Vibe coding · Day rate
A vibe coding development day: $1,150.
One day of AI-assisted development on your business app, run by a senior engineer under written guardrails. Nothing ships without a test that can fail, a green CI run and a review. AI speed, engineering discipline.
Vibe coding, with guardrails
Vibe coding means letting AI coding agents write the code from plain-language instructions. It is fast; it is only reliable when every behavior is proven by a test.
It is the same method we use on our own production software and on client projects. The tool speeds up writing code; quality comes from the process around it. For why each guardrail exists, read is vibe coding bad?
How the day works
Morning: scope and acceptance criteria
We agree on what ships by the end of the day and write the acceptance criteria before any code. Those criteria become the tests.
During the day: agents build, an engineer decides
Our lead engineer runs AI coding agents (Claude Code): one writes the tests, another the code, others review. Product, architecture and security calls stay with a human.
End of day: a reviewed pull request
One pull request per topic, with a digest: each criterion and its test, the red-then-green run, review findings and the security verdict. Nothing reaches production without your sign-off.
Your repository, your code
The work lands in your repository with its tests and CI. Standard TypeScript and SQL that any developer can pick up, with or without AI.
Eight rules: no behavior ships without proof
A behavior that is not proven by a test capable of failing does not ship. The rules hold whatever AI tool we use: when the tooling changes, the method stays.
1. One verification command
Lint, types, unit, integration and end-to-end tests, and a dependency audit run from a single command. The agent runs it while it works; CI only calls it.
2. The verdict comes from a third party
Every change is replayed on a CI machine. “Works on my machine” from an agent is not a verdict: a merge needs every check green on the latest commit.
3. Every bug fix starts with its test
A fix begins with the test that reproduces the bug. A defect that slipped past CI is labeled as such, so the checks get stronger where they missed.
4. The tester is not the coder
A separate agent writes the tests from the acceptance criteria alone and watches them fail for the right reason. The coding agent may never weaken them.
5. Proof scaled to risk
Access control, money, legal records and outbound messages get tests on a real database, at least 90% coverage of changed code and at least 70% mutation score.
6. A fresh database for every run
Tests run on a database rebuilt from the branch’s migrations, never the live one. A test missing a secret fails; it is never skipped.
7. Three independent reviews
Before each pull request: a code review (does every test prove the behavior?), a blocking security review and a behavior-preserving simplification pass, each by a separate agent.
8. Production check and rollback
After each merge we check the deployed version, key pages and error tracking for 30 minutes. If production breaks, the change is reverted first.
Pricing
Day rate
$1,150 per day (USD)
- Scope and acceptance criteria
- Development by supervised AI coding agents
- Tests written by a separate agent
- Code, security and simplification reviews
- A documented pull request in your repository
- Remote, with overlap during US East Coast mornings
Questions before we start
What can you ship in one day?
One complete feature of a business app: a screen with its forms and rules, a file import, a dashboard, or an integration with a tool you already use. Scope is set in the morning with the acceptance criteria; whatever is not proven by a test by the end of the day does not ship.
Do we need an existing app?
No. A day can start a new app (repository, database, CI, first screens) or move an existing one forward, as long as we can access the code.
Can we book several days?
Yes, that is the usual case. Days can run back to back or be spaced out to match your review pace. Beyond a few days, the work is split into batches, each shipped through its own pull request.
Does our data go through the AI model?
No. Development runs on synthetic or anonymized data, and tests on a database rebuilt for every run. Your real data stays in your application.
How is the day billed?
US$1,150 per day, invoiced by Atlas Software Partners LLC, a US company. No minimum number of days; the scope is confirmed in writing after a 30-minute call.
One day to move your app forward?
Book a 30-minute call. We’ll agree on what ships by the end of the day.