In 2024, Chris Paik memorably opined on a future where creating software will be as effortless as posting on Instagram. Where software becomes “disposable”; spun up to solve a specific task and discarded once complete.
Less than two years later, that prediction feels eerily prescient. AI coding tools have supercharged developer productivity while making software creation accessible to non-technical builders in a very real way. The recent bloodbath in public SaaS markets reflects a mainstream sentiment shift: software is easier to build and harder to defend. Founders and investors are now reckoning with multiple uncomfortable futures: price commoditization as lower barriers to entry flood the market with new entrants, the growing viability of the “build it in-house” competitive alternative, and the possibility that AI agents eliminate the human workflows many software companies exist to serve.
These shifts are real, but the “fog” created by AI’s rapid pace makes it difficult to evaluate the durability of any individual software company. As software investors, we are not interested in predictions. What we are most interested in is: what are the traits of the software companies that will still matter decades from now, no matter how good AI and development tools become?
Our thesis is that enduring software solutions will emerge from problem spaces that are complex, messy, and deeply tied to the real (physical) world, shaped by founders with hard-earned insights and a uniquely informed perspective on the problem. These conditions produce what we call “Thousand Paper Cut” moats; the more difficult it is to 'short-circuit' their development, the more defensible the advantage becomes, and the harder it is to displace.
Software built on thousand paper cut moats starts to resemble hardware: painful to build, but with structural advantages that compound over time.
These enduring software companies share three traits:
They solve a problem rooted in the physical world
That problem is extremely complex due to the chaos of its environment or the precision required to solve it
Solving it requires building a “Thousand Paper Cut” moat — value derived from the accumulation of many innovations that are difficult to replicate
Physical in this case does not mean hardware or brick-and-mortar businesses. It refers to software companies tackling problems that originate in the real world. The clothes we wear, the food we grow, the healthcare that treats us, the buildings we live in and so forth. These are problems rooted in our three-dimensional existence, not just digital workflows.
These domains are riddled with inefficiencies and are ripe for transformation through software and AI, but the underlying human needs aren’t going away.
In contrast, problems that exist purely in the knowledge-work space — digital workflows, internal collaboration, task management — are broadly vulnerable. When agents reshape or eliminate the underlying tasks entirely (hello Claude “skills”), what’s left? Software that only makes sense with a human in the loop is in trouble when the human leaves the loop. There’s nuance here. Critical infrastructure like security, payroll, payments, and deep systems of record isn’t going anywhere (no rational business is going to vibecode its payroll system to save on SaaS spend). But, the ground is shifting fast.
For founders building new companies, the knowledge-work landscape is a minefield. The line between “enduring infrastructure” and “soon-to-be-automated workflow” is blurry, and competition from both incumbents and AI-native upstarts is massive. Physical-world problems are a different bet: they’re more greenfield, deeply rooted, and the opportunity to build something durable is much broader.
Portfolio companies AssistIQ and Max Retail neatly fit these criteria. AssistIQ solves for the high cost of wasted and misused medical supplies during surgical procedures at hospitals. Max Retail helps independent retailers find buyers for their excess inventory through large online demand channels. The source of the underlying problems (surgeries, the need for apparel and retailers buying more inventory than they can move) will persist for the foreseeable future irrespective of AI progress.
Difficult problems push teams to build more robust solutions, and these problems resist shortcuts. The opacity, the information asymmetries, the sheer messiness — you can’t prompt your way through them. Founders have to wade in, eat glass, and iterate their way to understanding what to actually build. That process, and the hard-won insight it produces, creates durable advantage. We think about this complexity in two categories: complex environments and precision complexity.
“Complex Environment” problems are difficult because they exist within messy systems that involve many different stakeholders, steps in a process, or levels in a value chain that need integration and coordination for a product to deliver a step-change improvement to the status quo. They’re characterized by error-prone manual processes, rent-seeking middlemen, fragmented data sources, and decision-making driven by relationships and insider knowledge rather than accessible data.
Portfolio company Mercator AI solves a complex environment problem. General Contractors have no visibility into new developments in their regional market. To identify opportunities they rely on word of mouth, opaque insider networks, bid platforms, pulling permits from city hall. For GCs working in multiple regions, this problem compounds across every city they operate in. Mercator organizes this chaos and connects dozens of disconnected data sources to create real-time visibility into emerging projects, an extremely challenging technical undertaking.
“Precision complexity” entails technically intricate challenges where success is binary. A working solution must account for numerous variables, accommodate edge cases, and demand deep understanding of the domain. Unlike “complex environments,” the difficulty here doesn’t come from the system, it comes from the nature of the problem itself. Accuracy matters. It either works, or it doesn’t.
Portfolio company Vivid Machines reckons with precision complexity in predicting the health and yield of fruit trees. Doing so means dealing with massive biological variability, environmental noise, and unpredictable edge cases, in real time. To be useful for farmers the data and insights have to be extremely accurate.
Metafold is in the same boat. Their platform helps manufacturers instantly generate detailed volumetric data of any 3D shape, data that is completely missing in traditional representations of geometry. This enables decisions on manufacturability, performance, and cost. If the output isn’t extremely precise, the insights simply aren’t useful to the customer.
This complexity brings a barrage of unexpected technical hurdles to overcome. Products that successfully solve these problems don’t hinge on a single technical breakthrough. They come together by addressing dozens of small, messy challenges with a blend of straightforward fixes and novel solutions.
These “thousand paper cut” solutions stack up together to create a product that just works. As more “cuts” are resolved over time, the effectiveness and depth of these products compound and become harder to replicate.
Max Retail’s platform allows inventory sitting on a retailer’s shelf to be listed simultaneously across multiple e-commerce platforms. This sounds easy, but there are numerous innovations that make it possible. Max Retail integrates with retailers’ inventory systems for automatic product listing, dynamically generates and adjusts product images to meet the unique specs of each demand partner, and applies pricing intelligence to maximize ROI across channels. These many paper cuts stack together to deliver an experience that’s incredibly difficult to replicate from scratch and becomes more effective every day.
Critically, the AI opportunity increases the value of thousand paper cut products. The grimy, unglamorous innovation these tools are built on makes them uniquely capable of applying AI to solve problems that only they can. This protects them from a collision course with foundation model companies and dramatically expands the revenue they can capture in their markets.
By figuring out how to source, architect, and collate the most comprehensive city-by-city dataset of early signals of construction projects, Mercator can launch an AI sales agent for GCs. This is a product that only they can build, and allows them to deliver (and capture) more value to their customers than they could in a pre-AI era.
It’s extremely challenging to build software that fits the traits above: operating in the physical world, solving inherently complex problems that demand thousand paper cut solutions. It’s painful. But for founders who succeed, there’s a halo benefit that we like to call the “running the maze” effect.
In this context the initial product founders set out to build is rarely the one that ultimately works. The right path just isn’t visible at the start. Instead it’s uncovered through a long period of iteration, frustration, and tedious, hard-won progress.
Running the maze reveals unexpected opportunities. The persistence it demands leads to an evolution of the original idea into something larger: a pivot, a deeper insight into the problem space, or the realization that a product’s impact is far greater than first imagined.
It also means no shortcuts for competitors. The layered nature of Thousand Paper Cut solutions makes it very hard for new entrants to foresee everything they’ll need to build in advance. Competitors have to run the maze from the beginning, making replication difficult no matter how well-resourced the challenger.
—
The fog isn’t lifting any time soon. In a world racing to build software faster, the most valuable companies will be the ones that were hardest to build: solving messy, complex, physical-world problems with thousand paper cut solutions. If you’re a founder building something that fits this framework we’d love to hear from you.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.