Slide contents

1. 1. A factory on subscription

   Commercial products and an affordable pilot choice

2. 2. Buy a process, not a promise

   Four products, access conditions and one bounded pilot

3. 3. Names hide different units of work

   Compare input, coordination and output

4. 4. A mission separates work and verification

   A behavioural contract precedes the feature list

5. 5. Validation needs a runnable application

   A scriptable user journey is an entry condition

6. 6. Auto-Triage starts with a signal

   Auto-Triage: investigation, routing or repair

7. 7. Events alone do not form a factory

   Cursor Automations launch Cloud Agents

8. 8. 8090 preserves a chain of decisions

   Requirement, architecture, work order, implementation

9. 9. Product output defines the responsibility boundary

   Code, PR, acceptance and release are distinct states

10. 10. Cheap entry does not guarantee cheap tasks

    Official terms snapshot: 27 October 2026

11. 11. Missions require Extra Usage enabled

    A $20 plan + at least $10 credits gives a $30 entry estimate

12. 12. Two pages disagree about cloud computers

    Managed Droid Computers: Pro in docs, Plus on pricing

13. 13. Build the budget by cost source

    Illustrative arithmetic; this is not an actual bill

14. 14. Design access before connecting tools

    A teaching repository and synthetic data

15. 15. Without history, outcomes are unexplained

    Plan, diff, checks, interventions and spending

16. 16. Choose Factory for the teaching task

    Workflow fit matters more than an overall ranking

17. 17. Freeze the pilot specification before launch

    Matched task, starting revision and evaluator

18. 18. Choose fallback conditions in advance

    Feature access, environment, observability and budget

19. 19. Documentation does not replace our own run

    Separate claims, entitlement, execution and results

20. 20. The output is a testable pilot plan

    Next, run a bounded process and preserve evidence