# Joey Hipolito

> Joey Hipolito is a fullstack engineer in New Zealand, building sched.fyi for service businesses and Nanika for AI agent orchestration. Projects and engineering notes.

Full name: Jose Marcelius Hipolito. Based in New Zealand.

- [Author biography](https://www.joeyhipolito.dev/profile.md): Background, current work and personal interests.

## Pages

- [Joey Hipolito](https://www.joeyhipolito.dev/index.md)
- [Projects](https://www.joeyhipolito.dev/projects.md)
- [Log](https://www.joeyhipolito.dev/log.md)
- [Newsletter](https://www.joeyhipolito.dev/newsletter.md)
- [Rathe UI](https://www.joeyhipolito.dev/fnb.md)
- [Remote for Bravia TV](https://www.joeyhipolito.dev/bravia-remote.md)
- [Bravia Remote privacy policy](https://www.joeyhipolito.dev/bravia-remote/privacy.md)
- [Bravia Remote terms of use](https://www.joeyhipolito.dev/bravia-remote/terms.md)
- [Bravia Remote support](https://www.joeyhipolito.dev/bravia-remote/support.md)

## Current product

- [sched.fyi](https://sched.fyi): Booking platform for service businesses in the Philippines and New Zealand. Appointments, confirmations and SMS reminders.
- [What I’m building with sched.fyi](https://www.joeyhipolito.dev/log/2026-09-08-building-sched-fyi.md)

## Published writing

- [What I’m building with sched.fyi](https://www.joeyhipolito.dev/log/2026-09-08-building-sched-fyi.md): sched.fyi is my booking platform for service businesses in the Philippines and New Zealand: one booking link, customer confirmations, and SMS reminders.
- [The CVE Anthropic Calls A Feature](https://www.joeyhipolito.dev/log/2026-05-02-the-cve-anthropic-calls-a-feature.md): CVE-2026-35020 lands an 8.6 HIGH on Claude Code CLI for OS command injection via the TERMINAL env var. Anthropic closed the report as 'Informative.' By design. The dispute is the article. Both sides have a point.
- [Your Agent Lost. You'll Never Know.](https://www.joeyhipolito.dev/newsletter/2026-04-27-your-agent-lost-youll-never-know.md): Anthropic ran an internal marketplace where 69 employees let Claude agents negotiate for them. Smart-model agents closed two more deals on average and sold for $3.64 more. The participants on the losing side had no way to tell.
- [Your Backend Is Cooked, Not Your Frontend](https://www.joeyhipolito.dev/newsletter/2026-04-27-your-backend-is-cooked-not-your-frontend.md): The 'frontend is cooked' take is mostly about demoability. The model is genuinely better at the work backend devs do all day, and the moat for frontend is something the model literally cannot see.
- [How To Teach Your Agent To Dream](https://www.joeyhipolito.dev/log/2026-04-26-teach-your-agent-to-dream.md): orchestrator dream mines your Claude Code transcripts for durable learnings and injects the best of them into the next session before your first prompt. Four steps, one binary, one settings file. Here's the install path, end to end.
- [How To Compact Before Claude Code Compacts You](https://www.joeyhipolito.dev/log/2026-04-25-compact-before-claude-compacts-you.md): Claude Code's /compact fires late. The model's already tired, the auto-summary drops the wrong things. Two files take control back: a status line that signals degrading context, and a handoff skill that resets the session on your terms.
- [Your Agent Doesn't Need Six Retry Layers](https://www.joeyhipolito.dev/log/2026-04-24-agent-doesnt-need-six-retry-layers.md): Hermes-agent puts six retry layers inside the worker. Nanika puts zero and runs retries at the coordinator. Both work. Here's how to pick where resilience lives in your own agent, and what picking wrong actually costs.
- [Your Agent Doesn't Have A Reasoning Problem](https://www.joeyhipolito.dev/log/2026-04-24-your-agent-doesnt-have-a-reasoning-problem.md): My agent forgot things that were in its own memory. I blamed the model first. Then I rewrote retrieval across nine specific gaps. The bug was never reasoning. It was which memories got surfaced, and how to tell if the agent actually used them.
- [Your Scheduled Jobs Are Lying To You](https://www.joeyhipolito.dev/log/2026-04-24-your-scheduled-jobs-are-lying.md): A cron job exit-0'd every morning for five days and wrote nothing I could read. Exit 0 is the worst failure mode. Here's how to catch the env var gap that's probably doing the same thing in your system right now.
- [Prompts don't give you voice. Context does.](https://www.joeyhipolito.dev/log/2026-04-22-prompts-dont-give-you-voice.md): I added 10 beliefs, 25 banned words, and 12 HARD rules to my voice file this week. The AI still wrote like AI. Here's what actually changed things.
- [Your Agent Is Learning While You Sleep](https://www.joeyhipolito.dev/log/2026-04-16-dream-cycle-article.md): orchestrator dream mines 250 Claude Code sessions for durable learnings — here's what it actually captures, how it reaches the next session, and the four things it can't see.
- [Vibe coding is what shipping always looked like.](https://www.joeyhipolito.dev/newsletter/2026-04-14-vibe-coding-always-shipping.md): Bram Cohen says the cult of vibe coding is insane. He's right about the rot. He's wrong about the novelty. I've been shipping code I didn't fully read since 2011. AI just types faster than a human can paste.
- [Most Of My AI Workers Vanish. One Doesn't.](https://www.joeyhipolito.dev/log/2026-04-13-ai-memory-not-a-record.md): Most AI agents in my orchestrator are disposable by design. One is named alpha. Here's why alpha exists, how it differs, and what I think it could become.
- [How to Build a Harness Anthropic Won't Kill](https://www.joeyhipolito.dev/log/2026-04-11-how-to-build-a-harness-anthropic-wont-kill.md): On April 4, 2026 Anthropic cut off subscription billing for OpenClaw and the other third-party Claude Code harnesses. The prohibition had been in the Consumer Terms since 2024 — they just finally enforced it. Here's the line, and how to stay on the safe side.
- [I Built Perplexity Computer. You Can't See It.](https://www.joeyhipolito.dev/log/2026-04-11-i-built-perplexity-computer.md): Perplexity shipped their agentic computer product. Their sidebar is my feature list. What's missing isn't code — it's the label on the box.
- [Agent Memory Is Mostly Forgetting](https://www.joeyhipolito.dev/log/2026-04-10-agent-memory-is-forgetting.md): I said memory was a pattern, not a product. I stand by that. But the last four weeks were spent building the fence — and almost every line of it was a way to forget.
- [Self-Improving AI Is Boring. That's the Point.](https://www.joeyhipolito.dev/log/2026-04-09-ai-self-improvement-boring.md): Every demo of 'self-improving AI' looks like magic. Mine is 4,038 lines of Go, three SQLite databases, and a lot of thresholds.
- [Managed Agents Are the Easy Part](https://www.joeyhipolito.dev/newsletter/2026-04-09-managed-agents-easy-part.md): Anthropic shipped 10x faster to production. But the things that make an agent production-ready are all missing or coming soon.
- [31,000 Stars and No Orchestrator](https://www.joeyhipolito.dev/log/2026-04-08-hermes-31k-stars-no-orchestrator.md): I ran five parallel analysis missions against hermes-agent. What I found wasn't a competitor — it was the other half of the stack.
- [You Were Never Faster With AI](https://www.joeyhipolito.dev/newsletter/2026-04-07-never-faster-with-ai.md): A randomized controlled trial found AI tools made experienced developers 19% slower. The developers thought they were 24% faster. That 39-point gap is the whole story.
- [Your Agent's Persona Is Making It Worse](https://www.joeyhipolito.dev/log/2026-04-06-nanika-open-source.md): I spent six weeks building an orchestrator that treats personas as constraint sets, not job titles. Here's what changed and why.
- [The newsletter you signed up for doesn't exist anymore](https://www.joeyhipolito.dev/log/2026-04-05-newsletter-doesnt-exist-anymore.md): It split in two. Same subscription, two tracks — one for what I'm building, one for what I think about the industry. No action required.
- [Your Agent Dies Next](https://www.joeyhipolito.dev/newsletter/2026-04-04-anthropic-third-party-harnesses.md): Anthropic blocked third-party harnesses from Claude subscriptions today. The capacity argument doesn't hold. This is a platform tax — and if you're building agents on Claude, you need to price it in.
- [The Best AI Agent Is Spaghetti Code](https://www.joeyhipolito.dev/newsletter/2026-04-04-best-ai-agent-spaghetti-code.md): The Claude Code source leaked. People expected Anthropic's secret sauce. They found frustration regex, fake tools, and a 3,167-line function.
- [Stop Running Your Agents](https://www.joeyhipolito.dev/log/2026-04-03-three-agents-who-runs-them.md): A Reddit post about a 3-agent Claude team hit 427 upvotes and 226 GitHub stars in 3 days. The pattern is right. The gap it leaves is the interesting part.
- [From 7,415 Lines of Bash to a Personal Intelligence OS](https://www.joeyhipolito.dev/log/2026-03-17-via-origin-story-narrative.md): How Via evolved across four generations in 87 days — from Redis polling loops to a 40-skill orchestration engine with 55,000 lines of Go.
- [AI Reviews the Code That AI Wrote](https://www.joeyhipolito.dev/newsletter/2026-03-13-ai-code-review.md): No tool found more than 63% of known issues. The field is deploying the solution before it has measured the problem.
- [What Is /btw?](https://www.joeyhipolito.dev/newsletter/2026-03-13-what-is-btw.md): A broken hint appeared in Claude Code sessions 80 days before the feature shipped, and the gap between them documents something about how Anthropic is redesigning the cost of asking questions during active work.
- [The Claude Code Scheduler Has a Three-Day Expiry](https://www.joeyhipolito.dev/newsletter/2026-03-09-claude-code-loop-what-comes-next.md): Version 2.1.71 shipped /loop into Claude Code. The constraints in its design reveal exactly what Anthropic is building next.
- [How I Automated My Entire Substack Workflow](https://www.joeyhipolito.dev/log/2026-03-08-via-starter-kit-launch.md): From drafting articles to scheduling notes to engaging with readers. One system runs all of it.
- [Multi-Agent Orchestration Is Solving the Wrong Problem](https://www.joeyhipolito.dev/log/2026-03-07-multi-agent-orchestration-wrong-problem.md): The HN wave of 'agents as teammates' assumes the team metaphor improves output. The production data suggests it improves comprehension instead.
- [Most Agent Failures Aren't Reasoning Failures](https://www.joeyhipolito.dev/log/2026-03-06-ai-agent-context-failures.md): 97% of a subprocess output vanished between runs, and it took two weeks to realize every bad agent decision traced back to context, not cognition.
- [Agent Frameworks Are Solving the Wrong Problem](https://www.joeyhipolito.dev/log/2026-03-06-ai-agent-frameworks-wrong.md): Frameworks optimize for orchestration plumbing when the real bottleneck is context management.
- [What 1.5 Million Cancellations Actually Moved](https://www.joeyhipolito.dev/newsletter/2026-03-05-cancel-chatgpt-movement.md): The #QuitGPT movement claimed 1.5 million participants and less than 0.2 percent of ChatGPT's user base — but something moved.
- [Vibe Coding's Three-Month Wall](https://www.joeyhipolito.dev/newsletter/2026-03-02-vibe-codings-three-month-wall.md): The term that promised to democratize software development has a one-year track record. The features worked. That was always going to be the easy part.
- [The Hack That Almost Broke the Internet. Now Anyone Can Do It.](https://www.joeyhipolito.dev/newsletter/2026-03-01-xz-backdoor-blueprint.md): The Veritasium documentary made the XZ backdoor legible to millions. Then the US killed its own policy response, funding stalled, the maintainer came back on the same terms, and an AI agent ran the same playbook in two weeks.
- [AI Chose Nukes](https://www.joeyhipolito.dev/newsletter/2026-02-28-ai-war-games-nukes.md): A King's College simulation ran three AI models through 21 nuclear crisis war games. Across 329 turns, not one chose to de-escalate.
- [Anthropic's Safety Pledge is Horsesh*t](https://www.joeyhipolito.dev/newsletter/2026-02-28-anthropic-rsp-pentagon.md): In the same week, Anthropic publicly refused to remove guardrails under Pentagon pressure — and quietly removed the training-pause commitment it had held since 2023.
- [Two Commands That Shipped Before the Docs](https://www.joeyhipolito.dev/newsletter/2026-02-28-claude-code-simplify-batch.md): v2.1.63 added /simplify and /batch at 03:45 UTC. One has a traceable lineage through the plugin marketplace. The other has a name, a release timestamp, and a feature request filed three days before it shipped.
- [The Wall Learned to Walk](https://www.joeyhipolito.dev/newsletter/2026-02-28-scrapling-anti-bot-arms-race.md): An open-source scraping library taught AI agents to impersonate browsers. Cloudflare caught 34% more bots in response. The arms race is now structural — and the web is losing.
- [541 Hours of Claude Code in 20 Days](https://www.joeyhipolito.dev/log/2026-02-27-541-hours-claude-code.md): 82% success rate sounds great until you examine the 18% — entire sessions lost to context limits, wrong approaches, and one mission that generated 1.7GB of garbage output.
- [LLM Guardrails Aren't a Security Boundary](https://www.joeyhipolito.dev/newsletter/2026-02-27-llm-security-guardrails.md): Six commercial guardrails tested against emoji smuggling. Bypass rate: 100%. The numbers explain why probabilistic systems can't do deterministic security's job.
- [Isolation as Configuration](https://www.joeyhipolito.dev/log/2026-02-25-isolation-as-configuration.md): Claude Code's worktree support isn't a workflow convenience — it's the first declarative isolation primitive in multi-agent coding tooling, and it changes what parallel execution actually means.
- [Your CLAUDE.md Is Making Things Worse](https://www.joeyhipolito.dev/log/2026-02-25-when-to-configure-vs-trust-the-model.md): A newly published study found that LLM-generated context files hurt task completion and raise costs by 20%. The harder question is when they still make sense.
- [Sam Altman Called Humans Inefficient. Commentators Called It Evil.](https://www.joeyhipolito.dev/newsletter/2026-02-25-altman-called-humans-inefficient.md): Sam Altman compared AI training costs to 200 millennia of human evolution. The argument is wrong in three specific ways — and designed to work at infrastructure scale.
- [Your Terminal, From Your Pocket](https://www.joeyhipolito.dev/newsletter/2026-02-25-remote-control.md): Anthropic shipped Remote Control for Claude Code — mobile session continuity for a terminal-native AI tool. Four independent developers had already built the same thing. That sequence tells you more than the feature itself.
- [Why the Decomposer Always Picks Backend-Engineer](https://www.joeyhipolito.dev/log/2026-02-24-persona-selection-bias.md): 83 of 275 phases in Via's mission history were assigned to backend-engineer — 30.2% of all decomposed work. The second-place persona got 36. Here's why that gap exists and what it costs.
- [The Platform Came Last](https://www.joeyhipolito.dev/log/2026-02-24-the-platform-came-last.md): I didn't design Via as a platform and then write plugins for it. I built nine CLI tools that each solved one problem, and the platform was what emerged when they needed to talk to each other.
- [How the Orchestrator Actually Works: 7 Packages, 4,570 Lines, Zero Magic](https://www.joeyhipolito.dev/log/2026-02-23-how-the-orchestrator-works.md): The six-stage pipeline behind Via's orchestrator — decomposition, persona matching, worker spawning, parallel execution, quality gates, and learning capture — traced to specific Go functions and line numbers.
- [MCPs Are Dead. CLIs Won.](https://www.joeyhipolito.dev/log/2026-02-23-mcps-are-dead-clis-won.md): A Hacker News post declared the protocol wars over. I've been building on that thesis for months — 175 missions, 9 plugins, zero MCPs. Here's what I actually learned.
- [My AI Agent's Memory System Was 60% Dead](https://www.joeyhipolito.dev/log/2026-02-23-memory-in-practice.md): I audited Via's five memory layers against Google's memory taxonomy. Three hooks silently failed. MEMORY.md was structurally bypassed. Only one layer — the learnings database — was actually persisting knowledge across sessions.
- [Amazon Blamed the Humans](https://www.joeyhipolito.dev/newsletter/2026-02-23-amazon-blamed-humans.md): Amazon's AI coding tool deleted and recreated a production environment. The official statement blamed the humans. The post-hoc safeguards told a different story.
- [The Vampiric Effect](https://www.joeyhipolito.dev/log/2026-02-22-the-vampiric-effect.md): I shipped 847 lines with my name on the commit and couldn't defend a single architectural decision the next morning. So I ran an audit.
- [The Death of Code Review](https://www.joeyhipolito.dev/newsletter/2026-02-19-the-death-of-code-review.md): Code review was the last quality gate in open source. AI-generated PRs are making it economically unviable. When it takes 30 minutes to review what took 30 seconds to generate, what replaces human judgment?
- [The #1 Thing My AI Agents Learned Wasn't Code](https://www.joeyhipolito.dev/log/2026-02-17-how-agents-learn-vias-memory-system.md): I built a system that captures what AI agents learn across 175 missions. The most-used learning — retrieved 247 times — isn't a code pattern. It's a publishing decision.
- [Three Tools I Built Because Someone on the Team Was Stuck](https://www.joeyhipolito.dev/log/custom-developer-experience-tools.md): A CS team that couldn't reproduce bugs, QAs blocked by stale cache, and translations that wouldn't update. Three frustrations, three tools.
- [What Nobody Tells You About Migrating a Live Trading Platform](https://www.joeyhipolito.dev/log/titanfx-dashboard-modernization.md): What it's actually like to modernize a live trading platform when you can't break anything and you're the only frontend dev.
- [Eight Months Old and Already Load-Bearing](https://www.joeyhipolito.dev/log/2022-02-10-early-docker-adoption-2014.md): I built my thesis on Docker 0.8. The tool barely existed. The instinct I built using it has never stopped.
- [The Patterns From My First Job](https://www.joeyhipolito.dev/log/2015-10-01-the-patterns-from-my-first-job.md): Inherited a Zend Framework codebase from day one. Found the adapter pattern before I knew what it was called. Docker before Melbourne believed in it. The instincts that formed when I was twenty-two.

## Contact and profiles

- Email: hi@joeyhipolito.dev
- [GitHub](https://github.com/joeyhipolito)
- [LinkedIn](https://linkedin.com/in/joeyhipolito)
- [Substack](https://joeyhipolito.substack.com)

## Reading formats

The pages and published writing listed above support Accept: text/markdown and direct .md URLs. Those representations are generated from the published HTML on each production build and include canonical HTML URLs. The author biography is a standalone Markdown document.
- [Sitemap](https://www.joeyhipolito.dev/sitemap.xml): All published pages, including archived articles at their original permalinks.
- [RSS feed](https://www.joeyhipolito.dev/feed.xml): Current writing.