sethserver.com Subscribe
A retro-style orange and teal robot with a corrugated arms and legs dropping boxes into a hole

What Does CI/CD Actually Buy Us

By Seth Black • Updated: March 30, 2026

Programming · 4 min read

Like this kind of writing? Get one email a week with notes on startups, AI, and the occasional strong opinion about Python: subscribe to the newsletter.

You can tell a lot about a startup by how they deploy.

If the deploy process starts with "SSH into the box and run git pull," you're not doing DevOps. You're doing a ritual. And the ritual usually depends on The One Dev Who Knows The Script. That dev might be you. Congrats. I too have been that single-point-of-failure.

CI/CD shouldn't be fancy it should be stable.

CI

CI (Continuous Integration) is simple:

The key word is trusts. Not "trusts the developer." Trusts the system.

I trust backups more than I trust people. Test cases more than promises. Logs more than memory. That's how you make the wrong call and can afford to fix it. That's most of startup life.

CI is "trust but verify" as code.

I had Claude whip up a basic GitHub Actions example for me:

## .github/workflows/ci.yml
name: ci
on: [push, pull_request]
jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.11"
      - run: pip install -r requirements.txt
      - run: ruff check .
      - run: mypy .
      - run: pytest -q

Where AI fits (and where it doesn't)

AI assistants are great at generating copious amounts of tests. They will also happily generate tests that assert True == True with a straight face.

So let AI help you write tests. Then make CI enforce them. The pipeline doesn't care if the test is good. It just runs it. That's why you need both AI and CI: AI can draft the test, CI makes sure it actually runs (and keeps running). I like this AI+CI path, keep an eye out for a post about this. Weee!

CD

CD (Continuous Delivery / Deployment) is the other half.

Build something - a wheel, a container, a binary, then deploy it automatically when you tag a release. Make the rollout simple: rolling deploys, blue/green, health checks. The goal is to make shipping code simple, stress-free, easy.

The antipattern is generally:

  1. SSH and pray
  2. Write a bash script
  3. Put the script in git
  4. Have CI run it
  5. Stop letting anyone freestyle production

Each step feels like progress until you realize you're still one typo from disaster.

A concrete flow I like:

Even on a small setup (like a single Lightsail box or EKS cluster), this is a big deal. You go from "what changed?" to "we deployed the onboarding flow fixes built from commit 7aec2235." This is invaluable when it breaks exactly 42 days later.

Feature flags are not the same as deploys

Feature flags let you ship code that's turned off, or dark. CI/CD makes sure the code you ship doesn't break everything. Different problems.

Feature flags are for product risk.
CI/CD is for operational risk.

You want both. Otherwise you batch changes for weeks because deploys are scary, and then one deploy breaks six things and nobody knows which commit did it.

Startup anti-patterns that CI/CD exposes fast

1) "We have staging" (but staging is pretend)

If staging doesn't match prod, it's not staging. It's a dev environment you're lying to yourself about.

The pipeline will still run. Your deploy will still "work." And then prod will explode because prod has different env vars, different magical scripts, different data shape, different network rules, different "all the things".

2) No rollback plan (and migrations that make rollbacks impossible)

If your rollback plan is "SSH in and undo it," you don't have a rollback plan.

Clean rollback: redeploy the previous tag and done.

Messy rollback: you ran a migration that can't reverse, so now you're doing live data surgery while getting "what's the status, bud? need any help?" That's the gray hair special.

3) Communication failures dressed up as engineering

CI/CD doesn't fix incentives. It does make it harder to hide behind rehearsed answers. The build is green or it's not. The deploy happened or it didn't. Logs exist or they don't.

The quick comparison

Practice Without CI/CD With CI/CD
Tests Maybe run locally Always on every push
Deploys Manual, brittle Scripted, auditable
Rollback Hope and SSH Re-deploy previous artifact/tag
AI's role Writes code and "tests" Pipelines gate what reaches prod

What CI/CD actually buys you

It buys you the ability to ship any day, any time without worry. And if you're the One Dev Who Knows The Script, CI/CD means you can go to dinner without wondering if production is on fire.

Because it's not about crushing success, it's about recovering from catastrophic failure. And stable, automated pipelines mean you can recover faster—because the system just works.

Now, go forth and build something stable!

-Sethers

Share this post
Newsletter

One email, once a week.

Notes on databases, systems, and the occasional strong opinion about Python. No spam, unsubscribe anytime.

Seth Black
Written by

Seth Black

Engineer and founder based in Texas. Writes about databases, AI, and running things in production. Embeds in small teams as lead engineer.

More from Programming

View all →
Programming

What's the Differences Between Relational, NoSQL, Vectors, and Caches

Apr 14, 2026
Programming

Git and Source Control in AI Land

Apr 14, 2026
Programming

Domains, DNS, and why it feels fragile

Apr 14, 2026