"I think this approach is better."
Is it? Based on what?
Without evidence, opinions stay opinions. You can argue endlessly. Different people have different intuitions. Nobody knows who's right.
AI makes this worse. AI can generate unlimited approaches, all plausible. Which is actually better? Your gut? Someone else's gut?
Evidence is how you know.
The Pre-AI State
Traditional development had implicit evidence: the code worked or it didn't. The system handled load or crashed. The feature shipped or it didn't.
Evidence was embedded in outcomes. You didn't need to explicitly measure because reality provided feedback.
The AI Problem
AI produces plausible output. It looks right. It might be right. You can't tell by looking.
AI produces unlimited output. You could try ten approaches in an afternoon. All seem reasonable. Which is actually best?
AI produces confident output. It doesn't hedge. It doesn't say "I'm uncertain about this." Everything sounds authoritative.
Without explicit evidence, you're navigating by vibes in a sea of plausibility.
What Evidence Looks Like
Evidence connects claims to reality:
**Hypothesis:** "This prompt produces better output than the old one." **Evidence:** Side-by-side comparison on 50 inputs, rated by users. New prompt scored 4.2/5, old prompt scored 3.6/5.
**Hypothesis:** "This architecture will handle our scale." **Evidence:** Load test results showing X requests/second at Y latency.
**Hypothesis:** "Users prefer this approach." **Evidence:** A/B test with 1000 users. Version A: 12% conversion. Version B: 8% conversion.
Evidence is specific, measurable, and connected to the claim it supports.
Incorporating Evidence
**Before building:** What's our hypothesis? What evidence would confirm or disprove it?
**During building:** Are we generating evidence as we go? Test results, comparisons, measurements?
**After building:** What happened? Does evidence support our hypothesis? What did we learn?
This isn't heavyweight process. It's asking "how will we know if this works?" and then actually checking.
The Documentation Connection
Evidence should be captured and connected:
- Link decisions to the evidence that informed them
- Link hypotheses to the outcomes that tested them
- Preserve evidence so future decisions can reference past learning
Without this connection, evidence is ephemeral. You had data once. Now it's gone.
The Cultural Shift
Some teams resist evidence:
- "We don't have time to measure"
- "Intuition is faster"
- "Testing everything is overkill"
But consider: how much time is spent debating opinions that evidence would resolve? How many bad decisions happen because nobody checked?
Evidence isn't extra work. It's work that prevents wasted work.
Start Simple
You don't need elaborate measurement infrastructure.
For your next decision: - What's the hypothesis? - What would confirm it? - Go get that confirmation.
Write it down. Reference it later.
That's it. Evidence-based development, minimum viable version.
Build from there.