Architecture in practice
System design lessons, AI perspectives, and questions worth taking into your next interview or design review. Read the reasoning, then apply it to your own constraints.
An agent submits an approved refund, then loses the response. Design the retry, recovery, and audit trail without turning one customer request into two refunds. Includes a worked answer.
Rohit Bhardwaj
Before an AI agent takes action, decide what it may do, what evidence you will keep, and how you will recover when the outcome is uncertain. A practical review for your next design.
Rohit Bhardwaj
Design recovery for a refund or booking whose outcome is unknown. Separate a retry from a new business action and reconcile before guessing.
Rohit Bhardwaj
Use hard constraints, a small evidence table, and a reconsideration trigger to make a design choice that another engineer can examine.
Rohit Bhardwaj
Specify what an agent may prepare, what authorizes execution, and what happens when approval or verification is unavailable. Includes a printable review worksheet.
Rohit Bhardwaj
Separate relevance from applicability when retrieved policies conflict. Make authority, scope, effective dates, and uncertainty visible.
Rohit Bhardwaj
Compare two hypothetical AI workflows using the same quality bar, full run cost, and verified task outcomes. Includes a worked calculation.
Rohit Bhardwaj
Turn an ambiguous booking request into explicit outcomes, constraints, and a first decision. Includes a printable Design Decision Worksheet.
Rohit Bhardwaj
The AI Architecture Brief
Get practical architecture lessons, resources, and live-session updates. Choose the updates you want when you join.
Get the Architecture Brief →