Guardrails don’t slow you down — they let you go faster without crashing.

Next up in the series: “Make It Work, Make It Good — Pairing With Coding Agents.”

Same loop, different mindset. 🚀

Example:

# .git/hooks/pre-commit
ruff check --fix .
swiftlint
pytest -q -m "not slow"

Quick, local, automatic.

Don’t skip them. And never ever let your agent sneak in --no-verify.

That’s the “seatbelt click” before takeoff.

Guardrail #1: Pre-commit hooks.

I used to rely on CI after push. Too late — the mess was already in history.

Now I run lint + fast tests before every commit.

Local checks keep the refactor loop safe and fast.
No broken states. No surprises.

Feels less like “auto-coding” and more like collaborating with AI colleagues.

Full write-up (with diagrams + prompts):
👉 novibe-vibe-coding.hashnode.dev/building-a-m...

What I’ve noticed so far:

Agents keep context across steps (no constant resets).

They challenge each other, surfacing blind spots.

The workflow creates natural checkpoints where I step in, give feedback, and steer things.

Wrote up the full experience with lessons learned: novibe-vibe-coding.hashnode.dev/when-coding-...
#AI #SoftwareEngineering #Testing #CodingAgents

Key takeaway: AI coding agents are powerful tools, but they're not yet reliable enough for complex refactoring without close supervision. The tendency to declare victory prematurely is a real pattern developers need to watch for.

The agent even started marking failing tests as "skipped" instead of actually fixing them. That's like putting duct tape over a check engine light and calling the car fixed!

Just applied this to a server shutdown system—went from scattered logic across 4 files to clean, consolidated code. 75% reduction in lines.
Turns out AI is bad at preventing code bloat but excellent at identifying it after the fact.
novibe-vibe-coding.hashnode.dev/the-two-phas...

Google Docs: if the agent writes a doc (design doc, implementation plan ...) same thing: leave comments in the Google docs UI and it picks it up from there.

Anybody knows of such agents or adapters?

But this week, when the President of South Africa visits the US, he accuses him of “genocide” and “white rasism”.

So, we should overlook harassment of woman or suppressing minorities. But we draw the line when white people are the ones affected.

I'm not looking for a coding agent that makes the tests green.
I want one that understands why the test is there.

A development agent.

#AI #DevAgent #TestingMatters #SoftwareEngineering