AI Coding Agent
An agent that writes code, runs the tests, reads what went wrong and fixes it, round and round until everything passes. Pick a task and watch the loop.
New here? Read the short guide for this demo, with steps to follow and what to look for.
The idea in one minute
Think of a new developer who checks their own work.
A good developer does not just write code and hope. They make a plan, write a first version, run the tests, read what failed, and fix it. Then they run the tests again.
A coding agent works the same way, in a loop: plan, act, observe. The trick is the observe step. Real test results tell the agent what to fix next, so it does not have to guess.
- 1Pick one of six everyday tasks.
- 2Watch the agent write, test and fix.
- 3Try to break the final code with your own test.
This is a demo with made-up data, and it runs only in your browser. No AI model is called: the plan, the code and the agent's notes are written in advance. The tests are real, though. The code really runs in your browser, and every pass, fail and error message is worked out live. Nothing you type is sent anywhere or stored.
1 Pick a task
The tests it has to pass
2 Watch the loop
Agent notes
Code changed lines are highlighted
Test results worked out live in your browser
Trace: every action the agent took, in order
3 Try to break it
Let the loop finish first. Then you can run the final code on your own test case.
Enter your own values and run the final code on them. The checks below the result are worked out from your values, so you do not need to know the answer.
How the loop works
Plan, act, observe.
- Plan. Read the task and its tests, and list the steps.
- Act. Write the code, or change only the lines that need fixing.
- Observe. Run the tests and read exactly what failed. That decides the next fix.
- Guard rails. Each run is stopped after 1 second, and the agent gets at most 4 attempts before a person takes over.
Honest limits
What this demo does not do.
- The plan, the code and the notes are written in advance, so the agent cannot take on a new task. A real agent asks an AI model to write each attempt.
- If your own test finds a bug, the demo stops there. A real agent would add your case to its tests and loop again.
- The tasks are small on purpose. Real GST rules, for example, also depend on the place of supply and the type of goods.
Have an idea? Let's build it properly.
Tell us what you want to build or automate. We'll reply with practical next steps.