Sprint Retrospective
What Went Well
- Prototype demonstration was clear and functional – The customer understood the current behaviour and was able to give actionable feedback.
- Valuable technical guidance from the customer – The customer provided concrete next steps (diff-first workflow, context packaging, token size limits, backlog grooming).
- Team alignment improved – The meeting clarified responsibilities and the logical order of development steps (diff → context → LLM experiments).
Problems Encountered & Root Causes
1. Unclear approach to handling diffs vs full files
- Root cause: Assumptions were made without experimentation; team hadn’t validated LLM behaviour on diffs vs full-file context.
- Impact: Development stalled because it was unclear how to design prompts and context packaging.
2. Backlog lacked grooming and coherent prioritisation
- Root cause: Story points were assigned intuitively without relative sizing; the backlog wasn’t sorted by complexity or readiness.
- Impact: Harder to select appropriate tasks for the sprint; long-term tasks were mixed with near-term ones.
Changes for Next Sprint (Prioritised)
High Impact
- Perform hands-on experiments with diffs and full files using LLMs
- Compare outputs on 2–3 test cases.
- Establish what “good summary” means using real examples.
Medium Impact
-
Define the initial CLI command structure
-
Draft commands such as
cli changes-since,cli ask,cli diff. -
Set an internal token usage limit and warning system
- Apply a temporary personal token or free-tier LLM with mindful usage.
- Add a limit (e.g., 10k–20k tokens).
Low Impact
- Introduce backlog grooming sessions before sprint planning
- Sort tasks by clarity and size (small → medium → large).
- Apply relative story points only after grouping.