Engineering 2.0: Moving at the Speed of AI (Without Losing the Human Touch)

At Interactive Labs, we don't just use AI; we’ve woven it into the very fabric of our development lifecycle. Our goal is a new operating model: Build Smaller, Ship Faster, and Recover Instantly.

However, "faster" doesn't always mean "better" if you lose sight of the architecture. Here is how we are partnering with our clients, like UTR Sports, to define the next generation of software engineering while navigating the unique "gotchas" of generative AI.

‍

01. The AI-First Workflow: Our New Standard

We have moved away from the traditional "spec → code → review" model. Today, every feature starts with AI-assisted exploration and planning.

• Architecting, Not Just Typing: We use Claude Code and Cursor to generate first-pass implementations, allowing our engineers to act as architects who guide and refine the logic rather than manual typists.

• The Power of MCP: Through the Model Context Protocol (MCP), our AI tools connect directly to Figma. We can now translate design specs—colors, dimensions, and typography—directly into React Native code in minutes.

• Automated Guardrails: Every Pull Request (PR) is now subject to "Line Call"—an automated AI review stage that scores risk and flags performance or security concerns before a human even sees the code.

02. The "Gotchas": Why Human Oversight is Non-Negotiable

While the UTR Sports vision is heavily AI-leaning, we remain cautious and "pro-human" because AI has distinct limitations. Speed is only an asset if the direction is correct.

The Risk of "Speculative Implementation"

In a recent project—an At-Home Phlebotomy scheduling flow—AI generated code that was syntactically perfect and passed all lint checks, but was structurally wrong. It created screens that didn't exist in the design and used the wrong API endpoints because it lacked "full system understanding".

The Lesson: AI can miss deeper business context and edge cases, leading to "negative time savings" if a full rollback is required.

Avoiding Architectural Drift

Without strict boundaries, AI may:

• Over-generate: Modifying more files than necessary or introducing redundant logic.

• Deviate from Patterns: Introducing coding styles that don't match the existing architecture.

03. Our Rules for the AI Era

To reap the 10x benefits of AI while avoiding the pitfalls, we follow a strict discipline:

• Small, Focused PRs: We no longer submit 1,000-line PRs. Small, incremental changes are easier for AI to review and easier for humans to revert.

• Context is King: We use .cursor/rules and CLAUDE.md files to ensure AI understands our specific architecture and standards before it writes a single line.

• Feature Flags as a Safety Net: We ship everything behind feature flags. This allows us to "unship" or disable a bad feature in seconds without a new deployment.

The Future: The Compounding Flywheel

By combining Smaller PRs, Better AI Reviews, and Accurate Risk Data, we are creating a flywheel of higher velocity. We aren't replacing engineers; we are making every engineer dramatically more capable.

The goal isn't just to write code faster—it's to build a more resilient, transparent, and high-quality system for our clients.