← agentic systems
Agentic · Parallel · Cookbook 12 The Python 12_parallel_due_diligence.py

Four specialists, one verdict

A new supplier wants a contract. Four specialist AI agents could review it: security, privacy, legal and finance. Running all four on a stationery supplier is a waste of money, and skipping the wrong one on a patient-records vendor is how a breach starts. Jev decides which specialists this supplier needs, the chosen ones work at the same time, and a second Jev call turns what they wrote into one decision.

Jev: plan, then merge Specialist LLM agent (runs in parallel) Rule in code: contract value An issue nobody looked for

This supplier

Press Play.

    So far

    Specialists run
    0
    of 0 possible
    Specialist cost
    $0.00
    frontier calls
    Wall time
    0s
    in parallel
    One after another
    0s
    same specialists
    Issues missed
    0
    specialist not run
    Wrong verdicts
    0
    vs running all four

    Switch between Jev picks and Run all four and compare the cost. Then raise the selection bar and see when skipping a specialist starts missing real issues.

    The code · lines this supplier ran are highlighted

    
    
    
    
    
      
    Every supplier so far · at the current settings
    
      
    
        
    SupplierRanVerdict
    
        
      
    
    
    
  
    

  
    The numbers are illustrative, not measured: about $0.09 and 15–35s per specialist,
  0.25s per Jev call. ← 11 · Three desks, one claim ·
  Next: 13 · The foreman →