Coding assistants and Sahasa.AI
A coding assistant writes tests. It doesn't tell you whether you can ship.
Claude Code, Copilot and Cursor are very good at writing code, test code included, and your engineers probably use one already. So it's natural to ask whether handing one to your testers solves the testing problem. It solves part of it: the typing. The harder questions, like what to test, whether the tests prove anything and what actually ran, are still left with whoever wrote the prompt.
The same request, twice
"Write tests for the refund approval rule."
Both sides produce tests quickly. Only one of them tells you which of those tests would catch a broken approval rule.
You
Write tests for the refund approval rule in refund_handler.py and run them.
Assistant
I've added 12 tests in test_refunds.py covering approvals, limits and edge cases, and run them.
$ pytest tests/test_refunds.py 12 passed in 0.41s
All 12 tests pass. The approval logic looks correct.
What the session doesn't show
Four of the tests assert nothing, the boundary at exactly 10,000 is never tried, and nobody knows which requirement each test covers.
- 12 tests ran in the sandboxmeasured
- 4 tests have no assertionprove nothing
- Boundary at exactly 10,000 not coveredgap
- REQ-13 "no self-approval" has no testuncovered
PAY-114 · SHALLOW · NOT VERIFIED
What a release gate asks
Eight questions a session can't answer on its own.
None of this is a flaw in the assistants. They're built to help one person write code. These questions are about a whole team, over time.
01
Who decides what gets tested?
With a coding assistant
Whoever writes the prompt, one test at a time. Coverage depends on what they thought to ask for.
With Sahasa.AI
The requirements. Scenarios and cases are derived from them, and the platform measures coverage against every one.
02
Does anyone check the tests are real?
With a coding assistant
The person reading the output, if they have time. A test that asserts nothing still passes.
With Sahasa.AI
The platform. Every function and selector is resolved against your code and live app, and assertion quality is gated.
03
What actually ran, and what didn't?
With a coding assistant
Whatever was run in that session, on that machine. When the session ends, so does the record.
With Sahasa.AI
Every result says whether it ran, and the history is kept per repository and per product.
04
Who checks the code the assistant wrote?
With a coding assistant
Often the same kind of tool that wrote it, prompted by the same person.
With Sahasa.AI
An independent read of the diff: twelve categories of bugs with file, line and a fix, and the blast radius of the change.
05
Is security covered?
With a coding assistant
Only if someone asks, and knows what to ask for.
With Sahasa.AI
Code scanning, dependency CVEs, secrets, DAST and a threat model on every change, whether or not anyone asked.
06
Is it the same standard across 200 engineers?
With a coding assistant
It varies with each person's prompts, habits and model.
With Sahasa.AI
One set of checks for everyone, with the model for each module chosen centrally and recorded on every run.
07
Can a manual tester use it?
With a coding assistant
Only if they can read and write code, and judge whether the code is right.
With Sahasa.AI
Yes. Requirements to testability scores, scenarios, prioritised cases and a plan, with no code involved.
08
Can an auditor follow it?
With a coding assistant
There's a chat history, if someone kept it.
With Sahasa.AI
A traceability matrix from requirement to test to code, each line traced to a commit, and a project dashboard over time.
Why now
AI now writes a lot of the code. Checking it is still a manual job.
84% of developers use or plan to use AI coding tools, and 3% say they highly trust what those tools produce (Stack Overflow Developer Survey 2025). Reviewing it still falls on the same people, and adding another assistant to write the tests adds more output to review. What's missing is the step that checks it.
Keep your assistant
Use both. They do different jobs.
Sahasa.AI isn't a replacement for Claude Code, Copilot or Cursor. Your engineers keep writing with them. Sahasa.AI is the independent check on what they write, before it merges and before it ships.
Your coding assistant
- Writes application code and first-draft tests
- Works inside one engineer's session
- Moves as fast as the person using it
Sahasa.AI
- Decides coverage from your requirements
- Checks every test, finding and citation against your code
- Keeps the record: labels, history, traceability
Together Next
- Sahasa.AI callable from inside Claude Code, Cursor and Copilot
- The same checks run where engineers already work
- Nothing to switch between
Next step
Bring one repository.
We'll walk you through Sahasa.AI on a sample project and talk through how it would fit your stack. If it looks useful, we can scope a pilot together.
- 01A short call. Your stack, your frameworks, and where quality hurts today.
- 02A demo. The platform on a sample project, and how it would connect to your tools.
- 03A pilot, if it fits. Scoped with you around one service.