Strategies for Adopting Generative AI: Lessons from Cloud Adoption
While not every new technology trend looks the same, echoes of past transitions can be heard in the latest trends. Living through the early days of wide-scale transitions to cloud-based services like Amazon Web Services (AWS), I’ve started to hear familiar tunes as organizations adopt Generative AI. This led me to reflect on those early days in the move to the cloud and to look for similar themes that might help smooth the transition to an AI-enabled future.
Lesson 1: Wait to solve problems that are hard now
It’s easy to get caught up in the hype of new technologies and attempt to solve hard business problems with the latest technology. But as teams start to work on solving these challenges, they often find product and feature gaps in the new tools. This is natural, especially for products that have been built for individual or small team use, but aren’t mature enough to address the needs of larger or more mature organizations. This is particularly true regarding security, operations, and integration with other services. You may have people with strong technical skills who see a way to build around these limitations or problems. However, these solutions are often complicated, and, as a result, expensive to build and difficult to maintain. What’s worse, given how fast the field evolves, the solutions are often redundant within a matter of months as new features or complementary tools are developed and added by vendors. The solution? Wait. Unless delay is massively expensive–which is rarely true–you are better off scaling back your objectives or looking for other problems to solve until the product landscape evolves and what was once a challenging engineering problem is now just a product configuration option.
Lesson 2: Set ambitious goals, but don’t overplan your path to achieving them
This lesson also emerges from the challenges of working in a space where things are moving rapidly. Best practice would often tell us that you should have a clear path in mind when you start on a project. But the focus you want to maintain should be more on your intended destination, and less on the specific route you will take to get there. Take more of an agile approach and look at what you need to achieve in the short term and then solve for that first leg of the journey. When you reach that first waypoint, you can pause, look at how things have evolved since you started, and then make plans for the next stage that move you towards your eventual goal. Ideally, you want to break things up so that you get some incremental value with each segment. This also mitigates the risks from the first lesson, as it’s possible you will find yourself at a point where it is difficult to move on and it is best to wait. These shorter planning cycles also make it easier to backtrack if it becomes necessary without having gone too far astray.
Lesson 3: Make it easy for people to access, learn, and share experiences
When I helped with an organizational move to the cloud I organized what became known as the “Big Group.” We invited all of the staff in the technology organizations to regular meetings where we discussed what people were learning, the challenges they were experiencing, and organizational support they needed to help move us forward. We had also set up a monthly, one-day introductory training session led by our cloud vendor and created opportunities for people to join short-term projects focused on small projects. Together, all of these efforts built enthusiasm and capability at the front-line, and let us learn how people were actually working with the tools and what leadership support was required. Early indications are that this approach is extremely beneficial in the Generative AI era. The difference, though, is that instead of just technology teams, these efforts should include almost all staff in an organization given the general purpose nature of the tools.
Lesson 4: Reward failure
When new technologies emerge, the “right” way to solve a problem often isn’t known. Best practices have yet to emerge, and what might have been the smart way to tackle a problem yesterday could be an outdated approach by tomorrow. The frontier of capabilities is ill-defined under rapidly evolving conditions, and given the unusual workings of Generative AI this is even more true at the moment. This makes it difficult to rely on traditional measures of success–where you achieve your expected outcome–and instead need to reframe the objective as one where you ask if you have learned something as the result of your work. To ensure that learning happens in a rapidly changing environment, foster an environment where people are comfortable failing. The easiest way to do that is to reward failure and normalize it as a key component of the learning process. Perhaps you institute a “failure of the month” award, or go so far as to create a “failure budget” that is only spent when you try something that is unsuccessful. And if you aren’t spending down your failure budget, you’re not pushing your organization enough.
Lesson 5: Failure today doesn’t always mean failure tomorrow
Even among organizations with a focus on rewarding failure, there is often the mindset that to fail once is smart learning, but failing twice is not learning from mistakes. While in circumstances where the environment doesn’t change rapidly this is true, in dynamic conditions you need to keep revisiting your approaches to see if things have changed. This is really the flip side of lesson 1, because what was hard, and may have failed initially, could now be solved with new tools or systems. Early generative AI solutions were particularly bad at solving particular types of problems–math, counting, reasoning and logic–but advances in the technology have led to rapid gains in these areas. Where early systems were highly prone to hallucinations–and so wouldn’t be appropriate for customer-facing solutions–new models combined with new techniques now make the tools suitable for some of these jobs. As a result, it’s worth revisiting prior failures and trying again if there are new things that might address the prior limitations.
A Final Lesson: Just Get Moving!
In times of significant change and uncertainty, a common reaction can be to hunker down and wait, hoping things will settle and become clearer in the future. It’s an easy choice to make, as embracing the unknown can be unsettling, for both individuals and organizations. There are also many instances where the latest thing is just a fad, and after they hype wears off everyone moves on with little change. As I’ve argued before, I do not think this is one of those moments, and waiting will only make you less prepared for the future. Over time it will become harder to compete–harder to keep and recruit employees, harder to meet the needs of your customers, harder to secure a place in your industry. This is a lesson I see playing out now in organizations who didn’t embrace cloud migration earlier. They are doubly challenged because they don’t have the foundational cloud skills and systems to help with the adoption of AI and need to undertake a cloud migration journey while also considering how to adapt to AI. This doesn’t mean you need to go “all in” on AI, but neither should you wait around for the perfect time when the path is clear. By that time, everyone else may be so far ahead of you that your journey is effectively over before it’s even begun. So get started, try a few things, and open yourself up to the excitement and rewards of exploration.
Content Note: This article (em-dashes and all!) was written entirely by the Author, with light copy editing provided by Google Gemini.
