Key Takeaways

  • Investment focus is widening beyond foundation model builders to include applied tools like coding agents and answer engines.
  • Companies like Anysphere are demonstrating that developer productivity is currently the highest-value commercial use case for AI.
  • The market remains hungry for novel architectures and deep tech, maintaining strong interest in research-heavy outfits.

Money follows utility. For the last two years, the technology sector has been obsessed with the sheer size of foundation models—the bigger the parameter count, the higher the valuation. But the narrative is shifting. We are moving from a phase of "look what this can do" to "look how this changes my workflow."

The capitalization table is starting to reflect this fragmentation. While the giants continue their arms race for compute, a secondary layer of high-value startups is emerging, targeting specific verticals where accuracy and agency matter more than general chat capabilities.

Other fast-growing AI companies, including coding agent group Anysphere, search company Perplexity and AI research start-up Thinking Machines Lab, serve as the perfect cross-section of where smart capital is actually flowing. It’s no longer just about building the brain; it’s about building the hands that do the work.

Take the developer ecosystem. It is arguably the canary in the coal mine for AI adoption.

Anysphere, the company behind the AI-powered code editor Cursor, represents a pivotal shift in how we view "copilots." The early days of coding assistants were largely autocomplete on steroids. You typed a function name; it guessed the rest. Helpful? Sure. Revolutionary? Maybe. But Anysphere’s approach—and the valuation growth accompanying it—suggests the market is ready for agents that understand the entire codebase, not just the current file.

Here's the thing about coding agents: they offer immediate, quantifiable ROI. If a senior engineer costs $200,000 a year and an AI tool can handle 30% of their boilerplate work, the math solves itself. B2B buyers don't need to be convinced about the concept; they just need the tool to work.

Then you have the search dilemma.

For two decades, "search" meant a list of blue links. We were trained to act as the synthesis engine, clicking five different tabs to assemble an answer in our heads. Perplexity is betting that users are tired of doing that work. By positioning itself as an answer engine rather than a search engine, it challenges the ad-supported model that dominates the internet.

Is it risky to go up against the incumbents? Absolutely. But the traction suggests that for B2B research and quick information retrieval, the conversational interface is superior to the traditional keyword query.

It’s worth noting that this isn't just about software wrappers. The inclusion of Thinking Machines Lab in this cohort of growing companies signals that the industry hasn't given up on fundamental research. We haven't reached the end of history regarding model architecture. There is a persistent belief among investors that the current transformer-based approach, while powerful, might not be the final destination for artificial general intelligence. Labs that are tinkering with the underlying logic of how machines learn are still commanding attention, even if their products aren't as immediately consumer-facing as a chatbot.

There is a distinct rhythm to how these technologies mature. First comes the awe, then comes the application.

We are firmly in the application phase. The reason these specific companies are seeing growth is that they are solving the "last mile" problem. A massive LLM is great, but without a coding environment (Anysphere) or a citation-backed interface (Perplexity), it’s just raw potential. Businesses can't bill clients for raw potential.

This creates a new mandate for enterprise technology leaders. The question is no longer "What is your AI strategy?" That’s too vague. The question is now, "Which agents are you deploying to specific departments?"

If the coding agents are handling the software pipeline and the search agents are handling market intelligence, the human workforce is left with high-level strategy and synthesis. That is a radically different organizational structure than what we had in 2022.

The hype cycle is cooling down, but the deployment cycle is just heating up. Watching these specialized players—Anysphere in code, Perplexity in search, and research labs in architecture—gives us a much clearer map of the future than looking at the foundation models alone. The giants built the engine, but these companies are building the car.