The annual recurring revenue (ARR) numbers for AI coding tools look like a rocket ship. Cursor hit $500 million ARR, and Lovable went from $17 million in February to $75 million by July. On paper, this looks like unstoppable momentum.

But look closer, and you notice something different happening. Traffic to Lovable is down 40%, Vercel’s v0 has plunged 64% since May and Bolt.new slipped 27% since June.

When revenue climbs while usage falls, that’s not pure growth. That’s a warning sign.

Many of the people building these tools know it. Eric Simons, CEO of Bolt.new, said it plainly: “This is the problem across all these companies right now. The churn rate for everyone is really high… You have to build a retentive business.”

He’s right. And I think this moment reveals something bigger about where AI is headed.

The One-Size-Fits-All Promise

Three forces are working against generic AI coding assistants.

First, developer sentiment is cooling even as adoption grows. The Stack Overflow Developer Survey 2025 shows usage climbing from 70% in 2023 to 76% in 2024, but favorability dropped from 77% to 72% over the same period. Developers are using the tools more but liking them less, which suggests to me that the novelty is wearing off and the limitations are becoming clearer.

Second, enterprises are hitting compliance walls. Many AI coding tools were built for individual developers, not enterprise procurement, and that gap is showing. FINRA, GDPR and HIPAA all create requirements that many generic tools weren’t designed to meet. Some AI coding assistants, such as Augment Code, hold ISO/IEC 42001 certification, while other major tools still lack specialized AI compliance certifications.

Third, switching costs are simply too low. When every tool feels roughly the same, users bounce between them looking for something better. There’s no moat in being “pretty good at everything.”

The Growth In Vertical AI

While generic coding tools plateau, specialized AI applications are scaling fast, and the numbers are striking. The vertical AI market sits at $12.9 billion in 2024 and is projected to hit $115.4 billion by 2034, growing at 24.5% CAGR. Banking, financial services and insurance (BFSI) alone captures 21.5% of that market, processing over 2.5 quintillion data bytes daily.

The pattern is clear: Enterprises are buying AI that solves specific problems in specific industries, not generic tools they have to customize themselves.

There’s another shift happening beneath the surface as well. CB Insights reports that “Early-stage genAI companies with the fastest-growing headcounts are concentrated in AI agent applications—and more specifically in voice AI development.” The interface is moving from text to conversation, and the winners are getting more specialized, not less.

Big Tech Is Buying, Not Building

The strategic acquirers have made their bet, and M&A activity in AI hit record levels in Q2 2025 with 177 deals, nearly double the quarterly average of 89 since 2020. IBM made 3 AI acquisitions during the quarter, while Intuit, NVIDIA, Databricks and Salesforce each made 2.

The Moveworks deal is particularly instructive. ServiceNow acquired Moveworks for $2.85 billion. They bought it for the application layer: the front-end AI assistant, enterprise search and technology integrations. The companies already had 250 mutual customers, and ServiceNow can now expand its reach to “every requestor in an organization.”

The signal is clear: Large enterprise software players are buying specialized AI stacks rather than building generic copilots in-house.

What Comes Next

We’re entering Wave 2.0 of AI development, where the infrastructure layer is maturing and the next wave of value creation will likely happen in the application and middle layers.

CB Insights State of AI Report shows that among the 1,500-plus tech markets they track, those focused on specific industry or technical challenges saw the greatest number of AI deals in Q2 2025, while LLM developers tied for ninth place with just five deals. Investors increasingly expect greater value creation from applications than infrastructure.

Capital remains abundant, but it’s shifting. Funding to private AI companies reached $47.3 billion in Q2 2025, and combined with Q1’s record $116.1 billion, 2025 already surpassed 2024’s full-year total. But the top 10 rounds accounted for 60% of the quarter’s funding, which means mega-rounds for proven winners and seed capital for specialists. The middle is getting squeezed.

The Quiet Signal

The traffic decline isn’t a death sentence for AI coding tools, but it is a signal, and it’s worth understanding what’s actually driving it.

Part of the drop is simply the novelty wearing off. Many “vibe coders” tried these tools for fun, not because they needed them for daily work. We’ve seen this pattern before: Users play with the technology, share a few outputs and move on once the excitement fades. That cohort was never going to stick around.

But the deeper issue is on the professional side. AI coding tools still struggle to deliver production-quality code consistently. Engineers find themselves experimenting with prompts and specs just to get something viable, and even then, they often spend more time fixing AI-generated bugs than they would have spent writing the code themselves. The tools get you 90% of the way there, but that last 10% can be brutal.

I think the tools that succeed will be capital-efficient, strong on product-market fit and deeply embedded in specific workflows; losers will be generic tools competing on features alone, relying on hype without retention.

The next wave may not be won by the loudest tools. It could be won by the ones that become essential.

Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?

Source link