Notable Silicon Valley startup accelerator Y Combinator held a Demo Day for its inaugural Fall cohort this week.
The 95 startups in this latest batch looked quite similar to recent YC cohorts in the sense that it includes many AI startups. If I did my math right, 87% of the startups in this batch are AI companies. Similar to YC’s summer and winter batches this year, there was a noticeable focus on customer-service-related AI and AI agents.
But among these, four companies piqued my interest, and they all had something in common: They are building tools to help companies monitor their AI applications to quickly solve or prevent inaccuracies, which is preventing more widespread adoption of AI tools by enterprises. And enterprise companies should keep an eye on them.
What it does: API that enables AI agents to contact humans for help and approval.
Why it is a fave: AI agents can make a big difference when it comes to productivity — if they are working as intended. Having humans in the feedback loop helps prevent AI agents from going off the rails, but too much human oversight can slow down processes and diminish the efficiencies these AI agents are supposed to bring. HumanLayer seems like a nice happy medium; it brings in human oversight just when it’s needed and doesn’t require it when it is not.
What it does: Research agent for enterprise sales.
Why it is a fave: This is the first enterprise sales lead gen software I’ve had reason to get excited about (sorry). Raycaster’s approach is to find very specific details on a potential sales target, like what lab equipment the company uses or what the company’s CTO discussed at a recent conference, to pitch them at the right time and in the right way. This stands out among a wave of lead gen startups that seem to still be focused on just aggregating surface-level information.
What it does: Compliance guardrails for AI applications.
Why it is a fave: Galini gives enterprises a tool that makes it easier to set up AI guardrails based on both company policies and regulations for their AI applications. Plus, putting these controls in the hands of enterprises gives them more freedom and allows them to evaluate how effective the guardrails are.
What it does: AI tool set that helps enterprise customers manage hallucinations.
Why it is a fave: AI hallucinations are a big problem without an easy fix. While CTGT can’t prevent all hallucinations, its approach of actively monitoring and auditing an enterprise’s models, allowing it to better spot abnormalities and potential hallucinations, seems like a nice upgrade to the other options out there. The fact that the company is already testing its tech with Fortune 10 companies is also a good sign that potential customers are looking for a tool like this.