Dako's AI Builder Notes
⌘K
中EN
System
Color theme
Changsha HubOffline
大可 Dako
大可 Dako大可 DakoAI Builder · Agent PractitionerTurning AI into working software. Practical notes on agent engineering, vibe coding, and commercial delivery.
Platform links coming soon
Hand AI the repetitive; keep the critical for yourselfAI's 70%scaffolding, tests, docs, refactors,boilerplatefast, strong patternsrepetitive and predictableYour 30%architecture, complex logic, security, review,trade-offsjudgment can't be replacedsets the quality barAI owns volume; you own the critical
The split isn't by percentage; it's by whose judgment is worth more.

The 70/30 Split: What Goes to AI, What Stays with You

About 70% to AI—scaffolding, tests, docs, refactors; 30% stays with you: architecture, complex logic, security, review.

Is everything right for vibe coding? No. Judging "which task goes to AI and which stays with me" is the key move from beginner to senior. The 70/30 split is a useful place to start.

After reading, you should be able to answer:

  • What kinds of work is AI especially good at, safe to hand off?
  • What work must stay with you, AI can't replace?
  • Why is staring at the output line by line both tiring and wasteful?

The 70% for AI

AI is good at "patterned, predictable, high-volume" work:

  • Scaffolding / boilerplate
  • Writing tests
  • Writing docs / comments
  • Refactoring (when rules are clear)
  • Format conversion (JSON ↔ types, XML → structure)
  • Regex and one-off scripts

These are "fast output, strong patterns, nothing important"—put AI there, and keep your energy for real judgment.

The 30% staying with you

What AI isn't good at, or shouldn't decide for you:

  • Architecture decisions (should we go microservices)
  • Technology selection (which framework)
  • Complex business logic (pricing, rules, boundaries)
  • Security threat modeling
  • Competitor / requirement judgment
  • Key code review

These "judgments can't be replaced." Handing them to AI isn't impossible—it's inadvisable. Your judgment is where the value comes from.

The ratio is a direction, not an iron law

70/30 isn't a hard ratio; it's a sense of direction:

  • AI takes the high-volume, repetitive, verifiable work google-patterns
  • You keep the judgment, trade-offs, quality, and boundaries

The real question isn't "how much of my work did I give to AI," it's "do I want to outsource this judgment too?" The ones that are yours to make stay yours.

Don't review output line by line

  • Line by line is tiring, inefficient, and easy to miss the forest for the trees
  • Spend review energy on "the boundaries AI handled, the impact of changes, the structural gaps it left"
  • Let AI go back and fix details; you manage direction and trade-offs

The 70/30 split frees your hands to spend time on real judgment. That split finally lands in actionable checks: the review checklist.

References

  1. 4 engineering patterns behind the strongest AI agents (Google)