Scenario

A lead is potent and clean, but can it be made at scale? A molecule that needs a long, low-yielding, or exotic-reagent route is a program risk. The team wants an early synthesizability read so scale-up feasibility informs which analog to nominate.

Worked example. Two equipotent analogs differ sharply in route complexity; the program should nominate the one that scales.

Agent workflow

1. Read the candidate structure(s).
2. Synthetic accessibility screen — score synthesizability, estimate route length, and flag hard steps or uncommon building blocks.
3. Reactor kinetics screen — for the nominated route's key step, model conversion vs. time and temperature (a skill shared with the engineering Scale a Batch Reaction to Pilot use case).
4. Thermophysical property lookup — pull the physical-property data needed to sanity-check heat release and solvent behaviour at larger scale.
5. Compare candidates on a combined potency-and-makeability basis.

Demo output

A synthesizability report per candidate: accessibility score, estimated route length, and the specific steps or reagents that drive risk. The magic moment: a slightly less potent analog wins nomination because it is dramatically cheaper and safer to scale.

Deterministic vs LLM

• Deterministic — the synthetic-accessibility scoring and complexity analysis.
• LLM — the route-risk narrative and the scale-up recommendation.

Every score is a validated calculation. The agent reads and routes; the engine decides.

Limits

Screening-level makeability guidance, not a validated process route. It flags scale-up risk early; process chemistry designs the real synthesis.