A Roadmap for AI Products in an Age of Hyper Darwinism

In the midst of a generational technological shift, flavors of AI have unsurprisingly permeated every boardroom agenda and product roadmap globally. Simultaneously, the sands of technology continue to shift beneath our feet -- this AI era is one of “hyper-Darwinism”, where every quarter, new AI capabilities seem to negate an entire swath of burgeoning products.  

The question then becomes: In an AI world where change is the only constant, how can one build for the long run?  

Is this Platform Shift Enduring? 

Examination of prior platform shifts across internet, mobile and cloud illustrate the futility of denial. Quips such as “the internet is a fad”, “the iPhone is just a toy”, and “no one will compute off-premises” have aged poorly. In the time of generative AI, the story is similar. Acknowledging and prioritizing the realities afoot will likely appear prescient in time, especially when we consider that AI has been an ambient force in our lives for longer than most realize. Consider the following:  

Related:The 3 Things You Need to Modernize Ahead of AI Investment

This AI revolution has been brewing for decades, and hindsight reveals that time to adopt all new technologies is compressing, engendering a renewed sense of urgency for leaders.  

How to Endure Products during a Tectonic Platform Shift? 

Appreciating the likely permanence of AI in our lives, how should teams approach product development? We propose core principles for building AI products with durability in mind:   

Define the right tools for the right job.  AI is the acronym du jour, but it should not distract from the core focus of any product team, customer delight. Whether designing toasters or writing physics simulation software, customer centricity reigns supreme. Knowing which problems to solve and which technologies most appropriately solve those problems are the two most important concepts any product team can grok -- AI or not. However, in this environment, they assume differential complexity. 

Related:Go Big or Go Home: The CDO/CDAO’s Guide to AI Leadership

What problems are we solving for our customers? Soul-searching is critical in the midst of platform shifts. In the AI era, this takes the form of asking: Would my product exist were it not for the underlying technology? One can investigate by asking: Does my product augment existing capabilities, or does it create new capabilities altogether? 

This answer should directly inform product roadmaps.  Where AI augments existing capabilities, one can append an existing roadmap.  If it unlocks new capabilities, a first-principles product focus is required. We characterize these strategies as “plus AI, or +AI” and “AI plus, or AI+”.  Your +AI product strategy can be prosecuted "classically", while AI+ roadmaps require fundamental redefinition. 

How should we solve those problems for our customers? While generative AI -- in the form of transformer and diffusion-based models -- has driven the most recent wave of excitement, AI has a 70+ year history. The current shiniest objects are unlikely panaceas for all customer needs. It is then critical for product leaders to possess, develop or hire expertise required to invoke the most appropriate AI tools for their roadmap. A one-size-fits-all approach is likely insufficient because thousands of machine learning architectures exist, and each possesses cost and performance tradeoffs. 

Related:Does Your Organization Need a Dedicated AI Leader?

Seeing the Forest Beyond the Trees - Resist Feature Frenetics 

The arc of a product’s life cycle is long and resisting distractions along the way is paramount to success.  

Just because AI is easier than ever to integrate does not necessarily or universally mean that it should be. Whether building for a +AI or an AI+ endgame, driving principal understanding of products vs. features is more critical than ever. 

Injection of buzzy features will not make an AI product enduringly alluring. While many famous products certainly began as features, the product gestalt requires conscious effort to transcend merely chained features. The difference between an ephemeral product and an enduring product is defined by the focus of product teams. Are they focused on fleeting needs or considering the holistic customer journey? 

Meta has nearly two decades of product iteration under its belt and continues to grow its user base.  It developed a product first, then technology to support it, evolving with the fluid needs of its customers. 

Conversely, OpenAI developed technology first, with products to showcase it. ChatGPT was the fastest product in history to reach 100 million users, but consumer interest has meandered. This certainly may change, though it illustrates that technology on its own is insufficient. 

Build for 1x, Engineer for 10x, Architect for 100x 

Maximize malleability and modularity. Half the battle is self-awareness, The import of understanding the purpose and essence of an AI product cannot be overstated, but it’s time to get tactical.  

Building in a space where change is the only constant is a tall task. Several maxims can help reduce "technical paralysis" teams might encounter in this new era: 

Build for 1x: Start by building for today’s needs to see what works, instead of letting perfect be the enemy of good, prioritize rapid learning. 

If building for a narrow domain that can be addressed via a 7B parameter model, avoid investing in a 70B parameter. Get the minimally viable product (MVP) into the wild (responsibly) and begin gathering feedback. 

Experiment and fail fast and avoid getting hung up on sunk costs. "Mistakes", so long as they do not drain resources, should be celebrated for the clarity they bring to the next iteration. 

Engineer for 10x: As a product extends beyond today’s needs, build with flexibility in mind. Some example prompts: 

Does every model introduced to the product architecture require a rewrite? 

Can a product’s back end adeptly toggle between the latest state-of-the art models? 

Architect for 100x: As a product expands and teams consider a 5-to-10-year arc, build with modularity and malleability in mind. 

Avoid having all eggs in one algorithmic basket. AI architectures have ebbed and flowed for decades 

Build buffers and capacity that allow for ingestion of new datasets, new model types, and multiple models at a time. 

AI presents an exciting yet daunting environment in which to be a product leader. Self-awareness is key to ensuring that AI is accretive to customers. A focus on those customers requires knowledge of their needs today and a view of how those needs will change over time. In all cases, meeting them where they are with the right tools for the job is paramount. Throughout all of this, taking a "Lego block" approach to systems architectures will afford product teams the opportunity to cross the chasm from ephemeral to enduring.