🧠 Food for thought

Implications, context, and why it matters.

NeoCognition’s ‘expert’ agents aim to learn jobs like humans do

• NeoCognition says its agents keep learning how a company works, including processes, limits, and routines, to build a specialized “world model of work” for each setting 1.
• The company says this focus can make them quicker and more dependable than general-purpose models. It also says the approach can cut costs and improve safety in high-stakes business use 1.
• NeoCognition ties its case to the founders’ work on Mind2Web, MMMU, and SeeAct, which cover AI benchmarks and agent studies. It also says work from Su’s team is used in frontier large language models from OpenAI, Anthropic, and Google 1.

Investors are also backing smaller, task-specific AI approaches

• NeoCognition’s funding fits a wider move toward smaller AI models built for one job, which investors believe can beat general-purpose systems in some tasks 2.
• Fastino says it trains small models for specific tasks on low-end gaming graphics processing units (GPUs). Yoodli, a communication coaching startup, reached a valuation above US$300 million after a US$40 million series B round by focusing its AI on communication training 23.
• The shift suggests companies buying AI care more about dependable performance and lower costs when a system is built for one purpose 2.