SENAR Quick Start: 5 Minutes to Better AI Development

You use AI to write code. SENAR makes the result reliable.

No meetings. No mandatory certifications. These habits are the SENAR Core — 8 rules, 2 quality gates, 2 metrics — everything you need to start. Adoption overhead: under 1 hour, about 5 minutes per session after that.

The 6 Habits

BEFORE you tell the AI what to build

1. Write the WHAT and the DONE

Before starting, write two things:

• Goal — what must be accomplished (one sentence)
• Acceptance Criteria — how you’ll know it’s done (numbered list, each independently testable)

Bad: “Add login functionality” Good: “Implement email/password login. AC: 1. POST /auth/login returns JWT on valid credentials. 2. Returns 401 on invalid password. 3. Returns 422 on missing email field. 4. Token expires in 24h.”

Why this matters: AI output quality equals input quality. A vague goal produces plausible-looking code that fails in production. A precise goal with clear acceptance criteria produces testable, correct code on the first try.

2. Set boundaries

Tell the AI what NOT to touch:

• “Change ONLY the users/ directory”
• “Do NOT modify the database schema”
• “Follow patterns from auth/router.py”

Without boundaries, AI will confidently refactor half your codebase to “improve” things you didn’t ask about.

WHILE the AI works

3. Verify against your criteria, not your intuition

Don’t just glance at the code and think “looks right.” Check each acceptance criterion:

• AC 1: Does POST /auth/login return JWT? → Check the test or run it.
• AC 2: Does it return 401 on invalid password? → Check the test.
• AC 3: Missing email → 422? → Check the test.

If there’s no test for a criterion, the criterion isn’t verified.

4. Document dead ends

When an approach fails, write one sentence about why:

• “Tried bcrypt for password hashing — import fails on Python 3.14, switched to argon2”
• “SQLAlchemy async session: can’t use lazy loading, need selectinload”

This takes 10 seconds and saves hours — for you next week, for your teammate, for any AI that reads your knowledge base.

AFTER it’s done

5. Run the tests

If tests pass AND all acceptance criteria are met → done. If not → not done. No exceptions, no “it probably works.”

6. Capture what you learned

If you discovered something non-obvious during this task, write it down:

• A decision you made and why
• A pattern that worked well
• A gotcha that surprised you

Note — 6 habits vs. 8 Core rules: These 6 habits cover the essence of the 8 SENAR Core rules. The full Core adds two explicit practices: a tiered verification checklist for latent defects (Rule 5) and root-cause analysis before patching symptoms (Rule 7). If you are working in a regulated or high-stakes context, read the full Core document before relying on this Quick Start alone.

Before & After: Real Data

These numbers come from one project (6 microservices, 552 tasks, $989 in AI costs, 38 sessions). They are illustrative, not universal — your numbers will differ. SENAR requires you to establish your own baselines before setting targets.

Without structured process (sessions 1–3, ad hoc):

• Tasks started without acceptance criteria → AI produced code that “looked right” but failed edge cases
• No dead ends documented → same failed approaches repeated across sessions
• No session discipline → 200+ minute marathon sessions with noticeable efficiency decline
• Defects discovered after “done” → estimated rework cost ~$105 per escaped defect (estimated from project data: average rework cost of defects found after task completion)

With SENAR habits adopted (sessions 4–38):

• Every task has goal + AC before start → First-Pass Success Rate improved to 85%+ (tasks correct on first try)
• Dead ends documented → repeated failures eliminated for documented cases
• Sessions capped at 120 min with checkpoints → consistent throughput, no context crashes
• QG-0 blocks taskless work → zero “what was this for?” tasks

Caveat: This is a single team’s experience (N=1), not a controlled study. The improvement conflates methodology adoption with natural team learning. Independent replications are needed — and that’s why we published the standard, so others can measure too.

The cost of SENAR: ~5 minutes session overhead + 1–3 minutes per task. What you avoid: rework from vague requirements, repeated dead-end approaches, marathon sessions with declining output.

That’s It

These habits correspond directly to the SENAR Core rules — the self-contained subset of the SENAR Standard. Everything else in the methodology builds on this foundation.

What you get:

• Fewer “works on my machine” surprises
• AI produces correct code on the first try more often (we measure this — it’s called First-Pass Success Rate)
• Knowledge accumulates instead of evaporating between sessions
• You can hand off to another developer (or your future self) without losing context

What it costs: ~5 minutes session overhead + 1–3 minutes per task for goals, criteria, and verification.

Next Steps