Managing the Work
Managing Research Software Projects
Greg Wilson
2026
Learning Objectives
- Explain what “project management” looks like for a small research
team.
- Describe the progress/plans/problems format for status reports.
- Explain how to manage a work backlog and when and how to cut items
from it.
Knowing Who Is Doing What
- Project management at this scale is knowing who does what and what
blocks them.
- The manager’s job is to remove obstacles, not assign tasks.
- The Monday morning check-in reveals when someone is stuck.
- The planning meeting matters less than the check-in.
The Three-Column Status Update
- Progress: what they finished since last time.
- Plans: what they will finish before next time.
- Problems: what is blocking them, or might.
- The meeting discusses Problems; Progress and Plans are read.
- If the leader cannot resolve a problem, they must escalate it.
Making Issues Work
- Each issue needs an owner, a milestone, and a time estimate.
- Someday/maybe items get a backlog label, not a milestone slot.
- Close issues when the work is done, not when the PR merges.
- Research artifacts belong in the same tracker.
- Triage newly filed issues at least once a week.
Cutting
- Cutting scope feels like failure but deferring the cut costs
more.
- If a feature is 80% done and blocking a release, options are delay,
ship incomplete, or cut.
- Deciding to cut two weeks late is always more expensive.
Estimation
- You can’t estimate accurately unless you collect data to tell you
how accurate past estimates were.
- Which most teams aren’t going to do.
- The smaller the task, the smaller the relative error.
- So break things down in order to make estimation more
rewarding.
- Distinguish a wrong estimate from a choice to solve a related
problem.
- The second kind of change should trigger a conversation.
What about AI?
-
LLM sprint tools can estimate complexity and flag hidden dependencies
-
LLM task tools optimize for issues closed, not correctness or
maintainability
-
LLM prioritization nudges teams toward many small issues over deep
problems
-
Cutting is harder because saying no is not automated
Write a Status Update (8 min)
Write one three-column status update for yourself, then swap with a
partner. Read their Problems column and identify one thing you could do
to help.
Set Up a Milestone (7 min)
Prompt the LLM with something like this:
Given these open issues in my project [paste 5-8 real issue titles],
help me create a four-week milestone. Assign each to either the
milestone or a backlog. For each milestone issue, estimate the effort in
hours and identify dependencies.
Which assignments did the LLM get wrong?