As an AI development studio that has helped numerous companies implement impactful AI solutions, we often get asked – what skills are needed to build a great AI product? We’ve written this article to help you get a handle on what exactly it takes to build that AI product you want for your business.

1. ML engineers —> AI engineers

With large language models (LLMs) like GPT-3.5, GPT-4 and Claude 2.1 readily available, you can now more easily build AI solutions without having to hire machine learning engineers with deep (and expensive) expertise in model training. In fact, for most use cases, you don’t need them at all. What you do need is a new breed of AI engineers – people who can navigate the new AI landscape of pre-trained models and tools to deliver custom solutions to you that exactly fit your use case.

2. Expert prompt engineering still matters

Prompt engineering is one of the most crucial skills in AI development. Ok, we know what you’re thinking – there is so much hype about prompts right now! That’s true but there is at least a kernel of truth to the hype… crafting prompts that accurately and thoroughly communicate the intended model capabilities and behaviors is essential. Getting prompts right is critical for building useful, safe, and legally compliant AI systems. Now a lot of that involves best practices, but a lot is also experiential – we’ve been writing and testing what works practically since when GPT-3 first hit the scene in early 2023. One of the best resources on this comes directly from the horse’s mouth – check out this guide from OpenAI on the six strategies to get better results from your prompts.

3. Interweaving your subject matter expertise with the right prompts to maximize model performance

Remember the old mantra, you need to know something to be able to ask the right questions? That’s entirely relevant for effective prompts – you need to know something about what you’re asking to get the best answer out of the language model. Tight collaboration between engineers and your subject-matter experts is a key building block to ensure that model capabilities closely match end-user needs. And that ultimately comes back to good process… actively shaping prompts, evaluating model performance via rigorous testing and providing continuous feedback to the relevant stakeholders to incrementally improve the AI for your use case.

4. Full-stack engineering as the orchestration layer for AI components

While prompt engineering determines the breadth of model capabilities, full-stack development to handle integrating prompts, orchestrating model APIs and managing infrastructure is essential for efficient and cost-effective model serving. In many cases, optimization via retrieval-augmented generation (RAG) works well to maximize model performance; in other cases, fine-tuning may be the best route – we’ll explore this in depth in our next blog post.

We’ve got you covered on this piece of the puzzle – we excel in delivering the most cost-effective alternative for your use case.

5. Managing all the tools and workflows

Efficient workflows and tools help you connect everything we’ve talked about so far together. Leveraging the right tools and workflows helps click the lego pieces of prompts, subject matter expertise and fullstack orchestration together. The result of seamlessly iterating on prompts, bridging communication gaps across roles and enabling rapid experimentation is a fantastic AI product custom built for you.

6. Monitoring the pulse of the AI landscape

If there’s one thing we’ve learned is a constant in the AI landscape, it’s that things change pretty fast. GitHub and tweets are among the nontraditional resources you need to be savvy about to understand what’s really happening across this fast-moving space, not to mention news, YouTube vids, podcasts, blogs, tool documentation and academic papers. Don’t worry, we’ve got that covered too. We’ve been staying on top of all of this in our AI newsletter, The Scroll, since the beginning of this tech super cycle.

Check us out for your AI development needs.

Our AI studio RareInsight has integrated expertise across prompt engineering, full-stack development, infrastructure, tools and the broader AI landscape to make us uniquely positioned to help companies create capable AI products. We understand the skills needed and have proven experience executing ambitious AI projects. Get in touch here to learn more!