I tested an AI coding agent with a real university project spec — here's exactly what happened, token by token.
Artificial Intelligence is rapidly changing how we build software. Instead of spending hours scaffolding projects or writing repetitive boilerplate code, AI coding agents are now capable of planning, generating, and even refining applications with minimal input.
Recently, I had the opportunity to test Omniroute while helping a friend with her final-year university project. She already had a detailed project specification and needed a working PHP template to kick-start her development. Since I had never worked with Twig templating before, I decided this would be the perfect opportunity to see how far an AI-assisted development workflow could go.
This article shares my experience — from setting up Omniroute to generating a complete project and evaluating the results.
🛠️ Setting Up Omniroute
I already had an old Claude API key with approximately $0.20 of credit remaining, so I decided to use it for this experiment.
To install and configure Omniroute, I followed this setup tutorial:
After following the installation steps, my directory structure looked like this:
⚙️ Improving the AI's Coding Standards
One feature I particularly liked was Omniroute's ability to customise the AI's behaviour through the .agent configuration file.
I configured Claude with modern development guidance so that it would generate code following current best practices instead of relying on outdated programming patterns.
I used the following guide as inspiration:
🔗 How to Stop Your AI Coding Agent from Writing Outdated Code with Modern Web Guidance
Tip: Although this step is optional, I would highly recommend it. Small improvements to the agent's instructions can significantly improve the quality and consistency of the generated code.
📋 Preparing the Project
My friend had already written all of the project requirements, including both the functional and non-functional specifications.
I created a new project folder called wedmauritius inside my XAMPP htdocs directory and copied her requirements into a file named:
newProjectRequirement.txt
Once everything was ready, I simply asked Claude whether it could access and understand the requirements document.
Instead of immediately generating code, Omniroute analysed the project and asked whether I wanted it to create a development plan before executing the implementation.
I found this particularly impressive because it approached the project much like an experienced software developer would — understanding the requirements first before writing any code.
🚀 Generating the Application
After confirming the execution plan, Omniroute began building the project automatically.
The development process required very little manual intervention. It generated the project structure, organised the files appropriately, and implemented the requested functionality based on the specification document.
The final project directory looked like this:
📊 Omniroute Dashboard
When Omniroute starts, it automatically launches its web dashboard. From here you can:
- Monitor development sessions
- Configure quotas
- Review usage statistics
- Manage other application settings
⚠️ Security note: The default login password is
CHANGEME. If you're installing Omniroute for the first time, remember to change this before using it regularly.
✅ The Results
Overall, I was pleasantly surprised by the quality of the generated application.
| Metric | Value |
|---|---|
| Total tokens consumed | ~250,800 |
| Additional API cost | ~$0.50 |
| Manual interventions needed | 1 (CSS fix) |
| Output quality | Production-ready starting point |
Rather than producing a rough prototype, Omniroute generated a solid starting point that my friend could immediately continue developing. For anyone beginning a university project or building an internal prototype, this can save many hours of initial setup work.
🔢 Token Usage
One of the more interesting parts of this experiment was the token consumption.
Although my Claude account had virtually no remaining credit, Omniroute continued the development using available free-tier endpoints to complete the project.
An additional $0.50 worth of tokens was consumed during the process — remarkably cost-effective for what was generated.
🧪 Testing the Generated Website
Once the project had been generated, I tested the application manually. Overall, everything worked remarkably well.
The only issue I noticed was a small CSS problem — the navigation bar was not remaining sticky while scrolling.
Rather than fixing it manually, I simply opened another Omniroute session and asked Claude to correct the issue. It did so without any fuss, which demonstrates one of the biggest advantages of AI-assisted development: rapid iteration.
💭 Final Thoughts
This was my first real project using Omniroute, and I came away genuinely impressed.
What stood out most wasn't just its ability to generate code — it was the workflow. Omniroute analysed the requirements, proposed a development plan, and then executed that plan with very little guidance. That makes it feel less like an autocomplete tool and more like collaborating with a junior developer who can rapidly scaffold an application.
Of course, AI-generated applications still require testing, code review, and refinement. Minor issues — such as the CSS bug I encountered — are to be expected. However, those are small trade-offs considering the amount of time saved during the initial development phase.
If you're looking for a way to bootstrap small PHP applications, university projects, proof-of-concepts, or internal tools, Omniroute is worth exploring.
I don't see it replacing software developers anytime soon — but it certainly makes them significantly more productive.
Have you tried Omniroute or a similar AI coding agent? I'd love to hear about your experience in the comments below.
Top comments (0)