Skip to lesson content
Hromium
Course progress: 67

Map the system first

The team is moving order status updates from direct writes to a transactional outbox. The change touches 50+ write sites while the old and new processing paths must coexist.

lesson goalRecognize when planning mode must come before the first edit: map dependencies and agree on a safe change sequence.

12 minutes

What to change is clear. The order is not.

This is an end-to-end change: HTTP handlers, background jobs, and retries are connected, while an immediate status update becomes asynchronous event processing.

Rewiring a working hospital

The hospital cannot close, so the old and new grids switch over in stages. The engineer maps the circuits and agrees on the sequence first—just as planning mode maps dependencies before code changes.

electrical grid
order system
switchover plan
stages, checks, and rollback

How should the migration begin?

The old and new processing paths must run in parallel; the system cannot stop.

What must the team do before the first edit?

Map dependencies and plan first. Then edit code.

Choose the correct answer or reveal it to see the comparison.

When to use planning mode

To mark this lesson complete, choose the correct answer or reveal it.

source: Lecture 01 — Moving order processing to events

All course lectures

Find a lecture by title or theme and open it.

Agentic architecture

18

  1. Preserving the Evidence Link in a Risk Register
  2. Guaranteeing the Outcome of Request Handling
  3. Choosing the Next Action After a Device Inspection
  4. A Fixed Review Process for Grant Applications
  5. The Coordinator Agent Never Launches Its Reviewers
  6. An Independent Comparison of Architectural Approaches
  7. Formatting a Mixed Operational Report
  8. Parallel Analysis of Independent Studies
  9. Adaptive Investigation of Data Loss
  10. Closing Gaps Before an Investment Memo
  11. A Fast Answer from Context Already at Hand
  12. Synthesizing Divergent Research Findings
  13. Adaptive Investigation of a Build Failure
  14. Handing a Request Off to a Specialist
  15. Sourced Context for the Final Opinion
  16. Handing Results to the Synthesis Agent
  17. Tracking the Progress of a Multi-Step Task
  18. Resuming a Subagent Through SendMessage

Tools & MCP

12

  1. Preventing an Invented Report Identifier
  2. Finding Mentions of a Deprecated Flag
  3. Handing Off Context When Escalating a Booking
  4. Access to the Corporate Data Catalog
  5. Telling Seat-Reservation Errors Apart
  6. Launching an MCP Research Template
  7. Simplifying an Overlapping Tool Set
  8. Mandatory Approval for a Large Discount
  9. Surveying What MCP Servers Expose Before Acting
  10. Project and Local MCP Scopes
  11. A Direct Request for a Specialist
  12. Deferred Loading of MCP Tools

Claude Code

23

  1. Migrating Order Processing to Events
  2. Editing One of Several Repeated Configuration Blocks
  3. A Repeatable Release-Readiness Review
  4. Choosing a Background Processing Mechanism
  5. Limiting the Length of a Display Name
  6. Instructions Scoped to Parts of a Monorepo
  7. Resuming a Named Session
  8. Resuming Work After a Small Code Change
  9. Understanding a Message Retry Mechanism
  10. Tracking Down the Source of an Unfamiliar Warning
  11. Planning Observability in an Unfamiliar Codebase
  12. Targeted Handoff of Reference Files for a One-Off Integration
  13. Fixing an Importer from a Reproducible Example
  14. A Team-Wide SDK Refresh Command
  15. Attaching Guides to Firmware Packages
  16. When an Explicit Multi-Stage Workflow Is Needed
  17. Refining Printed Labels Against a Reference
  18. Separating a Prohibition, a Recommendation and an Automatic Action
  19. Iterative Development of a Scheduler
  20. Restoring Trust in Procurement Checks
  21. A Targeted Investigation Guided by the Trace
  22. Parallel Subagents and the Worktree Branching Base
  23. Symbol References versus Text Search

Context & reliability

14

  1. Returning to a Booking with Stale Data
  2. Context Drift During an Incident Investigation
  3. Preventing Repeat Findings in a Documentation Review
  4. Fast Review of Automated Check Results
  5. The Review Criterion for Operating Instructions
  6. Compressing Data Before Risk Prioritization
  7. Uniform Recommendations for the Support Desk
  8. Reducing Bulky Tool Results
  9. A Failure While Issuing a Temporary Badge
  10. Keeping State for Several Concurrent Cases
  11. Reconciling Metrics Before and After a Migration
  12. Moving from Feature Flags to Analytics Attribution
  13. The Session Storage Fork
  14. A Recurring Check Inside a Session