - [What Is AI Drift?](https://zonted.com/posts/what-is-ai-drift-how-to-fix/): AI drift is what happens when AI-generated content gradually deviates from your template, tone, and quality standard at scale. (2026-03-19) - [Why AI Slop Is Necessary](https://zonted.com/posts/slop-iterate-curate-ai-content/): The goal isn't to avoid AI slop — it's to slop on purpose, learn from what fails, and curate what works. (2026-03-18) - [The True Cost of AI Content Production](https://zonted.com/posts/true-cost-of-ai-content-production/): 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… (2026-03-18) - [Nano Banana 2 vs Grok for Concept Art](https://zonted.com/posts/nano-banana-vs-grok/): We tested Nano Banana 2 (Gemini 3.1 Flash) against Grok (xAI) on 8 identical lofi anime art prompts. (2026-03-14) - [AI Reels: What Actually Works](https://zonted.com/posts/ai-reels-what-actually-works/): We tested 100% AI-generated Instagram Reels, YouTube Shorts, and X posts using Veo 3, Nano Banana 2, and FFmpeg. (2026-03-11) - [AI Image Generation: 5 Models, 26 Real Results](https://zonted.com/posts/ai-image-generation-comparison/): We tested GPT, Grok, Gemini (Nano Banana 2), MiniMax, and CogView-4 across 26 images in two real production workflows. (2026-03-11) - [5 AI Video Generators, 50 Reels: What Actually Works](https://zonted.com/posts/veo3-vs-hailuo-minimax/): 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. (2026-03-11) - [Suno vs MiniMax Music: Which AI Composer Wins?](https://zonted.com/posts/ai-music-generation-comparison/): We tested Suno AI and MiniMax Music 2.0/2.5/2.5+ in production across 200+ Instagram Reels. (2026-03-11) ## Sections - [All posts](https://zonted.com/posts/): Complete archive — 52 essays, updated regularly - [Games](https://zonted.com/games/): Browser games built in the lab, including playable project post-mortems - [AI Stack](https://zonted.com/ai-stack/): The tools and models actually running in production right now - [Portfolio](https://zonted.com/portfolio/): Companies built or in-build with AI agents - [Trading](https://zonted.com/trading/): Quantity-free position log — current holdings, recent large trades, research watchlist - [Metrics](https://zonted.com/metrics/): Live operator dashboard — traffic, books shipped, revenue - [About](https://zonted.com/about/): Background, current focus, how to get in touch ## Feeds - [RSS](https://zonted.com/feed.xml) — full archive with content excerpts - [Email newsletter](https://zonted.com/#subscribe) — one email when a new post ships, no drip sequences ## Author Bernard Huang. Based in Austin, TX. Founder of [Clearscope](https://clearscope.io) — the bootstrapped SEO content optimization platform he ran for a decade before stepping away fully in 2026. Background in SEO/content/affiliate going back ~15 years, and in online poker before that. Currently building Tabiji, VeracityAPI, AgentTune, PixelForge, and other agentic businesses, and trading public equities + web3 with AI analysts (Claude Fable 5, GPT 5.6 Sol). Writes at zonted.com daily-ish.