LLMs are already being used in all kinds of ways, but new and interesting use-cases keep popping up.
In a recent post, legendary programmer and tech entrepreneur John Carmack floated an intriguing idea: what if the AI assistants we chat with daily could serve as job references? Carmack’s premise is straightforward but potentially transformative. While social media algorithms learn what content users prefer, conversational AI systems can learn how people think through extended dialogue. Over months or years of interaction, an LLM builds up a detailed understanding of how someone approaches problems, communicates ideas, and reasons through challenges—essentially conducting what Carmack describes as an “extended interview.”
The proposal would allow job candidates to offer their LLM chat history as a reference, similar to providing contact information for a former manager or professor. The AI could then form an assessment and represent the candidate without exposing private information. Carmack envisions hiring processes where one company’s LLM conducts an in-depth conversation with a candidate’s personal LLM, dramatically increasing the signal quality beyond what traditional resume screening provides.
Beyond Traditional Screening
For Carmack, this isn’t about introducing AI judgment into hiring—it’s about improving what already exists. He notes that LLMs already evaluate candidates in many interview processes today, working from the limited data in resumes. His argument is that richer data sources benefit everyone except those misrepresenting their qualifications, and questions whether AI assessment is truly worse than evaluation by what he terms “random HR people.”
The tech pioneer acknowledges that some candidates already generate strong signals through public work—open source contributions, academic papers, long-form writing, or active social media presence. However, he points out that most talented professionals lack significant public footprints. Even the resource-intensive interview processes used by major tech companies aren’t as predictive as organizations would like, making multi-year chat histories a potentially valuable signal.
A Two-Sided Market Opportunity
Carmack takes the concept further, imagining systems that could answer questions like “What are the best candidates in the entire world that we should try to recruit for this task?” He frames this as unlocking enormous economic value by optimizing person-job fit—emphasizing that the benefits flow in both directions, helping both employers find the right talent and employees discover better-matched opportunities.
The idea raises obvious questions about privacy and consent in AI assessments. It also assumes that how someone interacts with an AI assistant accurately reflects their professional capabilities—a premise that would need validation. Also, like other metrics, this one could be gamed as well — people could create fake smart personas while dealing with LLMs to trip up recruiters. Still, as organizations continue seeking better ways to identify talent and candidates look for methods to stand out in crowded fields, Carmack’s suggestion represents the kind of unconventional thinking that could reshape hiring practices in an AI-driven future. Whether LLM-based references become reality or remain a thought experiment, the proposal highlights how conversational AI systems are accumulating data that goes far beyond simple task completion—and how that data might eventually be leveraged in ways we’re only beginning to imagine.