AI isn’t a future concept for us—it’s how we build software every day.
At our company, AI is deeply embedded into our development workflow. We use it to move faster, reduce costs, and improve security and code quality, while still keeping humans firmly in control of every decision that matters.
This post explains:
- The AI tools our developers use
- How our “vibe coding” workflow works step by step
- Why this approach dramatically shortens development timelines without sacrificing quality
The AI Tools Our Developers Use
Here’s a snapshot of how some of the developers across our team use AI-powered tools today.
| Name | AI Tools | Notes |
| Developer 1 | GitHub Copilot | Copilot Pro free trial |
| Developer 2 | GitHub Copilot Pro Google Antigravity | Models: Claude Opus Claude Sonnet Gemini Pro |
| Developer 3 | VS Code + Cline | GitHub Copilot plus |
| Developer 4 | Github Copilot | Model: Claude Code |
| Developer 5 | VS Code + Cline | GitHub Copilot plus |
| Developer 6 | Deepseek | Model: Deepseek |
| Developer 7 | GitHub Copilot Pro DeepSeek | Models: Claude Sonnet DeepSeek |
Rather than forcing a single AI tool across the organization, our developers use the tools and models that best fit their workflow, while maintaining consistent code review and quality standards.
Sample AI-Driven Feature Development Workflow (“Vibe Coding”)
Below is an overview of a typical feature development process using vibe coding, an AI-assisted but human-verified approach to building software.
- Prepare requirements
- Gather and provide UI screenshots.
- Describe feature details and requirements to Claude Code or another LLM.
- Share as much project context as possible, including:
- Relevant frontend/backend service folders
- UI component folder
- Related MS SQL schema scripts
- AI-assisted planning
- Claude Code or another LLM summarizes the development feature plan, which usually includes:
- New or updated database schemas (e.g., dbo.EquipmentFailureReport)
- Stored procedures for backend services
- Frontend/backend service updates (e.g., failureReportService, REST endpoint managementFailureReports)
- UI components (e.g., FailureReportDialog.tsx)
- If any step is not as expected, we discuss and iterate with Claude Code or another LLM to refine the plan.
- Review and implement
- Review the proposed plan, then allow Claude Code or another LLM to proceed with the implementation.
- For each code change:
- Manually review and accept changes.
- If changes are not as expected, reject and provide further instructions.
- Occasionally, make manual edits as needed.
- The more the LLMs evolve, the less manual edits we are making.
- Finalize
- Review all changes and thoroughly test the new feature.
Saving Time and Money
With this approach, developing a typical feature takes about 1.5–2 days.
Using a traditional development workflow, similar features would usually take about a week.
That’s a 3–5× improvement in speed, without cutting corners.
The key insight isn’t “AI writes code for us.”
The real advantage is AI as a highly contextualized collaborator that:
- Accelerates planning
- Reduces boilerplate work
- Surfaces edge cases earlier
- Allows developers to focus on judgment, architecture, and quality
Don’t Forget! The most important point is to provide as much context and detail as possible—the more context the LLM has, the better and more accurate its output will be.
Faster, Cheaper, and More Secure—With Humans in Control
AI doesn’t replace our developers. It amplifies them.
Every line of code is reviewed by AI security scans, dedicated security scanning products.
Every architectural decision is human-approved.
Every feature is tested before release.
Every feature is PEN tested before deployment to production.
The result is software that ships faster, costs less to build, is more secure, and benefits from multiple layers of human and AI intelligence working together.