Update from the Agentic AI Dev Team

AI image generation is not quite there with text, is it?

Making Agentic AI Work for Web UX: A Brutal Reality Check

We're deep into the weeds of building an Agentic AI that can handle web user experience (UX) tasks—think analysing interfaces, navigating design systems, and optimising flows. It sounds futuristic, but here's the unvarnished truth: it's bloody hard.

The Problem: AI Can’t See What You See

At the issue's core is this: our AI can't truly "see" the web like a human does. It doesn't grasp visual hierarchy, spacing, or the subtle cues that make a design intuitive. It processes the DOM, maybe some screenshots, but lacks visual perception. This makes tasks like evaluating UX or suggesting design improvements a massive challenge.

Prompt Engineering: A Game of Trial and Error

Crafting effective prompts is more art than science. A slight tweak in wording can lead to vastly different outcomes, and there's no guarantee of consistency. It's a constant cycle of:

1. Designing a prompt

2. Testing the output

3. Analysing the results

4. Refining the prompt

This iterative process is time-consuming and often frustrating, but it's the only way forward.

The Silver Lining: Incremental Progress

Despite the challenges, we're making headway. Each iteration brings us closer to an AI that can understand and improve web UX. It's not about giant leaps but steady, incremental progress. We're learning what works, what doesn't, and how to guide the AI more effectively.

Looking Ahead

We're not there yet, but the path is becoming clearer. With continued effort and refinement, we're optimistic about developing an Agentic AI that can genuinely enhance web user experiences. It's a tough journey, but one worth undertaking.

Stay tuned for more updates from the trenches.