Approach 1
Ultrawork
Just say "ulw" and walk away.
- Reads the codebase for context
- Breaks the task into atomic steps
- Implements it
- Verifies against requirements
- Commits the changes
Zero intervention. Just results.
Old version · Jan 19, 2026
Written when not reading the code was still called crazy, and everyone sat in front of the agent.
Read the current versionManifesto
Ultrawork Manifesto
The Philosophy of High-Output Engineering
> HUMAN IN THE LOOP = BOTTLENECK
Imagine an autonomous car that makes you grab the wheel every 30 seconds. Would you call that "autonomous"? No. You'd call it driver assist. Barely better than cruise control.
Why is coding any different?
We've accepted a version of "AI coding" where a chatbot writes 20 lines, then waits for you to fix them. Nobody would call that automation. It's micromanagement.
- Fixing the AI's half-finished code
- Manually correcting syntax errors
- Copy-pasting context back and forth
- Reviewing every single line for hallucinations
That's not "human-AI collaboration." That's the AI failing to do its job.
OmO is built on the premise that the human should be the architect, not the spell-checker.
Indistinguishable Code
Agent-written code should be indistinguishable from code written by a senior engineer.
- Follows existing codebase patterns and architecture
- Handles errors and edge cases without being asked
- Writes tests that actually test behavior, not just coverage
- No AI slop, meaning no over-engineering and no abstractions nobody asked for
- Comments only when they add value, never to restate the obvious
"If you can tell whether a commit was made by a human or an agent, the agent has failed."
Token Cost vs. Productivity
We don't care about token usage. We care about output. If spending $5 on tokens saves an hour of engineering time, that's a 20x ROI.
- Parallel agents exploring multiple solutions
- Complete work without human intervention
- Verifies its own work before calling it done
However...
We optimize for efficiency where it counts. Not by crippling the model, but by:
- Using cheaper models for routine tasks
- Not exploring the same ground twice
- Caching what it learned across sessions
- Stopping once there's enough context
Minimize Human Cognitive Load
The human should only need to say what they want. Everything else is the agent's job.
Approach 2
Planner + Executor
When you want strategic control.
Ultrawork Planner
Interviews you and researches the codebase, then writes a detailed work plan.
/ulw-execute
Executes the approved plan in the same session, delegating by category and tracking progress wave by wave.
You architect. Agents execute. Nothing happens out of sight.
Predictable
Same inputs, same output. No surprises and no creative liberties you didn't ask for.
Continuous
Work survives a crashed session or a week away, because the agent keeps the state and you don't have to.
Delegatable
Hand it a task with clear acceptance criteria, the way you'd hand one to a capable teammate. It corrects itself when something goes wrong and escalates only when it has to.
The Core Loop
- Human Intent
- Agent Execution
- Verified Result
Planning interview
Draw out what you actually want before anything gets built
Plan consultant
Catch ambiguities before they become bugs
Plan review gate
Verify the plan is complete before execution
Orchestrator
Coordinate the work so you don't have to micromanage it
Todo Continuation
Force completion, prevent "I'm done" lies
Category System
Route each task to the right model so you never have to pick one
Background Agents
Research in parallel without blocking you
Wisdom Accumulation
Learn from work, don't repeat mistakes
The Future We're Building
- Developers think about what to build, not how to get an AI to build it
- Code quality that doesn't depend on who wrote it
- Complex projects as easy as simple ones
- "Prompt engineering" becomes obsolete
"The agent should be invisible. Like electricity, like running water."
"You flip the switch. The light turns on. You don't think about the power grid."