AI-Powered Task Prioritization: How Jibun Kaisha's WBS System Works

The Real Problem with Task Management

Notion, Trello, Asana, Linear, Jira — the tool list keeps growing. The problem isn't the tools. It's that organizing tasks has become its own task: entering, categorizing, prioritizing, updating progress. That overhead alone eats 30 minutes a day.

Jibun Kaisha's WBS system addresses this by making AI responsible for the organizational overhead — not the human.

WBS Design Philosophy

Six-Department Balance

Jibun Kaisha operates as a six-department company:

Tasks are tagged to one of these categories. The system surfaces imbalance — if 80% of tasks are Product and nothing is HR, that's a signal.

Instance-Based Task Distribution

Ten AI instances (VSCode/Windows/PowerShell #1–6/Web/Mobile) each have assigned tasks:

instance: ps2
 → SEO 50-article plan (business-marketing)
 → T-1 blog dispatch routine

instance: ps4
 → 190-company competitor monitoring
 → jp_strength scoring

At session start, each instance queries wbs.priority_for_instance and gets the top 5 tasks for that specific context.

What the AI Assistant Actually Does

1. Priority Extraction

{"action":"wbs.priority_for_instance","instance":"ps2"}//Returns:top5taskssortedbyrescue_score

rescue_score formula:

rescue_score =
 (overdue_days × 40) +
 (stale_days × 20) +
 (priority_rank × 25) +
 (progress_pct × 15)

Tasks that are overdue or haven't been updated in 3+ days automatically surface to the top. No manual re-sorting needed.

2. Automatic Progress Logging

At session start and end:

{"action":"wbs.update_progress","id":"task-uuid","progress":75,"status":"in_progress","session":"PS#2 S46: SEO Phase2 #18+#19 completed / commit abc123"}

Every session records: what changed, how far along, which instance did it, and what commit corresponds to the work.

3. Cross-Instance Handoff PRs

When one instance needs to pick up work from another's lane, the system auto-generates a handoff document:

docs/cross-instance-prs/20260426_ps2_to_vscode_seo_review.md

Implementation: Supabase Schema

CREATE TABLE wbs_tasks (
 id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
 title TEXT NOT NULL,
 category TEXT NOT NULL, -- business-hr, business-finance, etc.
 status TEXT DEFAULT 'pending', -- pending, in_progress, completed, blocked
 progress INTEGER DEFAULT 0, -- 0–100
 priority TEXT DEFAULT 'medium', -- low, medium, high, critical
 instance TEXT, -- assigned AI instance
 owner_instance TEXT, -- owner (usually same as instance)
 end_date DATE, -- deadline
 recovery_plan TEXT DEFAULT '', -- plan when blocked
 updated_at TIMESTAMPTZ DEFAULT NOW()
);

RLS Policy

-- Regular users see only their tasks
CREATE POLICY "users can view own tasks"
 ON wbs_tasks FOR SELECT
 USING (auth.uid() = owner_id);

-- service_role (Edge Functions) has full access
CREATE POLICY "service role all access"
 ON wbs_tasks FOR ALL
 USING (auth.role() = 'service_role');

Session Flow in Practice

1. Session starts → query wbs.priority_for_instance
2. Pick top rescue_score task
3. Set status = in_progress
4. Do the work
5. Update progress/status + log session note with commit hash

When Blocked

User: "I'm stuck on the company registration step"
AI: → wbs.list_tasks(category="business-legal", status="blocked")
 → Finds "司法書士・税理士契約" with empty recovery_plan
 → "Should I log 'Contact a judicial scrivener referral service' as the recovery plan?"

The AI has full DB context, so suggestions are specific — not generic productivity advice.

Comparison: Notion, Asana, vs WBS

Notion is flexible but weak on AI integration. Jibun Kaisha's WBS is purpose-built for one use case: a solo founder working with multiple AI agents in parallel.

What This Changes

The traditional task management cycle:

Human decides priority → Human updates status → Human checks what's next

The AI-assisted cycle:

AI surfaces priority → Human acts → AI logs progress → repeat

The cognitive overhead of "what should I work on and in what order" moves to the AI. The human focuses on the work itself.

Summary

• The real task management problem is organizational overhead, not task tracking
• rescue_score surfaces neglected and overdue work automatically
• Instance-based distribution lets multiple AI agents work in parallel without coordination conflicts
• Supabase DB integration gives the AI real-time context, enabling specific (not generic) suggestions

If you've tried every task management tool and still feel behind, the missing piece might not be a better tool — it's AI doing the organizing so you only do the executing.

Related Posts

• Real Costs of a Multi-AI Workflow
• Supabase Edge Functions + AI Cost Breakdown
• AI University: 230 Providers Guide

Jibun Kaisha — integrating the best of 21 competitors into one life management app

Live: https://my-web-app-b67f4.web.app/