Principal (Agentic) Investigator
Developed by the Poggio Lab team at MIT
We're building pAI, a multi-agent AI system that turns research hypotheses into high-quality, literature-grounded, experiment-backed papers with minimal human steering.
Current AI systems require on the order of 102 to 103 prompts to go from a research idea to a written paper. pAI aims to reduce this to up to 10 human interactions.
This is not about fully automating research or replacing human creativity. pAI focuses on the rigorous establishment of ideas: matching theory with experiments, distinguishing correlation from causation, proposing parallel explanations and testing them. The human provides the idea. The system does the structured work of turning it into a solid manuscript.
Cost: ~$2-10
Time: 15-40 min
Requires: one API key
# 1. Bootstrap environment (one-time, ~5 min)
./scripts/bootstrap.sh researchlab minimal
conda activate researchlab
# 2. Set your API key (only one provider needed)
cp .env.example .env
echo "ANTHROPIC_API_KEY=your_key_here" >> .env
# 3. Validate setup without spending tokens
python launch_multiagent.py --task "test" --dry-run
# 4. Run the included quickstart example (~$3)
python launch_multiagent.py \
--task "$(cat examples/quickstart/task.txt)" \
--output-format markdown \
--no-counsel \
--no-log-to-filesRecent notes, releases, and research updates from the team.
Sun Jun 21, 2026
We released a revised preprint of Distribution-Aware Algorithm Design with LLM Agents, joint work by Saharsh Koganti, Priyadarsi Mishra, Pierfrancesco Beneventano, and Tomer Galanti. The paper studies...
Read post
Mon Jun 8, 2026
Tomer Galanti has written a clear explainer of Distribution-Aware Programming, the blog post accompanying the paper Distribution-Aware Algorithm Design with LLM Agents. The post explains the central i...
Read post