Intent Engineering: The Patterns of Semantics and Iteration

"Prompt" is a way to guide LLM models into producing specific outputs, and prompt engineering has become the poster child you see all over the news articles and blogs. Though this has garnered significant attention, we may need to look at a more holistic approach. I am thinking of naming it "Intent Engineering."

Consider the relationship between Software Engineering and the 90s poster child Java. While Java is the primary language/tool to get things done, software engineering encompasses a broader spectrum, addressing the entire lifecycle, from requirements gathering to maintenance. Similarly, while prompt engineering focuses on crafting precise prompts to elicit specific responses, intent engineering emphasizes understanding and shaping the new design patterns of this embrace.

Boundaries of Intent

The advent of LLMs symbolizes a seismic shift in our approach to AI. We are transitioning from a world bound by syntax and processes to one driven by semantics and Intent. Much like software engineering has boundaries defined by system requirements, the underlying goals of the task bounds intent engineering. These boundaries are about getting a model to produce a particular output and ensuring that the output aligns with ethical, societal, and domain-specific constraints.

The "Software" Stack around Intent

When we talk about the software stack in traditional engineering, we refer to layers of technology, from hardware to application software. In the realm of Intent Engineering, the stack can morph to have:

1. Objective Layer: This is the core intent or goal. It's analogous to the core functionality of a software program.
2. Contextual Layer:  Here, we add context to the objective. It might involve understanding user preferences, regional differences, or current events.
3. Constraint Layer: This layer sets the boundaries. It could be ethical guidelines, domain-specific rules, or even user-defined constraints to ensure the output aligns with the broader objectives.
4. Feedback Loop: Like software needing debugging and updates, the intent stack requires feedback to refine, improve, and adjust for deviations.

Understanding the Paradigm Shift

Historically, we've viewed programs and interfaces as static entities. If an icon doesn't function as expected, we don't get to change its behavior. However, this notion is shifting. Perhaps you could have expressed your Intent more clearly, or maybe you changed your mind. This iterative approach to interaction will bring novel approaches and patterns. Whether through text chat, voice, gestures, or images, the future promises a more fluid and organic relationship between users and systems, thereby shaping the future of Intent and its engineering practices and patterns around it.

This article attempts to shed light on the evolving landscape of software engineering practice with the advent and availability of this new tool. As always, please put your thoughts in the comments!