Advanced Prompting Guide for AI-Assisted Engineering

Last year, when we published the original Guide to AI‑Assisted Engineering, most organizations were still in the earliest stage of AI adoption. Leaders were focused on getting developers to try these tools, understand where they helped, and build basic habits around using them. That guide met teams where they were: it offered simple, copy‑and‑paste prompts for common engineering tasks, and helped establish a foundational practice for AI‑assisted coding.

Since then, adoption has grown, but the bar has also risen. In our research, one of the most consistent blockers we see is trust, especially when developers are working in complex, high‑risk systems. They may be comfortable using AI for boilerplate or small refactors, but hesitate to rely on it for critical services, shared libraries, or data migrations where unintended changes are unacceptable.

This new Advanced Prompting Guide for AI Engineering is designed for those environments. Gathered from numerous sources, including interviews, educational talks, and community interaction, the techniques in this guide reflect what has held up across a wide range of real teams and codebases. These solutions are meant to be tool‑ and architecture‑agnostic: they can be applied whether you are working with agents, traditional coding assistants, or spec‑driven workflows on any modern stack.

Here’s how we recommend leaders use this guide:

1. Measure
Capture a baseline view of AI tool usage and impact in your organization using a multi‑dimensional set of metrics, such as those in DX’s AI Measurement Framework. This gives you a starting point for evaluating how enablement efforts and new prompting practices change outcomes over time.

2. Distribute
Start by distributing the original Guide to AI Assisted Engineering to build a foundational practice for AI use within your organization. Then, share this new Advanced guide for higher‑stakes systems and use cases.

3. Iterate
Treat the prompts and workflows in both guides as living templates. Adapt them to your stack and workflows, centralize the patterns that work well, and refine them using what you see in your metrics. Use the data to decide which techniques to standardize, where to expand usage, and where additional guardrails are needed.

As one final note for leaders: the definition of “developer” on your teams is expanding (and that includes you). Engineering leaders, designers, and product managers are increasingly using AI in the work of delivering software, from shaping strategy and drafting specs to reviewing designs and coordinating complex changes. As this work moves beyond simple code generation into managing cascading system rules, cross‑team dependencies, and critical transactions, generic prompts are not enough. The patterns in this guide are intended for your own use as well: to structure how you ask AI to help with planning, design, review, and decision‑making, and to give you a safer, more predictable way to apply AI in the parts of the system you’re accountable for.