Our development workflow at a major tech company using AI tools

Core Development Concepts
1

What’s up everyone! I keep seeing people say that AI coding tools can’t be trusted for real production code, but that’s just wrong.

I’m a software engineer with about 10 years under my belt. Been working at big tech companies for the last 5 years. Started as a systems guy but moved into development, though I’ve been coding for around 15 years total.

Here’s our current process for building features with AI assistance:

Step 1: Create technical specs document. This is where most planning happens. Start with a basic proposal and get key people to sign off on it.

Step 2: Design review meeting. Senior devs tear apart your design doc. It hurts but catches problems early.

Step 3: Build phase starts. Spend time documenting each piece that different dev teams will work on.

Step 4: Task breakdown with product managers. Figure out what each developer works on and in what sequence.

Step 5: Actual coding begins. We do test-driven development, so AI agent writes test cases first for my feature. Then I use it to build the actual functionality.

Step 6: Code review process. Need two other devs to approve before merging to main branch. AI is getting better at helping with reviews too.

Step 7: Deploy to staging environment first, then production.

We’re seeing about 30% faster delivery from idea to live feature. Pretty significant improvement.

Bottom line: Good architecture planning upfront is key. Build incrementally. Tests come first, always.

2

that’s really interesting! what AI agent are you using for test cases? im curious how it handles edge cases humans might miss. do code reviews catch more AI-generated bugs, or does it actually help reviewers spot issues better?

3

Same setup here, but our senior architects fought us hard on it at first. We hit that 30% improvement too after about six months. Biggest headache was keeping code consistent when everyone uses AI tools their own way. Fixed it by setting AI guidelines during design reviews and building shared prompts for stuff we do all the time. The test-driven piece you mentioned is huge - AI tests catch requirement gaps we’d never spot otherwise. Staging caught tons of integration problems that would’ve blown up in production with traditional dev.