Over the past three years, capital markets poured $200+ billion into AI companies, with nearly $80 billion specifically targeting GenAI and LLMs. We hit peak funding in Q2 2025, and we’re now sliding down the back slope of the hype cycle.
The numbers tell a sobering story: The entire LLM sector has generated perhaps $10 billion in total revenue against $200+ billion in investment: a 20 to 1 gap.
Making matters worse, companies in the LLM eco-system are touting annualized recurring revenue (ARR) figures without disclosing churn rates, upgrade patterns, or actual renewals. It is financial theatre designed to maintain momentum and secure more funding. We are heading towards what is calls the “trough of disillusionment” on the Gartner Hype Cycle.
But here’s the thing about descending from hype peaks: the view only gets clearer as the fog lifts. And what’s becoming clear is where the real value lies.
Attention Shifting from Foundation Models to Development Infrastructure
While headlines focus on the next GPT release or Claude upgrade, the organizations seeing actual ROI have shifted their attention somewhere more practical: their software development infrastructure.
This isn’t about implementing AI for AI’s sake. It’s about carefully making LLM-powered tools available across the development flow, from specification and discovery through QA, over CI/CD, and to deployment.
But, and this is critical, your ability to capture these gains depends entirely on your current software development maturity. AI won’t fix a broken development process. It will accelerate whatever process you have. If that process is already fragmented, bureaucratic, or siloed, AI will make those problems worse, faster.
Three Patterns Emerging: Which Organization Are You?
After engaging with dozens of companies over the past 12 months, I’ve observed three distinct patterns in how organizations are approaching AI-enabled development:
The Leaders: Orchestrated Experimentation
The best development organizations are building cross-functional AI tool chains right now.
Here’s what this looks like in practice:
- Product managers use Claude or ChatGPT to spot gabs and suggest improvements, as they iterate on epics, user stories and acceptance criteria, but they do not mistake that for proper discovery and data insights.
- Designers leverage Figma AI or similar tools to explore multiple design directions simultaneously, but do not drop UX best practices.
- Developers work with GitHub Copilot or Cursor to accelerate implementation, but keep control with code review, mentoring, and other best practices.
- QA teams employ AI to widen test coverage based on specification <=> code comparison
Critically, these organizations understand the risks. They know LLMs hallucinate. They know code suggestions need review. They’ve established guardrails, review processes, mentoring, and openly share views about what AI can and cannot do.
The result? Employees across these organizations report 20-50% productivity improvements—not from any single tool, but from the compounding effect of AI assistance across the entire value chain.
The Middle: Fragmented Adoption
Most organizations fall into this category. They’re experimenting with AI, but in disconnected pockets: A vibe-coded prototype in one team, an AI-enabled IDE in another, automated test generation in a third, and some designers using AI, others refusing to touch it.
Each initiative is educational for the individual, but there’s no coordination, no shared learning, and no way to measure aggregate impact. Teams can’t build on each other’s discoveries.
The real risk here isn’t standing still. It’s the illusion of progress. Leadership sees AI adoption happening and assumes they’re competitive, while in reality, they’re falling behind organizations with coordinated strategies.
The Strugglers: Dysfunction Accelerated
The worst development organizations fall into two camps:
The Ostriches: They’ve decided AI is hype and are ignoring it entirely. Their competitors are gaining 20-50% productivity advantages while they debate whether AI is “real” or another passing fad.
The Turf Warriors: They’ve bought into AI hype so completely that different functions are now competing to automate each other out of existence. Product is trying to use AI to eliminate the need for designers. Engineering is exploring whether AI can replace product managers. Design is investigating AI that can go straight from concept to code.
In both cases, a bad development culture doesn’t improve. It accelerates toward dysfunction. AI-generated code compounds technical debt faster. AI-written specifications amplifies a feature-factory mindset.
Get an outside in perspective
At Copenhagen Product Collective, we have close to three decades of international experience with software product development. We know how to take advantage of new technologies and ways-of-working.
If you want to accelerate the benefits you get from AI in your product development process, we can provide a maturity assessment of the organisation and recommend which steps to take first. Read here and reach out to hear more.
…but please remember: the trough is where sustainable businesses are built. The hype recedes, the tourists leave, and the builders get to work.
