From code to chemistry: using Kiro to tackle ADME-Tox, a key drug discovery challenge

This article describes building a production-ready ADME-Tox prediction agent for drug discovery using Kiro, AWS's agentic AI IDE, with a small team in one month.

• Small team (2 developers, 1 scientist) built ADME-Tox agent in ~60 combined hours
• ADME-Tox assessment eliminates ~90% of drug candidates before clinical trials
• System integrates open-source tools: ADMET_AI, RDKit, ChemProp, DeepChem, PubMed APIs
• Multi-agent architecture: ADME prediction, toxicity assessment, literature mining agents
• Tested on drug analogs; delivered comparative safety profiles in 1-2 minutes
• Kiro spec-driven development collapsed months of architecture work into days
• Results complement early-stage decision-making but don't replace in vitro testing

The project demonstrates Kiro's effectiveness for rapid agentic AI development in life sciences, enabling small teams to build production-ready solutions quickly without sacrificing quality.