Lovable announced their GPT‑5 integration and, in a less than 15‑minutes, rebuilt a working ChatGPT client complete with Stripe authentication. OpenAI is around a $500 billion secondary valuation; Lovable publicly raised $200 million at a $1.8 billion valuation in July. The comparison isn't about parity. Lovable rides OpenAI's models. Rather, it’s a signal. When the marquee product of our times can be recreated in just 15 minutes, it means one thing. Code is commoditizing.
Code and engineers were the blocker before. But the constraint has shifted. It's not about scaffolding or imagination anymore. It's about everything around the code. The data rights, distribution, workflow embedding, reliability, and the discipline to ship continuously. That's the real challenge. Those who ship products for a living have always known this truth.
Today, anyone with Cursor/VS Code and a capable model can conjure working software on demand. You don't even need the latest frontier model to get far. However, and this is crucial, not everyone can, and not everyone will, build software that lasts. The bottleneck was never the code itself; it's the systems powering it.
The Age of Instant Software
Since Anthropic's Claude Sonnet 3.7 release in February and OpenAI's summer updates, we've entered the era of instant software. Sam Altman calls this the "fast‑fashion era of SaaS." Social feeds overflow with emancipation headlines from content hustlers: "I rebuilt a multi‑billion‑dollar SaaS in two prompts." "Five apps in a weekend." "30 apps in 30 days." "Double your passive income with your personal CTO."
Code has become something you summon rather than learn through years of experimentation and commits. Yet empires of convenience follow predictable physics. When the promotional clips quiet and revenue speaks, the picture clarifies: what appeared to be a Cambrian explosion is more like cherry blossoms. Dazzling, brief, gone before you notice.
Meanwhile, the market adheres to the same power laws it always has. H1'25 saw generative‑AI downloads reach 1.7B with $1.87B in in‑app spend—yet revenue remained concentrated. ChatGPT dominates nearly every major market in downloads, MAU, and revenue; independent estimates place its mobile earnings orders of magnitude above its nearest competitor. Attention scattered. Money consolidated.
When code becomes a commodity, the act of making software becomes easy. But software was never difficult because of code. Code was never the true obstacle. The real challenge lies in transforming capability into systems that withstand users, budgets, and risk.
The Future Isn't Evenly Distributed
One year ago, a tidy thesis circulated widely: if everyone can create software, no one will buy it; personal SaaS would collapse under the weight of instant, bespoke apps.
That thesis now appears unrealized. Or at least incomplete. For now. As of today, the capabilities impact different groups differently:
Archetype A — Deeply technical. The productivity boost is real, but many professional software engineers treat AI as sophisticated autocomplete. They write code in chunks, controlling every edge case and function. Productivity rises, but the frontier hasn’t necessarily affected their day-to-day.
Archetype B — Technically savvy, not code‑proficient. This is the breakout group. They installed Cursor or VS Code, activated agent mode, and started shipping at a pace that overwhelms enterprise processes. Ideas transform into running features weekly. The surrounding systems—review, change management, access control—haven't caught up.
Archetype C — Uses tech purely as leverage. Lured by creator headlines, they attempted vibe‑coding, hit coherence walls, and quit. Often blaming the model rather than recognizing the craft of managing AI outcomes and their own AI literacy/skill.
A generational pattern is also emerging: Gen Z tends toward Archetype B; older decision‑makers gravitate to C. The latter group steers enterprise AI adoption and struggles with AI‑first practices. Eventually, Gen Z decision‑makers will ascend, displacing rigid habits. Yet democratized creation hasn't democratized distribution, data, or discipline. That's where separation begins.
There's an App for Everything Already
The market is saturated, and that's good. Mindfulness apps? Hundreds. Calorie-counting? Hundreds. Sun trajectory, astronomy, budgeting? All polished, affordable, and well distributed. Capability doesn't create need, and the infrastructure of excellent apps already exists.
Personal finance illustrates this perfectly. With today's tools, someone with minimal coding experience can ask a model to build a functioning money app: a few prompts, a payments integration, a hosted deployment—thirty minutes to a demo. But problems arise immediately:
Quality. The first version typically feels lifeless and brittle—merely imitating intent.
Evolution. The moment you attempt modifications, seams split; complexity compounds; the app breaks.
Distribution. Even if you reach "good enough," the path to app stores, web presence, and actual users is mountainous.
This explains why, despite skyrocketing usage of coding platforms and respectable ARR growth at tools like Cursor and Lovable, we don't see breakthrough products born from vibe‑coding that scale to significance. The most celebrated "wins" are shovels, not gold—platforms monetizing the desire to build, not the end products themselves. Case studies and showcases sound miraculous but reduce to familiar components—booking systems, forms, content management, an API or two—valuable locally but not globally transformative. Showcases aren't traction.
The other end of the spectrum is also being hit hard. The personal finance startups are panicking running one fire sale after another. Unable to justify their current employee counts. Monarch, Origin, Copilot Money can no longer justify a $129/year for beautiful products. They’re now forcing year long commitments at less than half their original pricing.
Creation costs collapse both ends of the spectrum. Let's examine the evidence.
The Big Lie
The creator headline economy sells aspiration. "I built five apps in a weekend" reads like a software P&L statement; in reality, it's almost always a media P&L. These creators are selling content, not SaaS. This doesn't make the demos fraudulent—it makes them effective advertising. The flywheel is straightforward: demo → audience → cohort → new demo. It's a media loop, not a product funnel.
Here's a specific, testable claim: there is no publicly verified, post‑2023, solo "vibe‑coded" clone, built by a founder with little to no prior engineering background, that has exceeded $5–10M ARR with durable retention, enterprise‑grade reliability, and customers beyond the creator's audience. If such an example exists, it should publish cohort data, churn rates, uptime statistics, security posture, and customer lists. The silence speaks volumes.
Now, instead of broad generalizations, let's examine the flagship case studies the platforms themselves highlight.
Lovable: what the "wins" actually say
Launched Gallery / Hall of Fame. Dozens of micro‑apps—dashboards, galleries, resume sites, SCORM packagers, "AI co‑founders." These demonstrate time‑to‑prototype, not time‑to‑traction. Most pages lack DAU, revenue, or retention metrics; many are single‑purpose CRUD applications with thin workflows and no moat.
Agency economics, not product economics. One of Lovable's most‑promoted stories features an agency owner "making $100k/month using Lovable." That represents services revenue. It validates Lovable as a tooling platform but offers no evidence of a Lovable‑built consumer or B2B product achieving audited multi‑million ARR independently.
"Case studies" without independent metrics. Branded project pages (ad‑ops "lift," growth‑lab "$42M+ managed," etc.) read like portfolio one‑pagers. They don't disclose verifiable user counts, cohort curves, churn rates, or uptime statistics. When "visit" counts appear, they represent vanity traffic, not meaningful usage evidence.
Video proofs of speed. Ten‑minute builds and agent‑mode demos compellingly demonstrate capability. They don't document distribution, conversion, or durability.
Lovable verdict: Impressive acceleration of making; limited evidence of scaled products with public, durable traction. The best outcomes are either (a) Lovable's own ARR growth (a platform story), or (b) agencies monetizing Lovable (a services story)—neither contradicts our claim about solo, vibe‑coded product businesses.
Replit: what the "wins" actually say
Northern Health (clinician; 4 days; 16+ mini‑apps; £100k+ annual savings). A genuine operational win—internal automation that reduces costs. It's not a public product with external distribution, audited ARR, or enterprise‑grade SLAs.
BatchData (CGO‑built tools: $62k+ annual savings; ~$30 prototype; ~1 hour to MVP). Again, this represents internal tooling savings, not market traction. Useful, real, but limited in scope.
GenAIPI ("$180k revenue in 6 weeks"). Revenue comes from education/content, not a SaaS product launched and scaled via Replit. It validates audience monetization, not product retention.
Startup showcases / hackathons. These highlight creative use and speed. They rarely include 12‑week retention, paying user counts, or uptime/security posture. Well‑publicized agent missteps in production underscore the gap between demos and dependable systems.
Replit verdict: Compelling time‑to‑first‑value for prototypes and internal tools; minimal public evidence of post‑2023, non‑technical founders shipping vibe‑coded products that achieve audited, durable multi‑million‑dollar ARR.
Bottom line: Getting to small demoable units is still the focus. Shipping valuable products is still a pipedream. Value concentration after 2023 occurred among distribution leaders at the application layer and platforms selling tools—not in solo, vibe‑coded clones growing to meaningful, verified ARR.
Inside the Enterprise: Pilots Everywhere, Value Rare
What we witnessed in the vibe‑coding boom—outsized promises, meager outcomes—is now playing out inside large companies. The push was genuine: boards demanded visible AI implementation, budgets expanded, and pilots proliferated across functions. The results disappointed. In a recent report, MIT starkly revealed that approximately 95% of enterprise gen‑AI pilots delivered no measurable ROI. S&P Global reflected the same pattern differently: the proportion of companies abandoning most of their AI initiatives rose to 42%, with nearly half of projects scrapped between proof‑of‑concept and broad adoption. The diagnosis remains consistent: technology isn't the primary limitation—integration, workflow design, and governance are.
Beneath the numbers lies a human story: pushback and fatigue. LinkedIn's mid‑2025 survey of U.S. professionals revealed a clear expectation gap—nearly half feel underprepared to use AI at work, and 41% report that the pace of AI change harms their well‑being. A companion study on training explains why: over half say company AI upskilling "feels like a second job," while only a minority receive structured instruction. Leaders, meanwhile, have raised expectations—multiple surveys show two‑thirds would reject candidates without AI literacy—yet most employees remain untrained and AI skills scarce. Misaligned expectations combined with skill shortages produce stalled adoption.
Within companies that do make progress, a pattern emerges: workflow redesign before UI reskinning; data contracts and identity mapping before model selection; governance and evaluation before launch. The pilot that successfully navigates procurement after months of data rights negotiations, observability implementation, and systems integration typically improves a single KPI by double digits—and becomes the template. The others become cautionary tales. Gartner's latest Hype Cycle places generative AI squarely in the trough of disillusionment. CIOs are responding with fewer experiments, stricter gates, and KPI ladders and unit‑economics tests before scaling. The spectacle fades. Budgets tighten. The only viable path forward is the unglamorous one: skills, governance, rights, and routing.
Results. Not Vibes.
The story of the next eighteen months is consolidation—of spending, skills, and patience. The winners won't resemble influencers; they'll look like operators. People who aren’t “vibe coding” but “context engineering.” Enterprises will hire for AI literacy and teach it; they'll connect models to systems of record; they'll route work across different model classes to protect margins; they'll pass audits with the same composure they bring to product launches.
If code is now inexpensive, the scarce resources are grit and execution. Most people don’t have that on a regular Tuesday. Even harder to find in an era where people have to juggle a day job while learning and experimenting constantly. Employees are being pushed against the wall by employers to adopt AI and exponentially become more productive. We need to focus less on the hype, and more on the results. The people who understand these AI systems deeply and know how to get what they want from it will be the ones who survive.