How to build agents for scientific discovery in 2026
Andrew White
FutureHouse, Edison Scientific
Unlock 2026
April 2026
AI Progress
Model intelligence doubles every 7 months
METR Task Completion Benchmark metr.org
Effect is Visible in Economy
1 NASA OIG Report oig.nasa.gov 2 US FHWA fhwa.dot.gov 3 US Telecom Capex Report ustelecom.org 4 Morgan Stanley AI Market Trends 2026
Edison Scientific
- Founded in 2025
- Spinout from FutureHouse
- Based in San Francisco with Wetlab
- Raised $70M in December
Our Timeline
Edison Platform
Edison makes a fully-portable AI Scientist platform
Placeholder for platform overview
| full-text papers | 175M |
| patents | 120M |
| queries since launch | >2M |
| users | 65k |
| coverage of existing biology data | 80% |
| API access | 50 RPM |
| internally deployable | Portable |
Literature
Data Analysis
Evaluation: Can it reproduce papers?
... Calculate Spearman correlations of the resulting log-fold change (logFC) values across conditions. Perform hierarchical clustering. Plot and visualize the clustering result as a heatmap to show how different ASD forms cluster together as development progresses
Creates a reproducible R notebook
Side-by-Side
Side-by-Side
Kosmos: An AI Scientist for Autonomous Discovery
Ludovico Mitchener, Angela Yiu, Benjamin Chang, Mathieu Bourdenx, Tyler Nadolski, Arvis Sulovari,
Eric C Landsness, Daniel L Barabasi, Siddharth Narayanan, Nicky Evans, Shriya Reddy, Martha Foiani, Aizad Kamal,
Leah P Shriver, Fang Cao, Asmamaw T Wassie, Jon M Laurent, Edwin Melville-Green, Mayk Caldas, Albert Bou,
Kaleigh F Roberts, Sladjana Zagorac, Timothy C Orr, Miranda E Orr, Kevin J Zwezdaryk, Ali E Ghareeb, Laurie
McCoy, Bruna Gomes, Euan A Ashley, Karen E Duff, Tonio Buonassisi, Tom Rainforth, Randall J Bateman, Michael
Skarlinski, Samuel G Rodriques, Michaela M Hinks, Andrew D White
arXiv:2511.02824, 2025
How do you validate systems like this?
Work with external groups. Input is their experimental data. Three discoveries reproduced in unpublished work. Four novel discoveries.
Entorhinal Cortex
Kosmos overview
independent expert annotation of task difficulty and correctness
Agents 2026
- Since early 2022, agents have mostly been stagnant
- Call tools, keep conversation history, use special prompts
- The outer scaffolding changed less than the demos suggest
Pure-Code Agents
- No special tools: new integrations are mostly just code to write
- No MCP-style context bloat: discover code like an unknown codebase
- No special harness: better code-writing agents get better at your tasks
Demo vs Production
- There is still a huge gulf between demo and production
- Sandboxed execution changes what is possible and safe
- Secrets management and filesystem boundaries become product problems
OpenClaw
- OpenClaw showed a different form factor
- A colleague that works independently, not just an assistant
- More autonomous, more asynchronous, and more accountable
Project Aries
- Kosmos world model plus a coding agent
- Persistent colleague-like interface
- Benefits of a coding agent applied to scientific work
Example
Metabolite prediction model
Find the Gloryx paper, build a bigger dataset, re-implement their model, then iteratively improve until better. Make a demo.
Metabolism Model Demo
Reaction Expansion
Ranked Metabolites
Beta Launches Tomorrow
