Everyone’s always complaining about AI being unreliable. Sometimes it’s brilliant, sometimes it’s garbage. But most people are looking at this completely wrong.
The issue isn’t really the AI model itself. It’s whether the system is doing proper context engineering before the AI even starts working.
Think about it - when you ask a question, good AI systems don’t just see your text. They’re pulling your conversation history, relevant data, documents, whatever context actually matters. Bad ones are just winging it with your prompt alone.
This is why customer service bots are either amazing (they know your order details) or useless (generic responses). Same with coding assistants - some understand your whole codebase, others just regurgitate Stack Overflow.
Most of the “AI is getting smarter” hype is actually just better context engineering. The models aren’t that different, but the information architecture around them is night and day.
Wrote up the technical details here if anyone wants to understand how this actually works: <a href="https://open.substack.com/pub/diamantai/p/why-ai-experts-are-moving-from-prompt?r=336pe4&amp;utm_campaign=post&amp;utm_medium=web&amp;showWelcomeOnShare=false" target="_blank" rel="noopener nofollow">link to the free blog post I wrote</a>

Stay at the forefront of AI with full access to my newsletters on cutting-edge techniques, exclusive tutorials, and extensive archives. Keep up-to-date with our open-source projects: RAG Techniques, GenAI Agents, and Prompt Engineering, and more

DiamantAI

AI system reliability depends more on context engineering than the underlying models themselves. Context engineering involves providing AI systems with relevant conversation history, data, and documents before processing requests, rather than relying solely on individual prompts. This approach explains why some AI applications excel (like context-aware customer service bots that access order details) while others fail (generic response systems). The perceived improvements in AI intelligence often stem from better information architecture and context management rather than advances in the core models.

Why AI Experts Are Moving from Prompt Engineering to Context Engineering

I think that’s a very good point. I am also constantly aware of the challenge of how to provide AI with the appropriate context. Although I haven’t used kiro yet, I find the approach of writing design documents and moving forward very interesting. Even if AI takes over the code implementation, I think the task of writing design documents will remain, and I believe that reading books and learning, as mentioned in other articles, will continue to write a good context.