The Breaking Point Architecture: Managing AI's Attention Limits

Architect for AI attention limits: know when to switch tools as complexity grows, and add evals plus regression checks to catch breakage.

Sep 29, 2025 6m 1s

Chris Rickard

Expert Insights

Chris Rickard proposes a crucial shift in thinking about AI-assisted software development, introducing the idea of a 'breaking point architecture'. This philosophy emphasizes the balance between optimization, system requirements, and feature deliverability.

Intriguingly, Chris suggests that this dance between AI and human input creates a fresh breed of job roles where human expertise navigates complexity when AI reaches its limits.

Listen to Chris delve into:

The phenomenon of AI 'attention' and how overloading it can lead to reduced accuracy.
The introduction of a 'breaking point architecture' and the strategic shift between tools at optimal complexity levels.
The emergence of new job categories and the need for developers to deftly manage transitions between AI and human expertise.
The concept of fractional development and a shift to AI editors.
The importance of a hybrid approach combining static analysis and AI in managing extensive codebases.
A sober view of AI's capability limits, advocating caution for systems where security is paramount.

This system here has 70 features, but we've interacted with an LLM about 5,000 times in order to get this information here. So for a user to do that themselves, that would mean writing 5,000 prompts to an LLM. I told it, hey, come on AI... let's get it working and let's say maybe it fixes it, but it might actually break something else on this screen. So you're taking one step forward, two steps back. You can go from zero to 80% really fast. And then there's other ways to fill up that 20%... You eject from these platforms and you use something like Cursor or Windsurf.

— Chris Rickard, Founder, UserDocs

THE NEW DEFAULT angle

Here are some actionable takeaways for teams adopting AI-assisted workflows:

Understand AI's attention limits: Educate your team on how AI models pay 'attention' to context and the impact on accuracy. Avoid overloading AI with too much information.
Embrace the breaking point architecture: Align your workflow to manage project complexity with different tools. Know when to transition from AI to human expertise.
Identify new job roles: Find opportunities to optimize skills and expertise within your team to navigate complexity where AI falls short. Drive a balance between AI and human intervention.
Adopt fractional development: Work to understand that AI can streamline 80% of the process, while the remaining 20% can be managed with AI editors such as Cursor or Windsurf.
Implement a combined approach of static analysis and AI in handling extensive codebases. Pre-processing relationships before AI interaction can help manage attention limits.
Be aware of AI limitations: Acknowledge the importance of applying caution when it comes to systems with major security implications. In such cases, a combination of human governance and AI mechanics is crucial.
Prepare for transition: Develop a clear understanding of the breaking point to know when to switch tools while maintaining optimal performance and accuracy.
Start with high abstraction tools and progressively move lower as complexity increases. The right orchestration of tools will help build an efficient, AI-powered software development process.