A.I. could cut time spent coding by 45%. But even seasoned tech professionals will need ‘extensive’ training to harness its full power, McKinsey experts say

Business and HR leaders looking to master these superpowers will need to train their people to work with A.I., hire the talent to build those A.I. tools, and keep that talent so they can grow your A.I. capabilities. For companies that can pull this off, the rewards in terms of productivity and new value will be significant.

Here’s the first reality check: Almost every job will have an A.I. tool that will allow an employee to do their job better, but that requires extensive training. Effective training starts with understanding how A.I. is changing the work people do. Our own experiments have shown that using generative A.I. tools can help reduce time spent refactoring code by 20% to 30% and generating code by 35% to 45%, but speed gains vary by task complexity and the developer’s experience. These tools perform best for relatively repetitive tasks and in providing a starting set of code that developers can work with and improve.

As the tasks get more sophisticated, however, A.I. tools become more like co-programmers helping developers create software. To work with these A.I. tools, developers will need new skills, especially when it comes to generative A.I. These new skills will include how to better understand the intent of end users, how to translate that intent into code and test the results with subject matter experts, and how to closely track and rapidly adjust models based on performance. They will need to evaluate the solutions A.I. suggests, and understand which A.I. tools are best for which tasks (and be able to combine various ones to unlock greater capabilities).

The power of A.I. to change how people work extends far beyond developers. Let’s take a retail merchant, for example. A merchant’s goal is primarily to build strategies that drive growth and expand margins in their category. A.I.-powered analytics can transform many of their underlying tasks—strategic and tactical pricing, SKU selection and assortment, size and timing of buys, inventory planning and management, promotion mix and timing, markdowns, et cetera. Generative A.I. will be able to do even more. Imagine if generative A.I. could help a merchant design different visual merchandising plans for their section of the store aisle, and then rank them by various criteria (cost, feasibility, probability of sale, et cetera). Or imagine if a negotiation tool could help a merchant select a negotiating position and supporting tactics (e.g., acceptable concessions; pricing guidelines) tailored to each key vendor.

In each case, this merchant is using superpowers provided by an A.I. tool to unlock insights from data sets they’ve always had but have not fully exploited. And in each case they’ll need to learn new skills. For the merchant, that would include prompt engineering to get the A.I. model to build the best category plan, image enhancement to develop initial options offered for the planogram of her shelf space, and A.I./human collaboration as she uses the bot as a negotiation coach in her vendor meetings. This power then frees her up to step back from their day-to-day work and do other work like considering the future of her category, how shopper needs and segments are shifting, and how the competition in her category is behaving. This, too, will require upskilling. 

Amazon recognized the importance of A.I. training when it upskilled its merchants so they could incorporate new information (e.g., data from new sources, use cases, and models) and vet prescriptive recommendations. Merchants learned to work with engineers to fine-tune the A.I.-based models that Amazon developed to do pricing, SKU selection, and promotions, for example. Because Amazon automates much of the work, the job of the merchant became one of guiding the development of these automations, while freeing up their time for other tasks, such as recruiting or negotiating with suppliers. Taken to its logical conclusion, we can see how some jobs will shift from managing the execution of A.I. recommendations to designing automated A.I.-based solutions that do the execution as well.  

Providing these A.I. tools to your employees requires finding talent who can build and maintain them. Our research shows that the most in-demand A.I. talent includes software engineers, data engineers, A.I. data scientists, and machine learning engineers. Finding these people is a big challenge. Respondents to a recent McKinsey report highlight that hiring for A.I. talent has been difficult. In China, demand for people who can build A.I. products, for example, will grow sixfold by 2030, while local and overseas universities as well as existing top-tier A.I. practitioners are estimated to supply just one-third of that need.

As we argue in our book, Rewired, finding good talent starts with understanding what skills you need. In too many cases, we have found that companies hire A.I. engineers only to have them not have work to do that matches their skills. Understanding the A.I. solutions you need to deliver on your business’ goals and what skills are needed to develop those solutions is the granular work that companies need to do to identify the right talent. With this clarity in place, companies should incorporate A.I. experts into the interview process to reassure potential talent that they’ll be working with top people. In addition, HR leaders will need to develop a recruiting program to optimize the candidate experience, not just improve processes.   

Importantly, CIOs and technology leaders will need to also ensure complementary talent is in place to enable A.I. talent and tools to be effective. For example, data engineers are needed to develop a data architecture that ensures A.I. tools access to quality data. Machine learning ops engineers are needed to refactor and manage A.I. solutions over time so they can be trained and “corrected” over time. Companies seeing the highest returns from A.I,, in fact, have hired more data scientists and machine learning engineers than any other digital role.

Hiring top A.I. talent isn’t enough. As stipulated in Rewired, the key to keeping top talent lies in creating an environment where A.I. talent thrives. Some of the lessons more broadly applied to tech talent are relevant when it comes to A.I. talent. One area where that’s particularly true is around career paths. While some digital colleagues want to progress into general management roles, more than two-thirds of developers don’t want to become managers. They prefer instead to keep their craft sharp and continue to build their skills. We cannot stress enough that skills are a top currency for A.I. talent, and being able to build skills is an important motivator for them. Companies seeing the highest returns from A.I. are nearly three times more likely than other respondents to say their organizations have capability-building programs to develop technology personnel’s A.I. skills.

For this reason, it’s crucial to invest in creating long-term learning journeys that support technology employees in developing the breadth and depth of their craft, as well as the behavioral skills that the organization also values. When designing learning journeys, it is important to differentiate between skill families. Resist the temptation to view all technical roles as interchangeable (“they are all engineers”).

The A.I. superpower is real and potentially transformative. But its full potential value will remain in check until companies commit to training their people how to use A.I. tools and finding the A.I. talent to create and manage them.

Eric Lamarre, Alex Singla, and Suman Thareja are senior partners at McKinsey and the authors of Rewired: The McKinsey guide to outcompeting in the age of digital and AI.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.