2026-01-30 15:03:46 +09:00
<!DOCTYPE html>
< html lang = "en" >
< head >
< meta charset = "UTF-8" >
< meta name = "viewport" content = "width=device-width, initial-scale=1.0" >
2026-01-30 17:10:21 +09:00
<!-- SEO Meta Tags -->
< title > Ultrawork Manifesto | The Philosophy of High-Output Engineering< / title >
< meta name = "description" content = "A philosophical framework for developers and AI engineers to achieve extreme productivity through deep focus, atomic execution, and the Ultrawork methodology." >
< meta name = "keywords" content = "ultrawork, developer productivity, AI engineering, deep work, tech manifesto, oh my opencode, sisyphus" >
< meta name = "robots" content = "index, follow, max-image-preview:large" >
< link rel = "canonical" href = "https://ulw.dev/" >
2026-01-30 20:53:23 +09:00
< link rel = "alternate" hreflang = "en" href = "https://ulw.dev/" >
< link rel = "alternate" hreflang = "ko" href = "https://ulw.dev/ko/" >
< link rel = "alternate" hreflang = "ja" href = "https://ulw.dev/ja/" >
< link rel = "alternate" hreflang = "zh" href = "https://ulw.dev/zh/" >
< link rel = "alternate" hreflang = "es" href = "https://ulw.dev/es/" >
< link rel = "alternate" hreflang = "x-default" href = "https://ulw.dev/" >
2026-01-30 17:10:21 +09:00
< meta name = "theme-color" content = "#0a0a0a" >
< meta name = "author" content = "Yeongyu Kim" >
2026-01-30 20:53:23 +09:00
< link rel = "icon" type = "image/png" href = "./images/favicon.png" >
< link rel = "apple-touch-icon" href = "./images/favicon.png" >
2026-01-30 17:10:21 +09:00
<!-- Open Graph / Facebook -->
< meta property = "og:type" content = "website" >
< meta property = "og:url" content = "https://ulw.dev/" >
< meta property = "og:title" content = "Ultrawork Manifesto | The Philosophy of High-Output Engineering" >
< meta property = "og:description" content = "A philosophical framework for developers and AI engineers to achieve extreme productivity through deep focus, atomic execution, and the Ultrawork methodology." >
< meta property = "og:image" content = "https://ulw.dev/images/og-image.png" >
2026-01-30 17:35:07 +09:00
< meta property = "og:image:width" content = "1200" >
< meta property = "og:image:height" content = "630" >
< meta property = "og:image:alt" content = "Ultrawork Manifesto - A Blueprint for Meaningful Work" >
2026-01-30 17:10:21 +09:00
< meta property = "og:site_name" content = "Ultrawork" >
<!-- Twitter Card -->
< meta name = "twitter:card" content = "summary_large_image" >
< meta name = "twitter:site" content = "@justsisyphus" >
< meta name = "twitter:creator" content = "@justsisyphus" >
< meta name = "twitter:title" content = "Ultrawork Manifesto | The Philosophy of High-Output Engineering" >
< meta name = "twitter:description" content = "A philosophical framework for developers and AI engineers to achieve extreme productivity through deep focus, atomic execution, and the Ultrawork methodology." >
< meta name = "twitter:image" content = "https://ulw.dev/images/og-image.png" >
<!-- Fonts -->
2026-01-30 15:03:46 +09:00
< link rel = "preconnect" href = "https://fonts.googleapis.com" >
< link rel = "preconnect" href = "https://fonts.gstatic.com" crossorigin >
2026-01-30 17:10:21 +09:00
< link href = "https://fonts.googleapis.com/css2?family=Cormorant+Garamond:ital,wght@0,300;0,400;0,500;0,600;0,700;1,400&family=Inter:wght@400;600;700&family=JetBrains+Mono:wght@400&display=swap" rel = "stylesheet" >
2026-01-30 15:03:46 +09:00
< link rel = "stylesheet" href = "styles.css" >
< / head >
< body >
< header >
< div class = "hero-container" >
2026-01-30 17:10:21 +09:00
< img src = "./images/hero.png" alt = "Ultrawork Hero" class = "hero-image" >
< h1 class = "hero-title" > Ultrawork Manifesto< / h1 >
2026-01-30 15:03:46 +09:00
< / div >
< / header >
< main >
<!-- Section 1 -->
< section id = "human-intervention" >
< h2 > Human Intervention is a Failure Signal< / h2 >
< div class = "gold-gradient-text bottleneck-text" >
HUMAN IN THE LOOP = BOTTLENECK< br >
HUMAN IN THE LOOP = BOTTLENECK< br >
HUMAN IN THE LOOP = BOTTLENECK
< / div >
< p > Think about autonomous driving. When a human has to take over the wheel, that's not a feature - it's a failure of the system. The car couldn't handle the situation on its own.< / p >
< h3 > Why is coding any different?< / h3 >
< p > When you find yourself:< / p >
< ul >
< li > Fixing the AI's half-finished code< / li >
< li > Manually correcting obvious mistakes< / li >
< li > Guiding the agent step-by-step through a task< / li >
< li > Repeatedly clarifying the same requirements< / li >
< / ul >
< p > ...that's not "human-AI collaboration." That's the AI failing to do its job.< / p >
2026-01-30 17:10:21 +09:00
< p > < strong > < a href = "https://github.com/code-yeongyu/oh-my-opencode" target = "_blank" rel = "noopener" > Oh My OpenCode< / a > is built on this premise< / strong > : Human intervention during agentic work is fundamentally a wrong signal. If the system is designed correctly, the agent should complete the work without requiring you to babysit it.< / p >
2026-01-30 15:03:46 +09:00
< / section >
< div class = "divider" >
2026-01-30 17:10:21 +09:00
< img src = "./images/orb-divider.png" alt = "Divider" >
2026-01-30 15:03:46 +09:00
< / div >
<!-- Section 2 -->
< section id = "indistinguishable-code" >
< h2 > Indistinguishable Code< / h2 >
< p class = "highlight-box" > < strong > Goal: Code written by the agent should be indistinguishable from code written by a senior engineer.< / strong > < / p >
< p > Not "AI-generated code that needs cleanup." Not "a good starting point." The actual, final, production-ready code.< / p >
< p > This means:< / p >
< ul >
< li > Following existing codebase patterns exactly< / li >
< li > Proper error handling without being asked< / li >
< li > Tests that actually test the right things< / li >
< li > No AI slop (over-engineering, unnecessary abstractions, scope creep)< / li >
< li > Comments only when they add value< / li >
< / ul >
< p > If you can tell whether a commit was made by a human or an agent, the agent has failed.< / p >
< / section >
< div class = "divider" >
2026-01-30 17:10:21 +09:00
< img src = "./images/orb-divider.png" alt = "Divider" >
2026-01-30 15:03:46 +09:00
< / div >
<!-- Section 3 -->
< section id = "token-cost" >
< h2 > Token Cost vs. Productivity< / h2 >
< p > < strong > Higher token usage is acceptable if it significantly increases productivity.< / strong > < / p >
< p > Using more tokens to:< / p >
< ul >
< li > Have multiple specialized agents research in parallel< / li >
< li > Get the job done completely without human intervention< / li >
< li > Verify work thoroughly before completion< / li >
< li > Accumulate knowledge across tasks< / li >
< / ul >
< p > ...is a worthwhile investment when it means 10x, 20x, or 100x productivity gains.< / p >
< h3 > However:< / h3 >
< p > Unnecessary token waste is not pursued. The system optimizes for:< / p >
< ul >
< li > Using cheaper models (Haiku, Flash) for simple tasks< / li >
< li > Avoiding redundant exploration< / li >
< li > Caching learnings across sessions< / li >
< li > Stopping research when sufficient context is gathered< / li >
< / ul >
< p > Token efficiency matters. But not at the cost of work quality or human cognitive load.< / p >
< / section >
< div class = "divider" >
2026-01-30 17:10:21 +09:00
< img src = "./images/orb-divider.png" alt = "Divider" >
2026-01-30 15:03:46 +09:00
< / div >
<!-- Section 4 -->
< section id = "cognitive-load" >
< h2 > Minimize Human Cognitive Load< / h2 >
< p > < strong > The human should only need to say what they want. Everything else is the agent's job.< / strong > < / p >
< p > Two approaches to achieve this:< / p >
< div class = "approach-container" >
2026-01-30 17:10:21 +09:00
< div class = "approach ultrawork-approach" >
< h3 > Approach 1: Ultrawork< / h3 >
< p class = "approach-tagline" > Just say "ulw" and walk away.< / p >
< p > You say: < code > ulw add authentication< / code > < / p >
< p > The agent autonomously:< / p >
2026-01-30 15:03:46 +09:00
< ul >
2026-01-30 17:10:21 +09:00
< li > Analyzes your codebase patterns and architecture< / li >
< li > Researches best practices from official docs< / li >
< li > Plans the implementation strategy internally< / li >
< li > Implements following your existing conventions< / li >
< li > Verifies with tests and LSP diagnostics< / li >
< li > Self-corrects when something goes wrong< / li >
< li > < strong > Keeps bouldering until 100% complete< / strong > < / li >
2026-01-30 15:03:46 +09:00
< / ul >
2026-01-30 17:10:21 +09:00
< p class = "approach-summary" > < strong > Zero intervention. Full autonomy. Just results.< / strong > < / p >
2026-01-30 15:03:46 +09:00
< / div >
2026-01-30 17:10:21 +09:00
< div class = "approach prometheus-approach" >
< h3 > Approach 2: Prometheus + Atlas< / h3 >
< p class = "approach-tagline" > When you want strategic control.< / p >
2026-01-30 20:53:23 +09:00
< p > Press < kbd > Tab< / kbd > to switch agents, then: < code > add authentication< / code > < / p >
2026-01-30 17:10:21 +09:00
< p > < strong > Prometheus< / strong > (Strategic Planner):< / p >
2026-01-30 15:03:46 +09:00
< ul >
2026-01-30 17:10:21 +09:00
< li > Conducts deep codebase research via parallel agents< / li >
< li > Interviews you with intelligent, contextual questions< / li >
< li > Identifies edge cases and architectural implications< / li >
< li > Generates a detailed YAML work plan with dependencies< / li >
2026-01-30 15:03:46 +09:00
< / ul >
2026-01-30 17:10:21 +09:00
< p > < strong > Atlas< / strong > (Master Orchestrator):< / p >
< ul >
< li > Executes the plan via < code > /start-work< / code > < / li >
< li > Delegates tasks to specialized agents (Oracle, Frontend Engineer, etc.)< / li >
< li > Manages parallel execution waves for efficiency< / li >
< li > Tracks progress, handles failures, ensures completion< / li >
< / ul >
< p class = "approach-summary" > < strong > You architect. Agents execute. Full transparency.< / strong > < / p >
2026-01-30 15:03:46 +09:00
< / div >
< / div >
< p > In both cases, the human's job is to < strong > express what they want< / strong > , not to manage how it gets done.< / p >
< / section >
< div class = "divider" >
2026-01-30 17:10:21 +09:00
< img src = "./images/orb-divider.png" alt = "Divider" >
2026-01-30 15:03:46 +09:00
< / div >
<!-- Section 5 -->
< section id = "predictable-continuous" >
< h2 > Predictable, Continuous, Delegatable< / h2 >
< p > < strong > The ideal agent should work like a compiler< / strong > : markdown document goes in, working code comes out.< / p >
< h3 > Predictable< / h3 >
< p > Given the same inputs:< / p >
< ul >
< li > Same codebase patterns< / li >
< li > Same requirements< / li >
< li > Same constraints< / li >
< / ul >
< p > ...the output should be consistent. Not random, not surprising, not "creative" in ways you didn't ask for.< / p >
< h3 > Continuous< / h3 >
< p > Work should survive interruptions:< / p >
< ul >
< li > Session crashes? Resume with < code > /start-work< / code > < / li >
< li > Need to step away? Progress is tracked< / li >
< li > Multi-day project? Context is preserved< / li >
< / ul >
< p > The agent maintains state. You don't have to.< / p >
< h3 > Delegatable< / h3 >
< p > Just like you can assign a task to a capable team member and trust them to handle it, you should be able to delegate to the agent.< / p >
< p > This means:< / p >
< ul >
< li > Clear acceptance criteria, verified independently< / li >
< li > Self-correcting behavior when something goes wrong< / li >
< li > Escalation (to Oracle, to user) only when truly needed< / li >
< li > Complete work, not "mostly done"< / li >
< / ul >
< / section >
< div class = "divider" >
2026-01-30 17:10:21 +09:00
< img src = "./images/orb-divider.png" alt = "Divider" >
2026-01-30 15:03:46 +09:00
< / div >
<!-- Section 6 -->
< section id = "core-loop" >
< h2 > The Core Loop< / h2 >
< div class = "ascii-art" >
Human Intent → Agent Execution → Verified Result
↑ ↓
└──────── Minimum ─────────────┘
(intervention only on true failure)
< / div >
2026-01-30 17:10:21 +09:00
< p > Everything in < a href = "https://github.com/code-yeongyu/oh-my-opencode" target = "_blank" rel = "noopener" > Oh My OpenCode< / a > is designed to make this loop work:< / p >
2026-01-30 15:03:46 +09:00
< table >
< thead >
< tr >
< th > Feature< / th >
< th > Purpose< / th >
< / tr >
< / thead >
< tbody >
< tr >
< td > Prometheus< / td >
< td > Extract intent through intelligent interview< / td >
< / tr >
< tr >
< td > Metis< / td >
< td > Catch ambiguities before they become bugs< / td >
< / tr >
< tr >
< td > Momus< / td >
< td > Verify plans are complete before execution< / td >
< / tr >
< tr >
< td > Orchestrator< / td >
< td > Coordinate work without human micromanagement< / td >
< / tr >
< tr >
< td > Todo Continuation< / td >
< td > Force completion, prevent "I'm done" lies< / td >
< / tr >
< tr >
< td > Category System< / td >
< td > Route to optimal model without human decision< / td >
< / tr >
< tr >
< td > Background Agents< / td >
< td > Parallel research without blocking user< / td >
< / tr >
< tr >
< td > Wisdom Accumulation< / td >
< td > Learn from work, don't repeat mistakes< / td >
< / tr >
< / tbody >
< / table >
< / section >
< div class = "divider" >
2026-01-30 17:10:21 +09:00
< img src = "./images/orb-divider.png" alt = "Divider" >
2026-01-30 15:03:46 +09:00
< / div >
<!-- Section 7 -->
< section id = "future" >
< h2 > The Future We're Building< / h2 >
< p > A world where:< / p >
< ul >
< li > Human developers focus on < strong > what< / strong > to build, not < strong > how< / strong > to get AI to build it< / li >
< li > Code quality is independent of who (or what) wrote it< / li >
< li > Complex projects are as easy as simple ones (just take longer)< / li >
< li > "Prompt engineering" becomes as obsolete as "compiler debugging"< / li >
< / ul >
< p > < strong > The agent should be invisible.< / strong > Not in the sense that it's hidden, but in the sense that it just works - like electricity, like running water, like the internet.< / p >
< p > You flip the switch. The light turns on. You don't think about the power grid.< / p >
< p class = "final-statement" > That's the goal.< / p >
< / section >
< / main >
< footer >
< div class = "footer-content" >
2026-01-30 17:10:21 +09:00
< a href = "https://github.com/code-yeongyu/oh-my-opencode" class = "cta-link" target = "_blank" rel = "noopener" >
Get Oh My OpenCode →
< / a >
2026-01-30 15:03:46 +09:00
< p > < strong > just ulw ulw< / strong > < / p >
2026-01-30 20:53:23 +09:00
< nav class = "language-selector" aria-label = "Language selection" >
< span class = "language-selector-label" > Language< / span >
< div class = "language-links" >
< a href = "/" class = "language-link active" lang = "en" > English< / a >
< a href = "/ko/" class = "language-link" lang = "ko" > 한국어< / a >
< a href = "/ja/" class = "language-link" lang = "ja" > 日本語< / a >
< a href = "/zh/" class = "language-link" lang = "zh" > 简体中文< / a >
< a href = "/es/" class = "language-link" lang = "es" > Español< / a >
< / div >
< / nav >
2026-01-30 15:03:46 +09:00
< / div >
< / footer >
< / body >
< / html >
