The purpose of this post is to better understand abundance. Our scarcity way of thinking and behaving is challenged as generative processes are able to deliver an almost infinite number of high-quality-outputs. This will lead to bottlenecks upstream (defining what we want) and downstream (evaluating and choosing the right output). Abundance challenges us to dive deeper and be more thorough in these areas.

When building a generative system for ideation in innovation, abundance is a crucial concept. Done well, generative processes naturally create abundance by producing large volumes of valuable output.

We should be cautious from the start. If you talk to an innovation or product manager, you’ll often hear that they already have an abundance of ideas. That can be true (although I disagree, see my article here), but this “abundance” is often about quantity rather than quality. Generally speaking, an abundance of low-quality ideas is neither difficult nor inspiring.

Scarcity drives value and is our default mode of thinking

Abundance is interesting because it runs contrary to our usual focus on scarcity. Scarcity drives value. It implies there’s one rare object, one golden nugget, we just need to find. This usually means there are many bad options and only one great option. Because that single best option is presumed to have near-mystical qualities, we assume it will be easy to recognize. The only needle in the haystack, once found, is scarce, valuable, and crucially easily distinguishable from the hay. In this classic example, we don’t need a sophisticated evaluation method to identify the needle as valuable; we already know it is. The challenge lies in the search through all the hay to find the needle in the first place.

Abundance changes the dynamics of evaluation and interaction

With generative processes, and the abundance they create, the dynamics change. Put differently, a stack of needles with no hay is a different situation. We tend to treat good ideas as scarce, and in certain environments (like uninspiring conference table brainstorming) that holds true. Done correctly, AI-based generative ideation processes increase both the quantity and the quality of idea output. Taken to its extreme, where we imagine a scenario where there’s a huge sea of options with good ideas, the needle in the haystack metaphor falls apart. Instead, we must ask: “How do we interact with this abundance?” Or, more precisely: “We have a stack of needles, there’s no hay, the question is which needle do we want to choose?”

Abundance moves the bottleneck downstream

When one bottleneck is solved, pressure and complexity moves upstream and/or downstream. For example, my ability to write many (good) notes creates pressure on my ability to organize them downstream. Distinguishing between needle and hay lets me get away with a simplistic evaluation process because they’re easy to tell apart.

By contrast, generative processes, and the resulting abundance (assuming a broadly high-quality distribution), puts downstream pressure on our ability to index, manage, and differentiate among ideas. This puts us in an interesting spot. It’s actually easier to proceed when we have one good idea and nine bad ones. Internal evaluation becomes trivial. If ease and speed of evaluation were the goal, having only one good idea would be ideal. Of course, ease and speed should not be the goal; selecting the most promising idea should be.

AI-based generation commodifies quality and puts ideas in a competition for differentiation

A scenario with multiple good ideas is different. The “needle in a stack of needles” puts us in a tougher position. The issue isn’t finding one best idea out of ten poor ones; it’s choosing among several high-quality options. We rarely have the capacity to pursue all ten.

So, how do we deal with idea abundance from generative processes? My current thinking centers on three levers.

1. Upstream: Changing the information architecture: This means describing and creating ideas based on a generative code. This generative code follows a structured logic in order to make the parts of the code reusable and more importantly combinable. By using a generative code, which I like to call Generative Sequences after Christopher Alexander, we are better able to deal with abundance as each output has clear connections to other outputs. Generative code is a lightweight way to store, transfer and explain a wide variety of outputs. (I will deep-dive into this topic in a separate post.)
2. Upstream: Define what we want and don’t want. This step clarifies what makes sense for the company and its environment. Consider vacation planning, There’s an abundance of options, so we predefine simple rules to narrow choices (e.g., “by the ocean,” “allows hiking”). This lever prompts deeper analysis of our capabilities and a clearer understanding of the external context. It’s best done upfront with simple rules and principles. Returning to the needle stack. I probably have a good preconceived notion of the needle I’m looking for. This helps me guide my choices and decisions. Given the indexing, I can then query for the right needle.
3. Downstream: Add information to increase differentiation. Treat this as an additional workflow for indexing and distinguishing ideas. Generated ideas—comprising their generative prompt sequence and full-text description—are enhanced with modules that add valuable information. The sum of the idea and its modules is the “idea stack.” Each module automatically adds signal across the idea set—for example, mapping each idea to higher-level patterns and themes, adding logic checks, and rating key assumptions. This helps us index and differentiate a large volume of ideas.

Toward the right interface for abundance

All levers feed into a broader challenge in generative ideation, and in generative AI more generally, which is finding the right interface. Chat-based interfaces are currently dominant and make sense for conversational interaction. When we think about generative processes and the abundance it creates, than there’s a need for a better interface as an intermediate layer to interact with the abundance. My current thinking is that the interface sits on top of the abundant component, e.g. ideas, and helps us formulate what we are looking for in order to better navigate the abundance. Building this interface will be the next challenge as chat is most likely not the right interface.

Learning from other domains

There’s also a broader question. How do other fields handle abundance? Some examples that come directly to mind include VCs selecting startups, companies screening job applicants, or individuals choosing a vacation destination. I will write another article about that, that I will add here.

---