Formations · Accelerated · Summer 2026
Research Lab
6 companies in the last four YC batches (Fall 2025 – Summer 2026), up from 2 in the four before.
0.8% of each new batch, up from 0.4%.
The label "Research Lab" has quietly shifted from a place that produces research to a product that automates it: Antinuous, Aster, and Ndea all center on generating discovery itself, while hillclimb sells the training data needed to make models recursively self-improve. The arXiv "AI Scientist Agents" cluster is still small at 107 papers but gaining 8 recent entries, suggesting academic groundwork is only now catching up to what these startups already sell. That follows directly from 2024-2025 reasoning models becoming reliable enough at multi-step tool use to attempt unsupervised experimentation.
Written by a model from the companies below.
Companies 6
Newest batch first.
- Induction LabsBuilding intellectually curious AIS26
- AntinuousAutonomous research agents for R&D and applied scienceSp26
- AsterAutomating open-ended research.Sp26
- NdeaBuilding AGI that can innovate.W26
- hillclimbTraining Data for Recursive Self-ImprovementF25
- Veria LabsContinuous AI pentesting that finds and fixes vulnerabilitiesF25
How sure: 6 of 774 companies in the recent batches, against 2 of 564 before. A group is only listed when its share rises by at least 35% with at least 6 companies.
How this is measured
The category comes from Description cluster · description-only. Counts compare the share of new YC companies in the latest four completed batches with the four before. First surfaced Summer 2026; observed October 24, 2026.
The summary are model-written (claude-sonnet-5, 2026-10-24) from the companies listed here. Any line on why this is happening now comes from the model’s general knowledge, not this dataset.