AT&T Labs Research
Before agents were Kevin's job, AT&T Labs was where he first built them: Mixture-of-Experts agents that cut enterprise document analysis time by 85 percent.
In Fall 2023, Kevin was an AI research intern at AT&T Labs Research in Middletown, NJ, working on NLP and intelligent agents. Source: Resume, Kevin B. Liu He designed autonomous agents with Mixture-of-Experts LLMs via Gentopia to parse enterprise documents, reducing analysis time by 85 percent for an 11x cost savings, then piloted the tool on AT&T enterprise data while iterating with data scientists and PMs across departments. Source: Resume, Kevin B. Liu
This is the earliest agents-research entry in Kevin's history, and it reads as the seed of everything later. Career Profile keeps the AT&T row as a core differentiator: MoE agents reducing doc analysis, published research. Source: Career Profile The lineage runs straight to the present. His current research with Professor Danqi Chen on agent orchestration, and specifically CoTCodec: Language Choice as an Orchestration Variable for Tool-Using LLM Agents on language as an orchestration variable, is the same question matured: the language model is one module, and the advantage comes from how you orchestrate it. Source: Resume, Kevin B. Liu; Kevin Liu
It also rhymes with World Models, Kevin's view that an LLM is one subsystem of intelligence, not the whole thing. The MoE framing, many specialized experts routed by a coordinator, is an early version of treating the model as a part rather than the whole. The internship was research, not product, but it set the through-line.
The research instinct predates the internship and outlasted it. Kevin already held two publications by 2023, in IEEE and IYRC venues, and the AT&T agents work became one more proof that he could do research that ships into a real enterprise pilot, not only research that publishes. Source: Resume, Kevin B. Liu; Career Profile That bridge from inquiry to working system is the same one What Is Engineering draws between research and engineering.
The pilot mattered as much as the model. Iterating with data scientists and PMs across multiple AT&T departments taught Kevin early that research only counts when someone downstream can use it, the same instinct that later shows up in the Forward Deployed habit of turning what you learn into shared infrastructure. Source: Resume, Kevin B. Liu
Timeline
- 2023-Fall | AI research internship at AT&T Labs Research; MoE agents via Gentopia, 85 percent reduction in document analysis time. Source: Resume, Kevin B. Liu
- 2026-Fall onward | Research lineage continues with Professor Danqi Chen at Princeton on agent orchestration and CoTCodec. Source: Resume, Kevin B. Liu
- 2026-05-31 | Story page synthesized from resume, career profile, and research pages. Source: compiled from wiki graph