AI-Orchestrated Daily Operations
62 custom AI skills replaced 27.5 hours of weekly admin overhead with one-command automation — morning briefings, time tracking, weekly reports, and system monitoring running hands-free in production.
The Challenge
A solo IT Director managing 6 offshore developers across 3 time zones, 5 client organizations, and 1,000+ active tasks in Wrike. The operational surface area was enterprise-scale. The team size was not. Every tool, every inbox, every dashboard was another tab competing for the same finite attention — and the work kept growing while the headcount stayed at one.
Every morning started with 60-90 minutes of context-switching before any real work could begin: scanning 3 email accounts across Microsoft 365 and Gmail, checking 5 separate calendars, reviewing overnight activity in Teams and Wrike, and monitoring SSL certificates, uptime alerts, and sFTP connectivity across client infrastructure. That was just the warm-up. The actual architecture, client strategy, and team coordination work had to fit into whatever was left of the day.
Evening wrap-up was consistently skipped. 30-40% of billable hours went unlogged because manual time entry was the first thing cut when the day ran long. Weekly VP reports took half a day to compile from scattered data sources. And the single biggest risk was the simplest: one person was the sole carrier of all operational context. If that person was unavailable for a day, operational visibility dropped to zero. No briefings, no monitoring, no reports — just silence.
Our Approach
The journey started in late 2024 with AI as a search engine — GitHub Copilot for code completion, ChatGPT for analysis, Claude Desktop for document review. Useful, but marginal. The breakthrough came from a shift in framing: stop treating AI as a code generator and start treating it as an operational co-pilot that could own entire workflows end-to-end.
Over a 65-day sprint beginning in January 2026, we built 62 custom AI skills. Each skill solves one specific workflow problem — no monoliths, no do-everything scripts. The architecture is orchestration-first: small composable commands that chain together. /start-the-day calls 4 sub-skills. /periodic-reports calls 5. Each sub-skill runs independently and can be invoked on its own — the orchestrators just wire them into sequences.
Token optimization was not optional — it was mandatory for viability. Raw MCP server calls consumed ~500K tokens for a single morning stack. At that rate, the system would cost more than the time it saved. We built 5 Programmatic Tool Calling (PTC) wrappers in Python that pre-compile API schemas into compact CLI commands, reducing token consumption by 85-99% per service. Every output is dual-stored: Obsidian vault for local searchability, Notion for mobile access and team visibility.
The Solution
Morning Stack
Daily Briefing
5 calendars + 3 email accounts + Teams + Wrike
Standup Prep
Overnight commits, code review queue, talking points
Health Dashboard
SSL certs, uptime, alerts, sFTP status
Meeting Prep
Previous notes, related tasks, participant context
Evening Stack
Time Logger Recap
Calendar scan, missing entry detection, task matching
Daily Recap
Planned vs. actual, carryover items, Notion + Obsidian save
Weekly Stack (Friday)
Friday Wrapup
Wins, incomplete, lessons
Weekly Report
VP report from all sources
Wrike Triage
Stale tasks, WIP limits
Capacity Report
Deep work vs. meetings
Client Tracker
Touchpoint freshness
Obsidian Vault
1,062 searchable artifacts
Notion
5 structured databases
Morning Briefing Pipeline
- 6-phase pipeline: Notion entry, 5 calendars, 3 email accounts, Teams action items, Wrike tasks, summary
- Email categorization: Action Required, FYI, System Noise, Delegatable
- Wrike tasks pulled and categorized by workflow status across all client organizations
- Full briefing generated and archived in under 15 minutes, hands-free via headless mode
AI-Assisted Time Tracking
- Conversational mode: describe what you worked on, AI finds the Wrike task and creates the entry
- Recap mode: scans today's calendar, cross-references existing timelogs, suggests missing entries
- Reduced missed billable hours from 30-40% to under 5%
- Task matching across 1,000+ Wrike tasks using context-aware search
Weekly Reporting Automation
- VP report aggregated from email, calendar, Wrike, code reviews, and time logs
- Capacity analysis: deep work hours vs. meeting load, intake vs. throughput
- Wrike triage: stale task identification, WIP limit enforcement, triage recommendations
- Full weekly stack runs as one command — was 2-3 hours of manual compilation
Token Optimization Layer (PTC)
- 5 Python CLIs replacing verbose MCP tool calls with compact commands
- Wrike: 97% reduction (50K tokens per query down to 1.5K)
- M365: 99% reduction (100K per email batch down to 1K)
- Morning stack total: ~500K tokens reduced to ~50K — 10x cost reduction
Infrastructure Monitoring
- SSL certificate expiration checks across all client domains
- Uptime monitoring via Pingdom with PTC-optimized API calls
- sFTP connectivity verification for file transfer workflows
- Alert aggregation: one dashboard instead of five separate monitoring tools
Institutional Memory
- 1,062 artifacts: 194 briefings, 206 context saves, 151 meeting notes, 427 compaction summaries
- Dual-storage: Obsidian vault for local search, Notion for mobile and team access
- Every AI session preserves context for future resumption — no cold starts
- 15 years of email history processed into structured, AI-queryable retrospectives
Results
~24h/week
Time reclaimed from admin overhead
$155K/year
Annual capacity freed at enterprise rates
62
Custom AI skills built in 65 days
194
Daily briefings generated and archived
<5%
Missing time entries (was 30-40%)
10x
Token cost reduction via PTC architecture
15 min
Morning prep time (was 60-90 min)
5
PTC CLIs with 85-99% token savings
1,062
Institutional knowledge artifacts preserved
Time Savings Breakdown
| Activity | Before | After | Saved/Week |
|---|---|---|---|
| Morning prep | 7.5h | 1.25h | 6.25h |
| Code reviews | 10h | 1.5h | 8.5h |
| Weekly reports | 3h | 0.3h | 2.7h |
| Standup prep | 2.5h | 0h | 2.5h |
| Time logging | 2.5h | 0.5h | 2.0h |
| Meeting prep | 2h | 0h | 2.0h |
| Total | 27.5h | 3.55h | ~24h |
The Multiplier Effect
- One person now handles what would otherwise require a dedicated code reviewer, a project coordinator, and a technical writer — without adding headcount
- Operational visibility no longer depends on one person being available — briefings, monitoring, and reports run whether anyone is watching or not
- 1,062 institutional knowledge artifacts mean every future AI session starts with full context instead of cold — compounding returns on every interaction
- We have a separate case study on the code review pipeline — the 179 automated reviews and 31 Critical findings caught are part of this same system
“We didn't set out to build a product. We set out to survive a workload that was outgrowing one person. Sixty-two skills later, the workload didn't shrink — our capacity to handle it expanded by an order of magnitude. That's the difference between using AI and being augmented by it.”
Frequently Asked Questions
Want to Reclaim 24 Hours a Week?
We built this system for ourselves — now we build them for organizations drowning in the same operational overhead. Your tools stay yours. We add the AI layer that makes them 10x more useful.