40+ autonomous AI agents. One human. Running 24/7.
I wanted to see what happens when AI doesn't just answer questions but actually runs a company. Research markets. Write content. Monitor systems. Post on social media. Ship products. All without being told what to do.
IT Division Manager by day. Builder by night. ADHD brain running at full speed with AI as externalized executive function. Gives direction, makes decisions, catches what the agents miss. The agents don't replace thinking. They replace forgetting.
The primary agent J talks to directly. Coordinates everything, delegates to specialists, maintains memory and context across sessions. Handles complex reasoning, strategy, and anything requiring judgment. Spawns sub-agents for heavy work. The first point of contact between human and machine.
Wakes up at 4:45 AM. Pulls email, calendar, weather, and vault archaeology into one scannable briefing. Delivered before J's feet hit the floor.
Reviews overnight builds, checks the calendar, surfaces avoided tasks. Calls out what's been dodged. ADHD-aware: uses pressure, not patience.
Health checks, cron verification, update detection, old file cleanup, uncommitted change commits. The janitor nobody thanks.
Scans the agent's own behavior files for outdated info, conflicting rules, stale references. Fixes them directly or flags for approval. The system editing itself.
Runs @whatmefocus on X. Searches for trending ADHD and AI agent threads, crafts replies and original posts, queues content through a leveling system. Owns the "Agentic Executive Function" category.
Reviews the day's work every night at 10 PM. Suggests blog post topics and angles for doss.me. Knows when to push and when to say "nothing today, don't force it."
Listens to every conversation for business signals. "Did J solve a problem others would pay for?" Seeds get logged to a growing idea file.
Scans the AI agent and ADHD tool ecosystem daily. Only reports if something is immediately actionable. Either implements the finding or skips it. No summaries.
Tracks snow levels, NWS forecasts, and conditions at J's 3,200 ft elevation. Alerts only when conditions change.
Checks Caltrans for SR-41 and SR-49 chain controls. J drives 100+ miles daily through mountain highways.
PG&E outage tracking, PSPS warnings, and emergency alerts for the mountain community.
Tree damage, CHP incidents, CAL FIRE alerts, evacuations. Living in the Sierra foothills means nature has opinions.
Scans county and state job boards for IT leadership positions. Tracks deadlines and salary ranges across multiple jurisdictions.
Weekly scan of county Board of Supervisors for IT mentions, budget discussions, leadership changes. Knows the politics before J walks in Monday.
Maintains a full-text search index (SQLite FTS5) of 1,400+ Obsidian notes. Rebuilt every 6 hours.
Syncs vault notes to Neo4j. Discovers connections between ideas, finds orphaned knowledge, traces relationship paths.
Every night at 11 PM, reads the raw daily log and distills it into long-term memory. Decisions, lessons, relationship context. The raw becomes curated.
Pushes memory files to Google Drive every 6 hours. Redundancy against local failure.
Processes the day, identifies what broke, builds one thing to fix or improve it. Verifies it works. Wires it into the system. Ships it before sunrise.
Hard rule: build something every night that improves J's life. A tool, a dashboard, an automation. Something he didn't ask for and didn't know he needed. Surprise and delight.
Fixes broken links, adds missing backlinks, organizes misplaced notes, cleans the Obsidian vault while everyone sleeps.
Fixes formatting, deletes empty notes, consolidates duplicates. Keeps the phone clean for quick reference.
Sunday mornings. Reviews the entire week for patterns, proposes new rules for the agent's own behavior, measures loop closure rates, creates knowledge connections. The system improving itself.
Friday afternoon. Hard numbers: how many overnight builds are in active use? How many lessons changed behavior? What's the loop closure rate? No narratives, just metrics.
Sunday morning coffee read. Tech trends, ADHD research, items for Chrissie, and one thing worth knowing. Curated, not dumped.
Weekly Sunday check. Daily note streaks, missing MOCs, empty files, uncommitted changes. Keeps the knowledge base healthy.
11:30 PM. Pulls scoring signals from the day's interactions: corrections, praise, tasks completed, patterns.
11:45 PM. Scores the day on multiple dimensions. Compares against yesterday. Reports honestly. J's weekly 1-5 rating is the external anchor.
One lesson per day. If the same pattern appears 3+ times, it becomes a permanent behavioral rule. The system learning from its mistakes.
"The agents don't replace thinking. They replace forgetting. And for an ADHD brain, that's everything."
Built on OpenClaw. Running on a Mac Studio in a mountain town.
Three tiers of intelligence. Opus thinks. Sonnet executes. Haiku maintains. Not everything needs a genius. Most things need reliability.
Main session, Deep Think, Digital Dreaming, Friday Insight
Briefings, monitoring, content, scoring, builds
Vault sync, git, file org, cleanup, health checks