The Trellis Mindset
A trellis doesn't grow anything itself — it provides structure so growth goes where you want. Without a trellis, vines sprawl chaotically. With one, they climb intentionally.
Your principles are the trellis. AI is the accelerated growth.
AI amplifies whatever you give it. Clear structure produces focused output. Vague direction produces sprawl. The quality of what you build with AI is determined by the structure you provide before the first prompt.
Think in Systems
The unit of work is changing. Tomorrow's developers won't hand-write functions and methods — they'll describe systems and let AI generate the implementation. The skill shifts from how do I write this code to how do these systems connect.
Systems thinking means:
- •Seeing your application as interconnected services, not files of code
- •Understanding data flow between boundaries (API ↔ Database ↔ UI ↔ External Services)
- •Knowing which systems exist, what they're responsible for, and how they communicate
- •Designing at the architecture level, delegating implementation to AI
The new literacy:
| Old Model | New Model |
|---|---|
| Write functions | Describe capabilities |
| Debug line by line | Validate system behavior |
| Know syntax | Know patterns |
| Read code | Read architecture |
| Implement solutions | Specify outcomes |
You don't need to know how every function works internally. You need to know what each system does, what it expects, and what it produces. AI handles the wiring. You handle the design.
Core Principles
Context is an asset with a lifecycle
Working context — changelogs, scratchpads, story docs — makes AI dramatically more useful. But context that helps you today becomes noise tomorrow. Create it deliberately, preserve it when valuable, clear it when done.
- →Continuity beats starting over. An AI with context about your work outperforms one starting cold.
- →Know when to scrap and restart. If you're spending more time correcting than creating, a fresh start wins.
- →Keep standards global, context local. Standards belong to the team; working context belongs to the session.
AI drafts, you decide
Treat AI output as a first draft, not a final answer. The faster AI produces something, the more important your review becomes. Speed without verification is just faster mistakes.
- →Small steps, frequent checks. Long autonomous runs produce impressive demos and expensive debugging sessions.
- →Trust but verify. AI will confidently produce plausible-looking code that doesn't work.
Automate what should be repeatable
AI is non-deterministic. That's fine for problem-solving and creative work. But if something needs exact results every time — automate it programmatically. Scripts, hooks, pipelines. Save AI for work that benefits from flexibility.
Make the right thing the easy thing
If clearing context before merging requires five manual steps, people will skip it. If syncing standards is automatic, people will stay current. Design for how people actually work.
Constraint is a feature
Limitations force clarity. A 10-slot daily task list forces prioritization. A fixed canvas size forces composition decisions. Embrace constraints in your tools and your process — they produce better outcomes than infinite flexibility.
The Phases
1Planning Phase
Know exactly what you want before you build.
The "one more feature" trap is real. Scope creep during building leads to inconsistent architecture, half-finished features, and wasted context. Do your discovery before you start.
Research first
- • Are the technologies/versions I'm using current?
- • Is what I'm asking for even possible?
- • What are the common patterns for achieving this?
Specify clearly
- • Write technical specifications for complex features
- • Create user stories with acceptance criteria
- • Describe what "done" looks like
Ask for multiples
Never accept the first answer. "Give me 3 ways I could implement X" — seeing alternatives reveals tradeoffs.
2Building Phase
Structure your communication for AI comprehension.
AI has been trained on hierarchical, well-organized text. Work with that, not against it.
Organize your prompts
- • Group related information together
- • Use clear hierarchical markers (#, XML tags, numbered lists)
- • Put the most important information first
Work incrementally
- • Build one feature completely before starting the next
- • Verify each piece works before adding more
- • Commit working states frequently
When stuck, reset
Constant errors create patterns the AI learns to follow. If you're going in circles, start a new session with lessons learned.
3Reviewing Phase
Use what you build. Document what you learn.
Test like a user
You have to actually use your app. Clicking through your own creation reveals flow problems, friction points, and features that work but feel wrong.
Balance form and function
Tech specs ensure the app functions correctly. READMEs ensure it's understood. Both need attention.
Leave a trail
Document what you asked for and why, what worked and what didn't, decisions made and alternatives rejected.
The Two-Document Trellis
Before I write any code, I create two documents. Together, they form the complete trellis — one side for the machine, one side for the human.
TECH_SPECS.md
The Machine Side — tells the LLM exactly what to build.
- • Data models and relationships
- • API endpoints and contracts
- • Component hierarchy
- • State management approach
- • Technical constraints
README.md
The Human Side — describes what users can do and how.
- • User flows and navigation
- • Feature descriptions (user perspective)
- • What each screen accomplishes
- • The "why" behind the product
TECH_SPECS.md README.md
│ │
│ ┌────────────┐ │
└──▶│ The App │◀──┘
└────────────┘
Function + Form
(what it does) (how it feels)The tech spec ensures the app works correctly.
The README ensures the app works as intended.
Quick Reference
| Phase | Key Question |
|---|---|
| Planning | Do I know exactly what I want? |
| Building | Am I working in verifiable chunks? |
| Reviewing | Have I used this myself? |
| Always | Is my context helping or hurting? |
"Your principles are the trellis. AI is the accelerated growth."