DESIGNING CONTROL AROUND AI - HOW TO MAKE AI CODE RELIABLE?

Master control over AI chaos: Delegate coding to AI, focus on system design, and establish deterministic checks for reliability and efficiency

Dec 16, 2025 6m 9s

Krzysztof Zablocki

Expert Insights

Krzysztof Zablocki underscores the need for a deterministic control layer around the inherently chaotic outputs of AI in software development. Rather than viewing AI as a means to automate code writing, Krzysztof highlights its power to aid ideation, scale system design, and facilitate rapid prototyping whilst maintaining reliability. He provides a vision for a new AI-enhanced workflow wherein the AI is given the implementation tasks, but the development process ensures each step is vetified.

Here’s what Krzysztof shares:

How delegating coding tasks to AI facilitates developers to focus on important aspects like system design and architecture.
The potential of AI to offer diverse design options rapidly, thereby promoting better decision-making and reducing attachment to a single, time-consuming coding

session.

How introducing AI doesn't just offer faster code writing but fundamentally changes the design and prototyping process, allowing developers to rapidly validate an idea before it's even approved.
The necessity of a control layer around AI to manage its unpredictable nature, ensuring the code generated is usable and efficient. This involves rigorous, system-driven verification at every step.
An emphasis on viewing the code as a means to an end, putting user experience first and challenging the old paradigms of code-attachment.

You always have to build, you always have to test. It's all deterministic.

Monterail Team Analysis

Adopt AI-assisted workflows in software development. Here are practical takeaways:

Embrace AI in code writing: Free up developers' time for critical tasks such as designing robust systems and architectures by delegating implementation work to AI.
Establish deterministic control layers: Given the unpredictable nature of AI, develop systematic checks at every step of the code generation process to maintain reliability and efficiency of the output.
Use AI for rapid prototyping: Beyond simple code generation, leverage AI for creating diverse design options quickly, which can aid in decision-making and faster idea validation.
Shift perspective in code ownership: Reduce attachment to code by viewing it merely as a means to an end. Focus instead on the quality, readability, and effectiveness of the code in providing user satisfaction.
Enable immediate feature testing: Utilize AI's rapid prototyping capacity to build, test, and showcase new features even before they receive formal approval. Use this as a means to garner stakeholder buy-in effectively.
Construct systems that facilitate iterative refinement: Plan your AI-based workflows to foster an environment for continual learning and improvement.
Humanise the review process: Despite heavy AI implementation, ensure that there’s a human overseeing and validating the process for high-stake projects to ensure desirable user experiences.