LLM Application

5/1328 → 17/964; +1.39 percentage points of formation share.

What stands out across this formation is how quickly the tooling layer has commoditized into infrastructure plumbing rather than product differentiation. Langfuse, Baserun, Parea, Helicone, and Traceloop are all, in practice, competing for the same job of tracing prompts and catching hallucinations before users see them, while LiteLLM and BricksAI solve the adjacent problem of routing calls across 100+ providers without rewriting code each time. That's a lot of redundant observability and gateway plumbing for one batch, and it signals a market still figuring out which layer actually deserves to be a standalone company. A smaller second cluster, Pincites for contract review and Datacurve for coding data generation, shows founders building actual vertical products on top of LLMs rather than tools for other builders, which is a meaningfully different bet. The 4.68x growth here dwarfs the 0.48x pace of Developer Tools generally, suggesting LLM-specific infrastructure is expanding far faster than developer tooling as a whole rather than simply riding that wave. Prompt injection research, with 602 papers and 25 recent, tracks closely with what PromptArmor is now selling as compliance and security, showing the security concerns are catching up to deployment rather than leading it. None of this could have happened before late 2022, when ChatGPT's release made LLM output reliability and cost suddenly a production problem for thousands of companies at once, rather than a research curiosity.

No comparable outside activity is connected yet.