How to find your next AI startup idea when everything seems possible
As an early-stage investor, friends often ask me something like, "Hey, I have this AI startup idea. Do you think it's a good one?"
The ongoing AI revolution offers incredible possibilities, but also brings uncertainty. The stakes are high: choose well and shape the future; choose poorly and fall behind. With limitless possibilities, how do you decide what to build?
A Simple Framework: the Critical Frequency Matrix
To cut through complexity, here’s a simple mental model I’m calling the Critical Frequency Matrix. In physics, critical frequency often refers to a threshold at which a medium—like the ionosphere or a waveguide—changes its behavior by reflecting or blocking waves instead of letting them pass through. Similarly, I used the Critical Frequency Matrix to identify pivotal areas within an enterprise where startups can break through, drive transformation and capture new opportunities.
To start, think of any enterprise as a collection of tasks or “jobs to be done”. We can then visualize the profile of the business as the distribution of these tasks along two core dimensions: frequency and criticality.
Frequency is straightforward, tasks can happen constantly (e.g. processing orders) or very rarely (e.g. M&A). Criticality is fuzzier—what’s vital for one business might be minor for another—but generally, criticality reflects the potential business consequences if a task goes wrong or is not executed to a high level of quality.
Every enterprise task can be mapped onto this grid, creating a simplified fingerprint of the business. While you might consider adding other dimensions like “technical complexity,” doing so actually weakens this framework's effectiveness. The power of this approach lies in what it deliberately excludes.
This intentional simplification recognizes a core truth in our world of accelerating computing power: technical moats rarely endure. At best, technical complexity provides a temporary head start measured in months, not years. Yesterday's PhD-level engineering challenge becomes tomorrow's API call. The enterprises that win aren't those with the most complex technology, but those that apply the right technology to the most valuable problems.
Finding AI’s Sweet Spot
Now that we have a framework, let’s return to the key question: Where exactly on this matrix can you find the most valuable AI opportunities?
While every business has its unique dynamics (i.e. different dot profile), certain task categories will have similar commercial implications.
Category 1: Humans excel at low-frequency, high-criticality tasks
Infrequent but pivotal events, think major corporate pivots, crisis management, or annual strategic planning, are where AI automation is least valuable and where human leadership shines brightest. These tasks demand nuanced judgment, creativity, and rapid adaptability. While AI can offer data-driven insights or scenario modeling, it struggles with the intangible, context-rich decision-making that defines success in unpredictable moments.
For example, deciding whether to acquire a competitor or fundamentally realign product strategy requires balancing quantitative metrics with cultural considerations, ethical implications, and an opinionated vision of future markets—all areas where a human perspective is essential. Similarly, when a PR crisis erupts, the ability to negotiate, empathize, and communicate transparently cannot be reduced to lines of code. Because such events occur too rarely to justify extensive automation, and because they hinge so heavily on human feelings, AI will likely remain a supporting player here for the foreseeable future.
Category 2: High-frequency, high-criticality tasks can be lucrative, but challenging for AI agents
Tasks that occur frequently and hold high criticality offer immense market opportunities but also present significant challenges for agentic solutions. The key question is whether agentic solutions, with their non-deterministic reasoning, are better than Robotic Process Automations (RPA) for these tasks. Generally, automation in these areas must achieve near-perfect reliability and accuracy, making technical execution paramount. These attributes make this region a great place for simple, deterministic automation. AI agents can still contribute meaningfully here, but their agency, their ability to reason and independently think, must be constrained. For example, an agent may still determine which predefined workflow to execute based on a given context; however, once a workflow is selected, the individual steps are likely to remain rigidly structured to guarantee predictability and minimize risk.
Category 3: Low-criticality, high-frequency tasks offer the broadest scalability
Tasks that have high frequency yet individually hold low criticality—such as routine customer support inquiries or basic bookkeeping—are ideally suited for automation through agentic systems. Their predictability and widespread presence across multiple industries create the opportunity for broad adoption, unlocking significant value previously trapped in these everyday processes.
For founders, there's an additional layer of strategic potential: Within enterprises, there are clusters of related tasks which require the same set of underlying capabilities. Identifying one of these task clusters that extends across multiple industry verticals is exactly the kind of high-leverage, transformative opportunity that investors dream of.
Category 4: Low-frequency, low-criticality tasks offer untapped value through generalized AI assistance
My intuition, though not yet fully validated, is that low-frequency, low-criticality tasks are typically fragmented, specialized, and disconnected. Said more plainly, I think the tasks in this category are relatively unrelated to one another. For example, executive assistants must execute a variety of tasks completely unrelated to one another: organizing occasional events, calendar coordination, even IT troubleshooting. Individually, these tasks do not justify dedicated automation and a solution that automates all of these tasks is challenging to develop due to their disparate nature.
Of course, a generalized agentic system capable of handling this diverse array of infrequent tasks would be immensely valuable–something close to AGI. Such a general system would require exceptional adaptability and intelligence to manage these unique, context-dependent activities that resist standardization. In short, building a big business here is possible but only with something resembling artificial general intelligence (AGI).
Conjecture: Market value for individual tasks generally grows with frequency and criticality
The market value of automating a job rises as you move “up and to the right” on our matrix—meaning tasks become more valuable when they are both more frequent (higher frequency) and more essential (higher criticality) to the business. Automating real-time fraud detection in banking, continuously monitoring patient vitals in hospitals, or running non-stop quality control checks in factories can yield significant returns precisely because these tasks are both critical and frequent. Conversely, occasional tasks with low criticality—like ordering office supplies quarterly or preparing annual internal memos—offer minimal benefit when automated.
Moreover, the market value of automating a task exhibits higher elasticity with respect to frequency than to business criticality, which is why the red curve is convex. Said more simply, my guess is that frequency is a greater contributor to market value than criticality. Why? Because the business value of solving a problem that causes constant pain is higher than solving episodic pain. Just look at the market capitalization of Uber versus automobile retailers.
Using the Framework to Spot Hidden Opportunities
As an investor, this framework helps me to focus on enterprise value and to highlight task areas with genuine potential for new AI-native ventures. Here’s a practical Charlie Munger-inspired checklist, to help apply this framework when thinking of your next AI-native startup.
Common AI-Native Startup Patterns
While evaluating individual tasks is important, building an AI-native business typically requires addressing a group of interconnected tasks. The sequence in which you tackle these tasks will significantly shape your business and go-to-market strategies. Here are a few common patterns in building an AI-native startup for the enterprise.
Remember, this is a tool, not a formula
The Critical Frequency Matrix isn't a magic wand—it's a tool. Like a hammer that drives nails effectively but can't cut wood, this framework is helpful within its intended use (AI-native, enterprise applications), but is limited outside of it. Apply it thoughtfully, not blindly.
To founders in this space: You're building during an extraordinary moment. If you're tackling these challenges, I'd love to hear about your journey. The world needs your ideas—now go build something amazing!