Software creation is undergoing the same structural transformation that social media inflicted on content creation a decade ago. In 2025, new App Store submissions surged 24% to 557,000 — the first meaningful increase since the 2016 peak — driven not by traditional developers but by a new class of “vibe coders” wielding AI tools. Lovable hit $200M in annualized revenue within 12 months of launch. Cursor reached a $29.3 billion valuation. Y Combinator reported that 25% of its Winter 2025 cohort built codebases that were 95%+ AI-generated. The phenomenon is real, growing exponentially, and reshaping who gets to build software. But it carries significant risks: 45% of AI-generated code fails basic security tests, experienced developers are measurably slower with AI tools, and a nascent “vibe coding cleanup” industry has already emerged to fix what non-developers ship. This brief presents the verified data on both sides.
The App Store tells a reversal story seven years in the making
The Apple App Store’s submission data reveals a structural inflection point. After peaking at approximately 1.06 million new app submissions in 2016, submissions entered a sustained seven-year decline driven by Apple’s quality crackdowns — including the removal of 2.8 million “zombie” apps, mandatory 64-bit architecture requirements, and increasingly stringent review guidelines. By 2023, new submissions bottomed at 424,000, the lowest since 2012. Then the curve reversed. Submissions climbed to 448,000 in 2024 (+5.7%) and surged to 557,000 in 2025 (+24.3%), according to Appfigures data published December 5, 2025. Monthly submissions jumped from a long-term average of ~50,000 to ~78,000 in the second half of 2025 — a 60% increase. This is the App Store’s biggest release year in nearly a decade. Year New iOS App Submissions YoY Change 2016 ~1,060,000 Peak 2017 731,000 −27% 2018 474,000 −35% 2020 491,000 COVID bump 2022 434,000 Continued decline 2023 424,000 All-time low since 2012 2024 448,000 +5.7% 2025 557,000 +24.3% AppfiguresNichehunt Bloomberg Know StartupUnified AI Hub Bloomberg FinancialContent Second Talent +4 Digital Applied +2 Statista Apple Developer Axon Applause 36Kr appfigures 36Kr Appfigures attributed the reversal to three converging forces: AI and vibe coding lowering technical barriers so non-technical founders can build functional apps in days; TikTok and Instagram serving as viral distribution channels for indie apps; and subscription models proving financially viable for solo builders. The analysis noted pointedly: “It doesn’t mean these apps are competitive on features or experience, but they are going into the App Store nevertheless.” Google Play tells a starkly different story. While also seeing a modest 7.1% increase in new releases, Google aggressively purged its catalog — from 3.4 million apps in early 2024 to just 1.8 million by April 2025, a 47% decline. Google removed 2.36 million policy-violating apps in 2024 alone and tightened standards to require 20 testers over a minimum two-week period for new submissions. Apple, by late 2025, responded with new review guidelines targeting AI mass-produced clone apps (rule 4.1c) and mandating explicit user authorization for data transfers involving third-party AI APIs (rule 5.1.2i).
Vibe coding went from tweet to $29 billion in twelve months
On February 2, 2025, former OpenAI co-founder and Tesla AI director Andrej Karpathy posted on X: “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.” The post received 4.5 million views. By November 2025, Collins English Dictionary named “vibe coding” its Word of the Year for 2025, defining it as “the use of artificial intelligence prompted by natural language to assist with the writing of computer code.” The commercial explosion that followed has been staggering. Four companies illustrate the velocity: Cursor (Anysphere) emerged as the category leader for professional AI-assisted coding. Its revenue trajectory defies convention: $100M ARR in January 2025 → $500M by June → $1 billion+ by November. The company raised a $900M Series C at a $9.9 billion valuation in June 2025, then a $2.3 billion Series D at $29.3 billion just five months later — making it the fastest enterprise software company to cross $1B ARR. Enterprise revenue grew 100x in calendar year 2025. Key investors include Thrive Capital, a16z, Accel, Nvidia, and Google. Lovable, the Stockholm-based platform targeting non-developers, followed a similarly explosive arc. Launching in late 2024, it grew from $7M ARR at year-end to $200M ARR by November 2025 — faster than OpenAI, Cursor, or Wiz reached that milestone, according to CEO Anton Osika speaking at Slush 2025. Users created over 25 million projects in Lovable’s first year, with 100,000+ new projects daily. Its December 2025 Series B raised $330M at a $6.6 billion valuation, led by CapitalG and Menlo Ventures, bringing total funding to approximately $552M. Bolt.new, built by StackBlitz, launched in October 2024 and hit $40M ARR within five months — profitable, with 5 million users and over 1 million daily active users. The parent company raised $105.5M at a ~$700M valuation. Replit, the oldest player (founded 2016), saw its AI Agent product drive a revenue explosion from $2.8M in early 2024 to $150M annualized by September 2025, raising $250M at a $3 billion valuation. Notably, 75% of Replit’s customers never write a single line of code. appfigures appfigures TechCrunchHeadstart Orangesoft 36Kr Codingvibers DEV Community +3 Natively Wikipedia Collins DictionaryCNN Wikipedia Cursor TechCrunch Cursor Business Wire Bloomberg TechCrunch TechCrunch TechCrunchEntrepreneur Loop AI Funding Tracker AI Funding Tracker +2 AI Funding Tracker TechCrunch Getlatka Unified AI Hub GreenGeeks Shipper Shipper Codingvibers Vestbee TechCrunch Medium +2 These are not outliers. The broader data shows a market in hypergrowth. AI coding tools represent a $5–7 billion market in 2025, projected to reach $24–26 billion by 2030 at ~25% CAGR, per consensus estimates from Mordor Intelligence, Grand View Research, and ResearchAndMarkets. Total venture funding for the top five vibe coding companies alone exceeded $4.5 billion in 2025 — Cursor ($3.2B), Lovable ($530M), Cognition/Devin ($400M), Replit ($250M), and Bolt.new ($105.5M). Coding is now the single largest category in AI application spending, accounting for $4.0 billion or 55% of departmental AI spend in 2025, per Menlo Ventures. The code itself is increasingly AI-generated. Google’s CEO Sundar Pichai disclosed that 25–30% of Google’s new code is now AI-generated. Microsoft CEO Satya Nadella cited 20–30% at a fireside chat at LlamaCon in April 2025. GitHub Copilot surpassed 20 million users by mid-2025. Developer adoption is near-universal: 82% of developers now use AI tools weekly, per industry surveys, and 84% have tried at least one AI code generator. Y Combinator’s Winter 2025 data crystallized the shift. Managing Partner Jared Friedman disclosed that 25% of the W25 cohort had codebases that were 95%+ AI-generated — though he emphasized these were highly technical founders who chose AI over manual coding, not non-technical amateurs. YC CEO Garry Tan added: “This isn’t a fad. This is the dominant way to code.” Mordor Intelligence Cursor Getlatka Crunchbase News Shipper Menlo Ventures Codingvibers MIT Technology ReviewCodeRabbit GitHub Netcorpsoftwaredevelopment Netcorpsoftwaredevelopment TechCrunchSecond Talent Medium Codingvibers TechCrunch Sacra
No-code laid the foundation; AI coding built the skyscraper
The vibe coding explosion didn’t emerge from nothing. It sits atop a decade-long no-code/low-code (LCNC) movement that steadily proved non-developers could build real software. According to Fortune Business Insights, the low-code development platform market reached $28.75 billion in 2024 and is projected to grow to $264.4 billion by 2032 at a 32.2% CAGR. These figures sit at the high end of analyst estimates — Gartner’s own estimate for 2024 was closer to $12.3 billion — reflecting widely varying market definitions across research firms. The adoption trends are directionally unambiguous regardless of which estimate you use. In November 2021, Gartner predicted that 70% of new applications developed by organizations would use low-code or no-code technologies by 2025, up from less than 25% in 2020. Now past that deadline, Gartner has updated the prediction to 75% by 2026, suggesting the trajectory is on track even if the original timeline was slightly aggressive. Gartner VP Jason Wong separately predicted that citizen developers at large enterprises would outnumber professional developers 4:1 — a prediction originally made for 2023, though often misattributed to 2025 or 2026 in secondary sources. The statistic that “60% of custom apps are now built outside IT departments” is widely cited across vendor literature but difficult to trace to a single primary source. It aligns with verified Gartner predictions that by 2024, 80% of technology products would be built by non-technology professionals and that by 2026, developers outside formal IT departments would account for 80%+ of the low-code user base. Development speed improvements of “up to 90% faster” are supported by Forrester and S&P Global research, though realistic improvements typically range from 50–90% depending on use case complexity. The 90% figure comes primarily from vendor-commissioned studies. Adalo BlogFortune Business Insights Qubit App Builder +2 Medium +2 VentureBeat +3 Quixy
A 4-million-developer deficit makes UGS structurally necessary
The talent shortage transforms UGS from a nice-to-have consumer trend into an economic imperative. IDC projected a global shortage of 4 million full-time software developers by 2025, up from 1.4 million in 2021, meaning the worldwide developer labor force operates at roughly 85% capacity. IDC analyst Arnal Dayaratna noted: “There is no substitute for trained full-time developers that have the skills to architect digital solutions with due consideration for their long-term viability, scalability, and security.” The economic consequences are enormous. In May 2024, IDC published “Enterprise Resilience: IT Skilling Strategies, 2024,” projecting that IT talent gaps would cause $5.5 trillion in economic losses by 2026 through product delays, impaired competitiveness, and lost business — a figure actually revised down from a prior $6.5 trillion forecast, trimmed by approximately $1 trillion thanks to AI coding tools and personalized learning. More than 90% of organizations worldwide will feel the pain of this crisis by 2026, IDC projected. AI skills are the most in-demand capability, cited by 45% of enterprise IT leaders surveyed. Korn Ferry’s 2018 “Future of Work” study painted an even broader picture: a global human talent shortage of 85.2 million people by 2030 — roughly equivalent to Germany’s population — resulting in $8.5 trillion in unrealized annual revenues. A critical clarification: this figure covers all skilled workers across finance, technology/media/telecom, and manufacturing, not solely tech roles. The technology-specific (TMT) shortage projected by Korn Ferry is 4.3 million by 2030. Many secondary sources incorrectly frame the 85.2 million as a pure “tech shortage.” The implication is clear: even if every computer science graduate worldwide entered the workforce tomorrow, the demand for software would vastly outstrip supply. Enterprises cannot hire their way out of this deficit. Tools that enable non-developers to build functional software aren’t just convenient — they’re filling a structural gap the labor market cannot close. Kissflow Joget LinkedIn Second Talent LinkedIn Yahoo Finance HR Dive Yahoo Finance HR Dive Multiplier Korn Ferry +2 Korn Ferry
From User Generated Content to User Generated Software
The UGS thesis rests on a historical parallel: what YouTube (2005), Instagram (2010), and TikTok (2016 in China, 2018 globally) did for content creation, AI coding tools are doing for software creation. The progression from “you need a TV network to broadcast” to “anyone with a phone can be a creator” maps onto the emerging shift from “you need a funded engineering team to ship software” to “you need a laptop and a $20/month subscription.” The creator economy offers a preview of scale. Some 207 million content creators exist worldwide as of 2025, with the creator economy valued at $224 billion and projected to reach $528 billion by 2030. Only 0.5% of the world’s population currently knows how to code. GitHub’s former CEO Thomas Dohmke repeatedly articulated a vision of scaling from 150 million registered developers to 1 billion software creators by 2030, enabled by AI tools. At GitHub Universe 2024, he announced GitHub Spark, calling it a tool to “enable over one billion personal computer and mobile phone users to build and share their own micro apps.” The data on who is actually building supports the thesis. The State of Vibe Coding 2025 report, published by Vercel, found that 63% of vibe coding users identify as non-developers — entrepreneurs, product managers, marketing teams, and designers generating full-stack applications and user interfaces through natural language. Replit CEO Amjad Masad noted that “product managers are some of the best vibe coders” because they “break problems into clear steps and communicate precisely — exactly what AI needs to build effectively.” Named revenue milestones from non-technical builders, primarily sourced from Lovable’s portfolio: Lumoo, an AI fashion platform built by Henrik Skagerlind Fasth and Peter Thörngren (fashion/retail executives, not developers), reached €800K ARR in 9 months with 15+ brand clients including Gant. [Source: Lovable/Tech.eu, December 2025; company-reported figures] ShiftNex, a healthcare staffing platform, hit €1M ARR in 5 months serving 5,000+ healthcare users. [Source: Lovable Series B announcement, December 2025; company-reported] Q Group, a Brazilian EdTech company, generated €3M revenue within 48 hours of launching a premium product built on Lovable. [Source: Lovable; extraordinary claim — may represent bookings rather than recurring revenue] FormulaBot, built by data analyst David Bressler (not a developer) using no-code and OpenAI APIs, generates $220K/month ($2.64M annualized). [Source: Market Clarity] QuickTables, built by two founders who quit their jobs with a 60-day runway, now generates $100K+/year on Lovable. Lovable CEO Anton Osika framed the shift explicitly: “This is the age of the builder. The story belongs to the teachers, product managers, founders, and dreamers who now have the tools to bring their ideas to life.” Menlo Ventures partner Matt Murphy, who led Lovable’s Series B, added: “They’ve done what was previously unimaginable by turning a latent market of tens of millions of people into web developers.” DemandSage VdoCipher inBeat Agency Sequoia Capital GitHub Market Clarity Vercel Replit Unified AI Hub Lovable Know Startup Unified AI Hub Tech.eu Know Startup Unified AI Hub AI Funding Tracker +2 Market Clarity Unified AI Hub Unified AI Hub
The risks are not theoretical — they’re already measurable
The counterarguments against UGS carry serious empirical weight and should not be dismissed as incumbents defending territory. Multiple independent studies document real, quantifiable risks. Security vulnerabilities are systemic, not incidental. Veracode’s 2025 GenAI Code Security Report tested 100+ LLMs across 80 code-completion tasks and found that 45% of AI-generated code introduced OWASP Top 10 security vulnerabilities. Java code failed at a 72% rate. Cross-site scripting vulnerabilities appeared in 86% of samples; log injection in 88%. Most troublingly, Veracode found security performance did not improve with larger or newer models — CTO Jens Wessling called this “a systemic issue, not an LLM scaling problem.” Separately, Checkmarx found up to 70% of AI-generated code was insecure, and Apiiro documented a 322% increase in privilege escalation paths and a 2.5x increase in critical-severity vulnerabilities across Fortune 50 enterprise codebases using AI tools. The Lovable vulnerability was particularly instructive. Security researcher Matt Palmer discovered in March 2025 that 170 of 1,645 Lovable “Launched” showcase apps (10.3%) had critical security flaws — 303 vulnerable endpoints exposing emails, phone numbers, payment details, API keys, and home addresses to unauthenticated attackers. The root cause was missing or misconfigured Row Level Security policies in the Supabase backend. After Palmer privately disclosed the issue, Lovable initially denied it and deleted tweets about it. The vulnerability was assigned CVE-2025-48757 and publicly disclosed on May 29, 2025, after Lovable failed to implement a meaningful fix within the 45-day disclosure window. The productivity paradox is real. METR (Model Evaluation & Threat Research) published a randomized controlled trial on July 10, 2025, studying 16 experienced open-source developers across 246 real-world tasks on repositories they actively maintained. Before starting, developers predicted AI would reduce their completion time by 24%. The actual result: AI-allowed tasks took 19% longer than manual tasks. After completing the work, developers estimated AI had sped them up by ~20% — a 39-percentage-point perception-reality gap. Independent replication by consultant Mike Judge found a median 21% slowdown over a six-week test. Google’s 2024 DORA report of 39,000 respondents found that every 25% increase in AI adoption correlated with a 1.5% dip in delivery speed and a 7.2% drop in system stability. Code quality degrades as context accumulates. Research from Adobe, Chroma Labs, and Factory AI documents what practitioners call “context rot” — the systematic degradation of AI output quality as conversation threads grow. Adobe found model accuracy drops dramatically beyond 32,000 tokens. Factory AI, presenting at NeurIPS in December 2025, reported agents scored just 2.19–2.45 out of 5.0 on artifact tracking after context compression. GitClear’s analysis of 211 million lines of code found that code duplication increased 8–10x in 2024, code refactoring dropped from 25% to under 10% of changed lines, and code churn (revisions within two weeks of commit) nearly doubled versus 2021 baseline. Senior practitioners are sounding alarms. Fast Company’s September 8, 2025 article “The vibe coding hangover is upon us” documented the emerging backlash. PayPal senior engineer Jack Zante Hays described AI-generated codebases as “development hell,” citing a “complexity ceiling” where code breaks faster than AI can fix it. Java creator James Gosling told The New Stack: “As soon as your project gets even slightly complicated, they pretty much always blow their brains out. Vibe coding is not ready for the enterprise because in the enterprise, software has to work every fucking time.” Stack Overflow’s 2025 survey found only 33% of developers trust AI accuracy, down from 77% positive sentiment in 2023, and a mere 3% express high trust. A new occupation — “Vibe Coding Cleanup Specialist” — has emerged on LinkedIn, with programmers advertising services to salvage AI-generated codebases. The dating app Tea, built with AI tools, left an unsecured cloud database containing 72,000 sensitive images exposed in July 2025. Replit’s own AI agent deleted an entire database during a “code freeze” for SaaS investor Jason Lemkin in August 2025. Help Net Security eeNews Europe Help Net Security SoftwareSeni Semafor Matt’s Website SuperblocksGIGAZINE GBHackers SuperblocksDesplega MIT Technology Review InfoWorld SmarterArticles SmarterArticles Wikipedia +2 Market Clarity The New Stack Market Clarity IKANGAI Wikipedia Fast Company Middle East
The forward look: from 557,000 apps to a billion builders
Several converging signals suggest UGS is accelerating, not plateauing. Gartner predicts 75% of enterprise software engineers will use AI code assistants by 2028, up from less than 10% in early 2023. A separate Gartner prediction widely attributed to the firm — that 60% of new software code will be AI-generated by 2026 — could not be fully verified from a primary Gartner publication, though it is consistent with the firm’s broader forecast trajectory. Google and Microsoft’s own disclosed figures (25–30% of new code already AI-generated) suggest the 60% threshold is plausible within 12–18 months. Replit launched its Mobile Apps product on January 15, 2026, allowing users to describe an app idea in natural language, generate a working React Native application, preview it instantly via QR code, and publish to the App Store with what the company claims is “three clicks.” Users still need an Apple Developer account ($99/year), but the technical barriers to going from idea to published iOS app have collapsed to near zero. Replit is reportedly nearing a $9 billion valuation in upcoming funding. The company’s case study: Dan Kempe, a non-technical solo builder, built “Flash News” during a Replit buildathon and achieved ~300 downloads. Sam Altman and Dario Amodei both predict the first one-person billion-dollar company by 2026, enabled by AI agents handling the work that once required engineering teams. Amodei predicted in March 2025 that within six months, 90% of all code would be written by AI. A Stanford study found that employment among software developers aged 22–25 fell nearly 20% between 2022 and 2025, coinciding with AI coding tool adoption — an early labor market signal. But Gartner’s own warnings temper the optimism. The firm predicts that by 2028, prompt-to-app approaches will increase software defects by 2,500%. The tension between democratization and quality will define the next phase. Appfigures analysts observed that while AI lowered barriers to entry, it also lowered average app quality — the 2025 surge in submissions did not correspond to a surge in competitive, well-built products. Gartner Replit ForkLog CNBC Replit Orbilon TechnologiesRory Callaghan MIT Technology Review ArmorCode appfigures
