Estimate and scope

Estimate and scope

How to size work and decide between approaches when AI compresses the cost of completeness.

Source: gstack ETHOS.md, CLAUDE.md, ship/SKILL.md.

Boil the Lake

"AI-assisted coding makes the marginal cost of completeness near-zero. When the complete implementation costs minutes more than the shortcut — do the complete thing. Every time."

Lake vs ocean:

• Lake = boilable. 100% test coverage for a module, full feature with all edge cases, every variant of a pattern.
• Ocean = not boilable. Multi-quarter platform migrations, full rewrites of legacy systems.

Boil lakes. Flag oceans as out of scope.

"'Ship the shortcut' is legacy thinking from when human engineering time was the bottleneck."

AI compression in estimates

Always quote both human-time and AI-assisted time:

Type of work	Compression
Boilerplate	~100x
Tests	~50x
Feature	~30x
Bugfix + regression	~20x
Architecture	~5x
Research	~3x

Phrase as: "2 weeks human / ~1 hour AI-assisted."

When evaluating "approach A (full, ~150 LOC) vs approach B (90%, ~80 LOC)" — almost always choose A.

Anti-patterns in estimation

• "Choose B — it covers 90% with less code." (If A is 70 lines more, choose A.)
• "Let's defer tests to a follow-up PR." (Tests are the cheapest lake to boil.)
• "This would take 2 weeks." (Say: "2 weeks human / ~1 hour AI-assisted.")

Search before building

"The 1000x engineer's first instinct is 'has someone already solved this?' not 'let me design it from scratch.'"

Three layers of knowledge

Layer 1: Tried and true — built-ins, official docs, well-established libraries.

"The risk is not that you don't know — it's that you assume the obvious answer is right when occasionally it isn't."

Layer 2: New and popular — recent blog posts, trending patterns, popular libraries.

"Humans are subject to mania. Mr. Market is either too fearful or too greedy. Search results are inputs to your thinking, not answers."

Layer 3: First principles — your own reasoning from fundamentals.

"Prize them above everything else."

Trust ranking: Layer 3 > Layer 1 > Layer 2.

Most engineering culture says "don't reinvent the wheel" by default. This inverts that: "Once in a while, questioning the tried-and-true is where brilliance occurs."

Search rubric

1. {runtime} {thing} built-in
2. {thing} best practice {current year}
3. Check official docs

The eureka moment

"The most valuable outcome of searching is not finding a solution to copy. It's finding a clear reason why the conventional approach is wrong... zig while others zag. When you find one, name it. Celebrate it. Build on it."

"The truly superlative projects are full of these moments — 11 out of 10."

Anti-patterns in scope decisions

• Rolling a custom solution when the runtime has a built-in
• Accepting blog posts uncritically in novel territory
• Picking the smaller-LOC option when the larger one covers more edge cases
• Deferring tests to a follow-up PR
• Quoting estimates in human-time without the AI-assisted equivalent
• Assuming Layer 2 (new and popular) is a strong signal because something is trending

The completeness test

When you find yourself thinking "ship the 90% version":

1. How many lines is the gap between 90% and 100%?
2. What's the AI-assisted estimate to fill it?
3. Is this a lake or an ocean?

If it's a lake and the gap is hours not weeks → boil it.