A profound shift is underway in how software gets made, for whom, and at what scale. For most of computing history, building software required deep technical expertise, and the economics demanded large addressable markets to justify the cost. That constraint is dissolving. AI-powered coding tools are collapsing the time and skill needed to build from weeks and years to hours and minutes — and with that collapse comes a dramatic transformation: ==the minimum viable market size for software is approaching one.==
This is the central thesis emerging from multiple thinkers, researchers, and investors simultaneously: the world was starved of niche software not because the demand didn't exist, but because the supply chain — skilled engineers, long dev cycles, expensive infrastructure — made it economically irrational to build for small audiences. That supply chain is being disintermediated by AI. What follows is not just a new generation of startups but an entirely new category of software: personal software, niche software, software for communities of ten, software built in an afternoon for exactly one person's workflow. ==The "long tail" of software — which Chris Anderson articulated for commerce in 2004 — is now exploding into what thinkers like Scott Brinker call the hypertail.==[^1][^2]
To understand why this is such a deep philosophical shift and not just a productivity upgrade, you need to trace the idea through its intellectual lineage — from the dreams of early computing visionaries, through the economics of the internet, to the current collision of vibe coding, agentic AI, and platforms like Wabi.ai.
In 2004, Chris Anderson, then editor of Wired magazine, published an essay that would reshape how people thought about markets. He called it =="The Long Tail". His observation was simple but profound: the distribution of demand for almost any product — books, music, films — looks like a power-law curve. A small number of "head" products capture the majority of attention, while a vast "tail" of niche products sits largely unserved, not because nobody wants them, but because the economics of physical distribution made it too expensive to stock, promote, and sell them.==[^3][^4]
The internet changed those economics. Amazon could stock millions of books. Spotify could offer every obscure album. Netflix could carry films with tiny audiences. Anderson's thesis was that selling a few of many things could eventually rival selling many of a few things — and that liberating this latent niche demand would produce a more diverse, vibrant cultural and commercial ecosystem.[^5]
The key lever was cost. As costs of production, distribution, and discovery fell, niche supply could finally meet niche demand. The pattern repeats in every domain where friction collapses.[^4]
When the web expanded in the 2000s, thinkers began applying Anderson's framework to software itself. The SaaS era produced a long tail of commercial applications — by 2025, Scott Brinker's annual martech landscape alone counted over 14,000 distinct commercial tools. This seemed like the long tail coming alive in software.[^2]
But there was a ceiling. Even in the SaaS era, building software still required engineers. The economics of software development constrained how far down the tail anyone could go. You simply would not hire a team of developers to build a product for 100 users — the unit economics didn't work. The "long tail" of software was still bounded by a hard floor: the minimum team size, the minimum development time, the minimum addressable market that could justify building at all.[^6]
That floor is what AI is now removing.
Long before AI existed as we know it, some of computing's foundational thinkers had a radical vision: that ordinary people should be able to build their own computational tools. Alan Kay, the computer scientist behind object-oriented programming and much of the graphical user interface as we know it, articulated this in 1984: "We now want to edit our tools as we have previously edited our documents."[^7]
Kay's vision — what he and others called "end-user programming" — imagined a world where the distinction between "using software" and "building software" dissolved. Just as anyone could write a Word document or a spreadsheet formula, anyone should be able to create a custom digital tool. For decades, this remained an aspiration without a technical path. Programming languages required formal training. IDEs required context. Building something real required understanding systems architecture, debugging, deployment, security — an entire professional discipline.[^7]
The bottleneck, as Geoffrey Litt of Ink & Switch and Notion articulates, was the translation layer: converting fuzzy human intent into precise, executable code. LLMs are dismantling that bottleneck.[^7]
Geoffrey Litt, a design engineer at Notion and researcher at Ink & Switch, has arguably done more than almost anyone to articulate what "software that can be reshaped by its users" would look and feel like. In his influential essay "Malleable software in the age of LLMs" (2023), he wrote: "I think it's likely that soon all computer users will have the ability to develop small software tools from scratch, and to describe modifications they'd like made to software they're already using."[^7]
==Ink & Switch's broader vision of malleable software is captured in a manifesto from 2025: software where "modification becomes routine, not exceptional. Adaptation happens at the point of use, not through engineering teams at distant corporations." The goal Litt describes is a world where the barrier between having an idea and building a tool approaches zero.==[^8][^9]
He draws a powerful contrast: today, when you want to move furniture in your house, you just do it. But when you want to change how a piece of software works, you typically can't — you must wait for a development team at a distant company to prioritize your request, if they ever do. Malleable software would feel more like furniture: yours to rearrange.[^9]
Crucially, ==Litt distinguishes between disposable software (throwaway apps you use once) and malleable environments — stable, evolving tools you craft over time that become genuinely yours, reflecting your workflows and mental models.== Both matter in the new paradigm, but the latter is the more profound shift.[^9]
On February 2, 2025, Andrej Karpathy — co-founder of OpenAI and former head of Tesla Autopilot — posted what he called a "throwaway tweet." He wrote: "There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists."[^10]
The tweet described his experience using AI coding agents like Cursor with Claude Sonnet. He was building projects by dictating what he wanted in plain English, accepting all AI suggestions without reviewing diffs, pasting error messages in without commentary, and watching working software emerge. "I'm building a project or webapp," he wrote, "but it's not really coding — I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works."[^10]
Collins Dictionary named "vibe coding" its word of the year for 2025. Online searches exploded. The term gave a name to something millions of people had begun doing or dreaming about: building software not by knowing how to code, but by knowing what you want.[^11]
A year later, in February 2026, Karpathy returned with a refinement. The term was "agentic engineering". While vibe coding was mostly playful — weekend projects, throwaway experiments — agentic engineering describes the more serious, professional-grade practice that has emerged as AI coding agents have become dramatically more capable. Engineers no longer write code directly; they direct and oversee agents that do.[^11]
The distinction matters philosophically. Vibe coding was about access — suddenly everyone could build something. Agentic engineering is about leverage — experienced builders can now accomplish in hours what previously took months. The two phenomena together create a spectrum: from the absolute beginner building a dog-feeding tracker via Replit to the experienced engineer using Claude Code to architect complex distributed systems with a small army of AI agents.[^6]
Both ends of this spectrum are building more niche software than was ever economically possible before.
In traditional manufacturing, high fixed costs require large markets to recover those costs. The same logic applied to software for its entire history. Building a professional-grade application required: hiring engineers (expensive), maintaining a codebase (time-consuming), handling deployment and infrastructure (specialized), providing customer support (ongoing cost). This meant that only products serving large enough markets could justify the investment.[^6]
Addy Osmani (engineering lead at Google Chrome) captured the transformation crisply in a widely-shared 2025 essay: "When the time and expertise required to create software drops dramatically, the minimum viable market size for a software project approaches one. This isn't just a quantitative change — it's a qualitative transformation in what constitutes a 'marketable' software product."[^6]
This reframes the entire economics of software. In the old model: software development was like industrial manufacturing, with high fixed costs requiring large markets to justify production. In the new model: it's more akin to craft production, where tools can be created for markets of any size, down to individual users.[^6]
The most vivid intellectual framing of this shift comes from Anish Acharya, a General Partner at Andreessen Horowitz, in a pair of essays from late 2025 and early 2026.[^12][^13][^14]
Acharya asks us to think about video creation. For most of history, making video content that could reach an audience required a studio, expensive equipment, and industry gatekeepers. Then came consumer cameras, YouTube, and the smartphone. Suddenly, anyone with a camera and editing software could find their audience. The supposed "long tail" of video turned out to be vastly larger than anyone imagined — from Mr. Beast to thousands of niche cooking channels to Dwarkesh podcast. The world, it turned out, was short video in 2006. We just didn't know it yet because the supply constraints were so severe.[^14]
==Acharya's thesis: "The world is short software."[^12]==
==Just as the world was short content in 2006 — when anyone pointing to 100 cable channels would have said "surely that's enough" — the world is short software today. We can point to hundreds of enterprise SaaS tools and say "surely that's enough." But we are nowhere near saturation of actual human need. There are entire categories of software that have never been built because they would serve too small an audience. There are workflows optimized for 20 people in a specific industry. There are tools a single doctor would use every day that no startup has ever built because the TAM was too small.==[^12]
LLMs change this. "You wouldn't hire a bunch of engineers to build a product for 100 people, but you can build (and monetize!) smaller-TAM products using app-gen and coding tools."[^12]
Andreessen Horowitz's broader institutional view, captured in a March 2026 essay, frames what's coming as a bifurcation. One class of software — thin UI wrappers around commodity data, incumbent systems of record that survive on switching costs rather than genuine value — will face real pressure. Another class — ==software that delivers genuine, defensible, compounding value — will expand dramatically as AI lowers the cost of building and unlocks new markets previously too small to serve.==[^15]
The essay makes an important point: "==The world is still short software. We are nowhere near saturating the world's demand for high-quality software. And as code becomes cheaper, we should just expect to see the market demand more==."[^15]
Scott Brinker, the marketing analyst behind the annual "Martech Landscape Supergraphic" (which by 2025 had catalogued over 14,000 commercial martech products alone), introduced the concept of the "hypertail" to describe what's happening beyond the already-long tail of commercial software.[^2]
The long tail was the universe of commercial apps you could find in an app store or on Product Hunt — still built by developers, still requiring investment, still targeting some minimum viable audience. The hypertail is the layer beyond: billions of custom micro-apps, nano-apps, and automations built by individuals and teams for specific workflows, specific contexts, specific moments. Not for general distribution. Not for a market. Just for a particular problem, a particular team, a particular need.[^1][^2]
Brinker writes that we may be approaching "a turning point where the number of commercial apps in the tech stack peaks and future growth of the stack — which overall we think could be exponential — will come from custom software, a cornucopia of custom apps, agents, and automations."[^16]
The Gartner prediction encodes this numerically: by 2030, 40% of enterprise application portfolios will consist of custom applications built on AI-native development platforms, up from just 2% in 2025. That is not incremental change. That is structural.[^17]
Commercial software in the "long tail" still followed certain rules: it needed marketing, distribution, customer support, version management, backward compatibility. The hypertail is liberated from most of these constraints. A hypertail app:
This creates something genuinely new: ==software that is personal infrastructure, as natural and individual as how you arrange your desk, organize your notes, or structure your morning routine.==[^9][^6]
Eugenia Kuyda is one of the most important figures in this story. She founded Replika in 2017 — an AI companion app that anticipated the emotional relationship people would form with AI systems years before ChatGPT existed. In late 2025, she launched Wabi.ai, securing $20 million in pre-seed funding led by Andreessen Horowitz.[^18][^19]
Wabi describes itself as the "YouTube of apps" — a social platform where anyone can create, share, remix, and discover personal mini-apps using simple natural language prompts. The vision is explicit: "This was really made to help people who have nothing to do with coding or the tech world to very quickly create apps from their daily lives," Kuyda told TechCrunch.[^18]
==a16z's Anish Acharya, who led the investment, framed it succinctly: "It's very rare to find someone who's got a track record for predicting what consumers will want, and we think she's doing it again."==[^18]
What Wabi is building — and what makes it philosophically distinct from earlier no-code tools — is not just an app builder. It's a social layer for personal software. Apps can be liked, commented on, remixed. You can find your best friend's custom workout tracker and adapt it for yourself. The discovery mechanism mimics how content spreads on TikTok, but what spreads is functional software, not passive media.[^18]
Kuyda's most arresting phrase for the vision is: "Think of it as an operating system built on the platform of you."[^20][^21]
This is a significant philosophical claim. Today's apps are generic — they exist at a fixed-context layer, built for the average user, knowing nothing about who you actually are. A fitness app doesn't know what book you're following, what your gym looks like, or how your nutrition habits intersect with your training goals. A finance app doesn't know your specific risk psychology, your career trajectory, or your family obligations.[^22]
Personal software, in Kuyda's vision, is different in kind: it accesses your context, your data, your goals across all dimensions of life, and builds tools that are expressions of you rather than generic utilities you adapt to. The software becomes a layer that is as individual as your digital life itself.[^22]
There is a deeper philosophical shift underneath the product trends. For most of human history, the division between tool-users and tool-builders was sharp. A carpenter used hammers; blacksmiths made them. A musician played instruments; luthiers built them. The specialization that made industrial civilization productive also created this division: you used the tools your society provided, and you adapted yourself to those tools.
Computing inherited this division. Software companies decided what tools existed. Users adapted. The GUI was an enormous advance in access — it brought billions of people into the user category — but it did nothing to close the gap between users and builders. If anything, the complexity of modern software engineering widened it.[^23][^7]
What AI tools are doing, at their best, is collapsing that division for software specifically. The person who knows the most about what a tool should do — the domain expert, the user who lives inside the workflow — can now increasingly build that tool. A doctor who has an idea for better patient tracking can prototype it in an afternoon. A teacher who wants a custom adaptive quiz system can build it herself. A fintech founder who understands a specific underserved market segment can build the exact product that segment needs without hiring a ten-person engineering team.[^23]
This is not just efficiency. It is a shift in agency.
Addy Osmani and others repeatedly invoke the spreadsheet as the closest historical analogue. When VisiCalc and then Excel gave non-programmers the ability to perform complex calculations and data modeling, it didn't replace software engineers — it created an entirely new category of capability that hadn't existed before. Business analysts, finance teams, and operations professionals built things that no professional developer would ever have built for them, because the TAM was too small, the use case too specific, the workflow too idiosyncratic.[^6]
Spreadsheets were "software" in the broad sense — they executed logic, stored state, produced outputs — but they operated in a different layer than traditional applications. They were personal infrastructure.
AI-assisted development is the next layer up: full-stack personal software infrastructure. Not just formulas in cells, but complete applications, with interfaces, databases, integrations, and AI functionality, built by the person who will use them.[^6]
Acharya adds a dimension that tends to get overlooked in efficiency-focused discussions: software as a medium of expression. We already live in a world where having a personal YouTube channel, a Substack, or a Twitter presence is normal — a way individuals express ideas, build audiences, and create leverage. Software has not historically been a medium for personal expression in this way, because the barrier to entry was too high.[^13]
That is changing. Acharya cites developers who have built viral "software posts" — apps that went viral not because they were enterprise products but because they were expressive, zeitgeisty, funny, or personally resonant. Visualizing a city's parking enforcement patterns. Rendering declassified documents interactively. A manifestation practice app built by a non-programmer for their community.[^13][^14]
==Software value also has a structural advantage over content value: it compounds rather than decays. A YouTube video has a 2-5 year depreciation schedule. An app that solves a real problem can accrue users and value indefinitely after a single launch. This makes software creation an economically powerful form of personal expression — potentially more durable than content creation.==[^13]
This shift has a disruption side, not just a creation side. In February 2026, financial markets experienced what Jefferies traders called the "SaaSpocalypse" — a violent selloff that erased $285 billion from software stocks in 48 hours. The trigger was a growing realization that AI would fundamentally challenge the traditional SaaS business model: per-seat licensing, horizontal tools with low switching costs, and software that substitutes for human labor at scale.[^24]
The threat vectors are multiple:[^25]
Forrester's analysis distinguishes which software survives: "Vertical- or domain-specific SaaS vendors will have a greater chance of survival. This market is already large and expanding, with vertical software projected to grow from roughly $133.5B (2025) to $194.0B (2029 forecast). SaaS vendors that offer differentiated solutions that address complex industries or that control unique, proprietary data will survive."[^25]
Horizontal commodity tools — thin UI wrappers around generic workflows — are most vulnerable. Deep vertical software, proprietary data moats, and network-effect platforms are most defensible.[^15][^25]
The SaaSpocalypse narrative focuses on what breaks. But the more interesting story is what gets built in its place. Thoughtworks documented eliminating three narrow-feature-set SaaS platforms in 2025, replacing them with custom AI-native internal tools. This is not the destruction of value — it is the redistribution of value from generic commercial software to custom personal and organizational software.[^26]
Scott Brinker framed it as "instant software" — just add ideas. The entire model of how software gets created is shifting from a product-centric model (build once, sell many times) toward a generation-centric model (generate on demand, exactly as needed). In many ways, this resembles the shift from manufactured goods to services — except that software as a generated artifact can still be durable, composable, and compounding in value.[^2]
If anyone can build software, what does competitive advantage look like? The answer that emerges from multiple thinkers is: deep domain understanding. When the coding skill is commoditized by AI, the differentiator shifts upstream to: knowing what to build, for whom, with what constraints, and why.[^27]
This is actually an enormous advantage for domain experts who have never been able to build before. A doctor who understands exactly what patient workflow friction exists can now build exactly the right tool. A trader who understands exactly what behavioral patterns produce losses can now build exactly the right risk management software. An Indian retail investor who understands exactly what financial education gaps exist can now build exactly the right fintech tool for that community.
==The a16z framing of "narrow startups" and "extraordinary specialization" points in the same direction: the AI era advantages those who can get deep into a specific problem and build exactly the right solution for it, rather than general solutions that sort-of work for everyone.==[^28]
One practitioner captured the practical reality vividly in a January 2026 LinkedIn post. He described spending nearly all of his professional time building AI software products for single users — one person at a time. Someone would describe their workflow, he would help them think clearly about how their work actually flows, and then build a tool for that single person. That tool would then spread organically when others saw it. "The bottleneck isn't building any of this. That part is fast now. The bottleneck is getting feedback from the stakeholders. And knowing how to pull it all together."[^29]
This is the inverted startup model. Instead of building for a market and hoping individuals adopt it, you build for one person and let network effects pull you into a market. The unit economics of building changed fast enough that this is now feasible.
Not everyone celebrates this shift uncritically. Experienced developers have noted what one observer called the "==competency illusion" — the less expertise you have in a domain, the more impressed you are by AI output, because you lack the knowledge to identify its flaws.== A non-technical founder who vibe-codes a product may not realize the security vulnerabilities, the architectural dead-ends, or the technical debt baked in.[^6]
Anish Acharya himself, despite leading Wabi's investment round, stated flatly that "vibe coding everything is a lie" when it comes to serious enterprise software. The distinction matters: vibe coding is genuinely revolutionary for personal software, niche tools, and prototyping. It is not (yet) the same as professional software engineering for systems that require reliability, security, and scale.[^30]
The explosion of the hypertail also creates what Scott Brinker calls a "==Big Ops" challenge: governing and orchestrating a universe of billions of custom apps, agents, and automations running in parallel across an organization is itself a massive technical and managerial problem. The very democratization that creates richness also creates chaos. Who owns the custom app when the person who built it leaves? Who secures it? Who updates it when the underlying model changes?[^31]==
==These are not deal-breakers for the thesis — they are the second-order problems that create the next generation of infrastructure businesses. But they are real.==
There is also a genuine question about monetization. Traditional software had clear value capture: licenses, subscriptions, seats. Personal software built for one person has no obvious commercial model. The YouTube analogy is actually imperfect here: YouTube content captures value through advertising and sponsorships tied to audience attention. A custom app built for your own workflow doesn't have a natural audience.[^32]
The answer may be that most personal software doesn't need to capture commercial value — it creates private value (time saved, problems solved, leverage gained) that benefits the builder directly. The commercial opportunity is at the platform layer: Wabi, Replit, Cursor, and others that enable personal software creation. Just as the real money in the YouTube era accrued to YouTube itself as a platform (and to a layer of professional creators), the real money in the personal software era may accrue to the platforms and to those who build niche software at the right scale.
Financial services is among the most domain-knowledge-intensive industries, and historically one of the most underserved by general-purpose software. The workflows of a retail trader in Bangalore are not the workflows of a hedge fund manager in London. The compliance requirements for an Indian NBFC are not the same as those for a US fintech startup. The behavioral patterns of a first-generation investor are not the patterns embedded in software built for sophisticated market participants.
Vertical SaaS in fintech has been growing because domain-specific solutions outperform generic ones in high-stakes domains. But the hypertail opens something even more specific: tools built by people who are the user — founders who are also domain experts, building for communities they genuinely understand.[^27][^25]
The a16z insight applies directly: AI niche revenue models can transform previously unprofitable small markets. As one founder noted, AI is making previously unprofitable niche markets viable by dramatically increasing customer value while driving costs down — in vertical SaaS, AI can turn a $120M market into a $1.2B opportunity by serving it more precisely.[^27]
For a technical founder in fintech, the implication is specific and actionable. The person who can combine:
...has a structural advantage that simply did not exist before 2024. The cost to build a working product for 500 niche users has collapsed from $500,000+ (a team, months of time) to potentially $50,000 or less (one or two people, weeks of time). This makes previously uneconomic markets economically rational.[^33]
The relevant ==Sam Altman prediction is not just about the "one-person billion-dollar company" as an abstract aspiration — it is about the structural economics that enable it. When AI agents can handle research, marketing, customer support, and code generation, the bottleneck to building shifts entirely to insight and judgment about what to build. Those who have that insight — especially in underserved, niche, domain-specific markets — can now execute on it without the traditional capital requirements.==[^34][^35]
The arc from Chris Anderson's Long Tail through Alan Kay's end-user programming dreams through Andrej Karpathy's vibe coding moment to Eugenia Kuyda's "OS built on the platform of you" is a coherent story. It is the story of software becoming a genuinely human medium — not just a product category built by specialists for mass markets, but an expressive, personal, composable layer that anyone with a problem and an AI can contribute to.[^21][^13]
The "long tail" of software that Anderson gestured at in 2004 when talking about open source was a commercial long tail — many niche products, but each still built as a product. The hypertail that is emerging is something different: it is the personal layer, the community layer, the moment layer — software that exists for reasons other than commercial distribution, for audiences smaller than any product team would ever have justified, built faster than any sprint cycle would have permitted.[^36]
==The world was short software. The supply chain was too expensive. The minimum viable market was too large. AI has broken those constraints.==
What gets built in the space that opens up will be — must be — the most human software ever created, because for the first time it will be built by the people who know most intimately what they need.
What do Amazon, Netflix and Spotify have in common? They're making gold on products...