The problem: assignment by loudest voice or habit
New work lands on whoever answered Slack first — or whoever carried the last release. Critical paths cluster on two seniors. Juniors wait for "easy" tickets. PMs ask "who can take this?" in a channel and get polite yeses that ignore real load.
AI task distribution suggestions should reduce that friction — not replace managers with an auto-dispatcher that ignores skills, growth goals, and blockers.
Why naive auto-assign fails
Skill mismatch — Available hours ≠ ability to release production infra.
Hidden overload — Open-task counts miss cross-project work.
Hero reinforcement — Auto-routing urgent work to last resolver burns your best people.
No audit — Silent reassignment breaks trust and traceability.
Good distribution combines capacity signals, skill context, and explicit human approval.
Framework: triage → capacity → suggest → assign
1. Triage — Priority, due hint, dependencies, required role (backend, design, QA).
2. Capacity — workload_heatmap remaining hours per candidate; portfolio_overview for matrixed staff.
3. Suggest — AI lists two or three people with headroom and relevant open context — citations, not gut feel.
4. Assign — Manager confirms; assign_task via UI or MCP; comment on task why owner changed.
| Input | Source in WKFGo |
|---|---|
| Who is overloaded | workload_heatmap |
| What blocks the task | get_task, dependencies |
| Cross-project load | portfolio_overview |
| Time calibration | time_summary (estimates vs logged) |
Where WKFGo fits: suggest, do not silently dispatch
workload_heatmap as suggestion input
Before sprint planning or intake triage, scan heatmap:
- Red cells → do not suggest without trade-off
- Green cells → candidates for new work
- Amber → pair or split task
MCP prompt: "For task 1842, list team members with lowest remaining hours this week on Project Atlas who have done similar labels — cite task IDs."
assign_task with accountability
WKFGo supports assignment via UI and MCP assign_task. Recommended policy:
- AI proposes owner in draft comment
- PM or tech lead approves
- Assignment logged in task history
Stage 4 on the MCP adoption ladder — after quarters of read-only Q&A.
Sprint and package context
list_packages, sprint objects, and release trains add horizon — do not assign a large epic to someone already committed to release cutoff.
Practical rituals
Intake triage twice weekly — New items get owner or explicit "unassigned + reason."
Rebalance standup — One red cell addressed per day beats heroic weekend.
Post-assign comment — "Assigned to Sam — headroom plus prior API work on #1201."
Retro on misfires — Wrong assignee? Fix estimates or skill tags, not blame.
Getting started this week
At intake triage, run heatmap before assigning any unowned task. For each assignee candidate, verify label history with list_tasks — not just lowest hours. Post assignment comment citing why. Review misfires in retro weekly. Stay on MCP stage 2–3 (suggest + human assign) until audit culture is solid.
Release week: freeze casual reassignments — only trade-offs logged in decision_inbox so history stays explainable.
Capacity bands that match reality
Utilization at 100% is not sustainable — meetings, sick leave, and context-switch belong in the denominator. Many IC teams plan at 70–85% of calendar hours for committed delivery work. When AI suggests assignees, apply that band: green is not "zero work" — it is healthy headroom for support and review. time_summary from prior sprints calibrates whether estimates assume fantasy capacity.
Anti-patterns
- Fully automated assign on production projects without review
- Suggesting purely by lowest hour count ignoring skill
- Ignoring PTO and calendar blocks in capacity
- Assignment changes without task history comment
Auto-dispatch without human review erodes trust. Use AI to rank assignee candidates by open load and skill tags visible in the project—then let the team lead confirm. workload_heatmap and open task counts ground suggestions; never reassign silently on model confidence alone.
Run a monthly audit: do AI answers cite task IDs from MCP? If not, tighten prompts to require citations before any stakeholder-facing draft leaves the team.
Permission-aware tools mean the same question gets different grounded answers for PM vs contractor—that is a feature, not a bug. Train teams to expect scoped truth.
FAQ
Does WKFGo auto-assign with AI out of the box?
No autopilot product — use heatmap plus MCP/chatbot to suggest; humans execute assign_task.
Can AI balance across projects?
portfolio_overview helps; matrix managers must still judge strategic priority.
What about team fairness?
Rotate visible opportunities; heatmap prevents silent overload on willing volunteers.
Bilingual teams?
Assignment comments can follow team language; data is locale-neutral.
Suggest from data, assign with judgment
Use AI and workload_heatmap to suggest task distribution — then assign deliberately with skill context and audit trail.
Try it now
Put these patterns on live project data—not slide decks.