CONTEXT ENGINEERING FOR BETTER AI - CONTEXT IS THE NEW PROMPT

Harness context engineering: control AI model's context for cleaner coding and progressive task disclosure for efficiency gains.

Dec 16, 2025 5m 30s

Krzysztof Zablocki

Expert Insights

Krzysztof Zablocki advocates for the power of "context engineering" in AI-assisted software development, arguing that this approach can streamline workflows and improve the end product. He implies that careful control over the model context—as opposed to simply focusing on prompts—can increase efficiency and lead to better project outcomes.

Krzysztof argues that successful AI implementation comes down to controlling what information the model learns and how it sees it. He champions keeping the model's context as clean and straightforward as possible and adaptively feeding policies based on the task at hand - a concept he terms as "progressive disclosure".

Hear Krzysztof explain:

The benefits of context over prompt-based engineering: Emphasizing that less cluttered and unnecessary information in the context can lead to better language models.
Progressive disclosure: How tailoring the context based on the specific tasks can enhance the development efficiency.
The role of "skills": How high-level task descriptions can guide the AI model without overwhelming it with detail.
The pitfalls of poor context engineering: Highlighting that inadequate model guiding can lead to inefficient and counter-productive code.
The need for new conventions: With AI-evolving rapidly, he underscores the need to establish standards to facilitate context engineering.

It used to be prompt engineering, and now it's like the context window engineering. I think that's one of the biggest things.

Monterail Team Analysis

Here's how development teams can tap into the context engineering in AI-assisted workflows:

Encourage shift from prompts to context: Promote a development ethos that values context management, shifting focus from prompt-based to context-based AI engineering.
Employ progressive disclosure: Leverage the strategy of progressive disclosure by loading only the necessary rules for a given task, thereby optimizing token usage and ensuring efficient code production.
Experiment with sub-skills: Use high-level task descriptions to guide the AI but prevent information overload. This could enhance the new "AI at the core" building process, making it more manageable and efficient.
Formulate context guidelines: Create a set of standards regarding the use and management of context in AI development. This can ensure consistency and improve overall effectiveness.
Incorporate resets between tasks: Regularly reset the model context to avoid redundancy and ensure lean, targeted, and accurate model responses.
Consider meaning of failure: Evaluate failures not as shortcomings of the AI model, but as an indication of a need for improved context engineering. Use these as learning opportunities to refine the guidance and training process.
Stay agile with conventions: With AI evolving rapidly, it's crucial to remain proactive, flexible, and adaptable regarding the establishment and modification of conventions in context engineering.