How to assemble a generative AI dream team

The question of the day is no longer “What’s the best cloud?” It’s now “How do we build a team of people to build a net-new generative AI system?”

Budgets have dropped. The board of directors wants one of these nifty generative AI systems that everyone is talking about to fundamentally change how the business functions. The expectation, fueled by many business articles, is that this new system will redefine the business. They want one badly, and you’re tasked with building and leading a team that can pull it off. So, how the hell do you do that?

Not a new problem

Ten years ago, we faced a similar problem, stemming from the interest in cloud computing and the need to gather a team to get you migrated to this beautiful world of cloud. Those who understood what was good for their career wanted to get certified with specific cloud brands, perhaps took one of my cloud courses, or did what many people did: learned enough online to get through an interview with someone who knew less.

Although there have been successes, for the most part, the cloud architecture, design, and deployment teams, whether working on migrations or net-new systems, achieved a grade of D. Many of the migrations did not produce value because they hadn’t re-engineered the applications moving to the cloud.

Also, some of the architectural decisions could have been better. Many picked technology for the wrong reasons and are now locked into that technology until there’s money to move somewhere else. We know now that digital transformation using cloud-based resources does not provide the value we thought it would. Generative AI is just a continuation of that story.

A who’s who of team members

There are a few things to remember and some lessons learned from cloud failures and successes. Assembling the right talent, whether for cloud computing or generative AI, will be no different.

First, let’s create a prototypical generative AI architecture and development team, understanding that every team will reflect the needs of the specific business. I’ll make general assumptions here. Also, the team’s makeup will differ by industry. Generative AI development for financial services will vary slightly from healthcare. Yes, the “It depends” answer that people hate.

Here are the roles we’ll likely see on this team, either as new hires, internally trained, or from an outside consulting firm. Let’s assume a single generative AI development project that will likely last six months to a year.

The project manager oversees but does not lead the project, ensuring that it aligns with business goals and is delivered on time and within budget. This person is invaluable in keeping you out of budgetary trouble and managing very complex delivery. For example, the platform of the generative AI system needs to be selected before you deploy. A good project manager will help you avoid making dumb mistakes.

AI researchers and data scientists are tasked with developing “cutting-edge models” tailored to specific business needs. They also continuously improve the models’ performance. These people are the brains of the project, thinking up ways to build and deploy models with the greatest likelihood of success.

Many organizations are demanding PhDs for these roles. That’s a little shortsighted. I work with many talented people who don’t have advanced degrees. You want people who are effective because they work well as a team player. That’s just an anecdotal judgment on my part.

AI engineers are the professionals who bring the models out of the research phase and into production, focusing on scalability and maintainability. They need a deep understanding of AI frameworks (including generative AI frameworks) and the ecosystems that support them.

The challenge here is to hire someone trained in a variety of frameworks. Other people will be limited to solutions geared to what they know. That’s going to be wrong most of the time.

Data engineers are responsible for designing the pipelines and feeding high-quality data into the models. At the end of the day, generative AI systems are just data-oriented systems; thus, this person is essential.

Again, don’t hire people focused on a single database or ecosystem (e.g., only AWS-native databases) for the same reasons as the AI engineers. A one-trick pony will miss the ultimate best-of-breed solution.

Platform engineers choose and design the platform. They understand cloud and non-cloud platform options, as well as memory, storage, and processors. It’s good to have engineers who understand the differences between CPUs, TPUs, and GPUs and who can create a solution that drives the most value.

It’s tough to find these people, as you’ve probably discovered. Some may show up and proclaim that they only use this cloud provider or that server cluster. Yeah, no. Again, your architecture does not care about your biases.

The AI ethics specialist role ensures that the AI system adheres to ethical standards and fair practices while proactively addressing bias in data sets and models. It’s not if you’ll get sued; it’s when. These people can put the rigor in place to ensure you’re not doing nasty things with your new generative AI toy.

AI product managers/owners translate business requirements into technical specifications and ensure product development meets the enterprise’s goals.

Security and compliance officers address potential security vulnerabilities in AI systems and ensure the company complies with all relevant data protection laws and industry regulations.

User experience (UX) designers ensure the output of the genAI system is intelligible and helpful to users and stakeholders.

Integration specialists are developers or engineers who integrate the genAI system into existing IT infrastructure and workflows.

Other roles include support and maintenance staff, business analysts, devops engineers, and legal advisors. This is similar to more traditional development, but they are still needed on these genAI projects.

Oh yeah, the generative AI architect. This person leads the project and the team and makes the tough calls to lead to success.

Of course, the team your company needs to be successful is likely a bit different than this one. However, I think I hit most of the major roles. Now, you only need to find the talent. That’s a rant for another post.