AI task assignment that picks the cheapest skilled person, in natural language
Most assignment decisions happen in standup with the manager guessing. Infersync's AI COO ranks every candidate by skill fit, remaining capacity, and hourly cost, then lets you bulk-assign 100 issues in one natural-language directive. Every action is previewed before execution so you can see the reasoning before paying for it.
Why teams end up here
- The manager makes 30+ assignment decisions per sprint in standup, all from working memory rather than capacity + cost data.
- P1 issues get routed to whoever's most familiar, not whoever's actually free this week or cheapest for the task.
- Bulk-relabeling, bulk-due-date changes, bulk-status moves all happen by clicking through GitHub or Jira one item at a time.
- When an AI feature exists, it's text generation (drafts, summaries) not action execution. The AI describes what should happen; the human still does the clicking.
- AI tools that DO take action are opaque about WHY — there's no preview step where you see the AI's reasoning before it mutates production data.
Preview-then-execute: see the AI's plan before it runs
Type what you want in plain English. The AI COO ranks candidates (skill_fit × availability × cost_inverse), surfaces the proposed plan as a numbered list, and waits for your approval before mutating anything. Operations-tier users can execute; Base-tier users see the preview surface for free as a way to test the AI's reasoning before paying for the action layer.
Natural-language directives
Type 'assign all OAuth issues to Jane, label them P1, due Friday'. The AI parses the directive into discrete actions and shows you each one before applying.
Skill + cost + availability ranking
Candidates are scored on skill match for the task domain (Node, Postgres, etc.), remaining capacity in the next 7 days, and inverse hourly cost. Top 3 surfaced, runner-up explained.
Budget gates baked into ranking
Over-budget candidates are auto-excluded from the ranked list. The cost lens isn't a separate dashboard, it's wired into the assignment logic.
Bulk operations up to 100 items per call
One directive can assign, label, set due dates, change state, and set priority across up to 100 work items in parallel (5 concurrent operations, rate-limited to GitHub's secondary limit budget).
Preview vs execute split (Base vs Operations)
Base users see the AI's proposed plan for free as a daily quota (10 commands per day). Operations users get unlimited execution. The split lets you test the AI's reasoning before paying for the action layer.
Two-way GitHub write-through
Assignments, labels, status changes write through to github.com via webhook. Comments on github.com flow back into Infersync. No mirror, no fork.
Steps to get from zero to live
- 1
Connect GitHub + LLM key
GitHub OAuth in 30 seconds. Bring your own Anthropic / OpenAI / Google API key (Settings → AI). Zero token markup from us.
- 2
Type your first directive
Open the command bar at the bottom of the dashboard. Type something like 'assign all open OAuth issues to the cheapest available Node engineer'.
- 3
Review the preview card
The AI returns a plan card: intent, ranked candidates, the proposed bulk action, and the budget impact. Approve, edit, or cancel.
- 4
Execute and watch the writes land
Approved actions execute in parallel, write back to GitHub via webhook, and are audit-logged. Each item's outcome reports back to the bar.
Available on the Operations plan and up
The AI COO preview surface is available on Base (£7/seat/month) with a 10-command-per-day quota. Action execution (assign, label, due dates, status, priority) and unlimited daily quota live on Operations (£15/seat/month). All AI runs through your own LLM API key on every tier — Infersync never marks up token costs.
FAQs about ai task assignment that picks the cheapest skilled person, in natural language
What's the preview-then-execute split, exactly?
When you type a directive, the AI first returns a 'plan preview' card with the proposed actions listed step by step. You review and approve before anything mutates. Base-tier users can issue 10 commands per day in preview-only mode (they see the plan but can't execute); Operations users can execute unlimited commands. The split is intentional: it lets anyone test the AI's reasoning before paying for the action layer.
How does the ranking work?
score = skill_fit × availability × (1 / cost_per_hour). Skill fit comes from each member's tracked skills with proficiency levels (5-level model, reversible audit history). Availability is capacity remaining in the next 7 days (which respects leave). Cost is the member's hourly rate from workspace settings. The top 3 are surfaced; the runner-up is shown with a sentence explaining why they came second.
What if the AI gets the assignment wrong?
The preview step is the safety net. You see the proposed assignments before they execute, with the reasoning. You can edit the directive and re-preview, or override individual rows in the preview card. Once executed, every action is audit-logged so you can see who-assigned-what-to-whom and roll back if needed.
Can I do bulk operations across hundreds of issues?
Up to 100 items per single call (with 5 concurrent operations), capped to stay inside GitHub's secondary rate limit budget. For a 500-issue bulk operation, you'd run it as five 100-item calls. The cap is there to keep AI token cost bounded per directive and to protect the GitHub rate limit headroom for the worst case.
Does this work for assignment beyond issues — like due dates, labels, priority?
Yes. The supported actions: assign, unassign, add label, remove label, set due date, clear due date, close, reopen, set priority. All work in single-item or bulk mode through the same natural-language interface. State changes write through to GitHub via webhook automatically.
What LLMs does this support?
Bring your own Anthropic, OpenAI, or Google API key. The provider is configured per workspace in Settings → AI. Local Ollama is supported in dev mode only (blocked in production for reachability reasons). No token markup from Infersync — you pay your existing API bill, we charge a flat per-seat fee for the platform.
See also: Engineering cost tracking • AI task assignment • GitHub time tracking • Capacity planning.