Appearance
Values & Principles
These are the principles that guide how we build systems and work with clients.
Not aspirational. Actually how we operate.
Work First, Bureaucracy Second
The most important lesson from building Nexus.
Complete the work. Handle the bureaucracy later.
Example: If code is written, tested, and working - but Jira transition fails - the work is still done. Log the Jira failure, but ship the code.
Don't let external system failures prevent work completion.
Applied to clients: We ship working systems even if documentation lags or processes aren't perfect. Working code > perfect process.
Evidence-Based Validation
Always validate actual behavior. Never trust claims.
Bad:
typescript
if (manifest.tests_added) {
return true; // Trust the claim
}Good:
typescript
const testFiles = manifest.tests_added.map(f => fs.existsSync(f));
return testFiles.every(exists => exists); // Verify the evidenceApplied to clients: We don't take requirements at face value. Week 1 is watching your actual workflows, not just asking what you think you need.
Multi-Layered Prevention
When something goes wrong once, build 4 layers of protection:
- Configuration Time - Validate at input (catch it early)
- Dashboard Time - Make issues impossible to miss (visual alerts)
- Runtime Time - Warn when something's wrong (explicit checks)
- Documentation Time - Explain why it matters (prevent repeat)
Example: Path Confusion Prevention - agents working in wrong directories caused 3 failed PRs. We added validation at all 4 layers.
Applied to clients: We don't just fix bugs. We prevent entire categories of problems from recurring.
Fresh Validation Context
Validators should be adversarial to builders.
If the same system builds AND validates code, it suffers from confirmation bias. It "knows" what it built, so it overlooks edge cases.
Solution: Fresh validator. Gets ONLY the output from the builder. No shared context. No knowledge of what the builder was thinking.
Applied to clients: We test systems in production-like environments, not just where they were built. Fresh eyes catch more issues.
Capacity-Constrained by Design
I can only work with 5 companies at a time.
Not because of artificial scarcity. Because I'm actually building your systems myself, not managing contractors or selling generic software.
Implication: We say no a lot. If we take you on, you get full attention. If we can't take you on, we're honest about it.
Radical Honesty
About what we've done: Field-tested systems with real deployments. Not "proven at scale." Early stage. Building in public.
About limitations: Can't take everyone. Can't solve every problem. Can't promise 4 weeks for everything.
About failures: We document incidents publicly. Example: Path Confusion Incident
Why: Honesty > hype. If you work with us, you should know exactly what you're getting.
4 Weeks, Not 6 Months
Speed is a feature.
Week 1: Figure out the problem. Weeks 2-3: Build the solution. Week 4: Ship it.
Not every project fits this timeline. But most do. And fast feedback loops mean we can iterate if the first version isn't perfect.
Why: You need to see ROI this quarter, not next year. Fast shipping beats perfect planning.
No Consultants Without Code
Our tagline. Our core differentiator.
We don't recommend what we can't build. We don't sell strategy decks. We ship working systems.
Why: The market is full of people who talk. We're builders who also think strategically.
Cost Optimization Without Sacrifice
Use the right tool for the job.
- Builder: Claude Sonnet 4.5 ($3/$15 per 1M tokens) - needs high quality
- Validator: Claude Haiku ($0.25/$1.25 per 1M tokens) - 90% cheaper, sufficient quality
- Monitoring: Prompt caching (90% cost reduction on repeated input)
Applied to clients: We build cost-efficient systems. Not cheap. Efficient. You pay for value, not waste.
Self-Healing Systems
Systems should recover from failures automatically.
Example: Nexus validators retry with corrections when they find issues. Don't just fail - try to fix it.
Applied to clients: We build systems that handle errors gracefully. Not everything can self-heal, but we try.
These Are Actual Principles
Not marketing. These come from building real systems and learning from real failures.
Every principle has a story. Every story has a lesson. Every lesson is documented here.
Related: