← Alle Artikel

16. Februar 2026

5,700 Messages in 16 Days

I built a skill that analyzes my AI agent sessions. 5,700 messages, 65 projects, 111 context switches per day. Here are the numbers.

Built a skill today that reviews my pi agent sessions over any date range. Ran it on Feb 1–16. Here is what came out.

The Numbers

Messages to AI 5,700+
Active days 16
Unique projects 65
Avg messages/day ~360
Avg projects/day 10
Avg context switches/day 111
Avg active hours 11
Peak day Feb 3, 630 messages, 13 projects

~360 messages a day. One terminal. 65 different projects. Client work, infrastructure, personal tools, finance, content.

This is only my pi agent. Not counting my OpenClaw instance, not counting Codex sub-agents. Just pi.

Where Time Goes

Category Count
Shipping (commit/push) 1,119
Building features 1,023
Testing/debugging 848
Building skills/agents 616
Code review/cleanup 304
Project management 240
Communication 148
Finance/billing 119
Deploying 111
Research 78

I spend almost as much time building skills (616) as building features (1,023). Every skill I build saves hours the following week. The system keeps building itself.

Feb 6

One day. 529 messages. 13 projects. 157 context switches.

  • Built an analytics platform from scratch (Wachturm), deployed it
  • Set up Tailscale private infrastructure, security reviewed
  • Built a lead gen pipeline for a client. Apify scrapers into Supabase into Slack
  • Cleaned up 1,500+ Readwise articles with auto-tagging
  • Drafted client emails in German about invoice matching
  • Analyzed December spending, migrated expense tracking to a new skill
  • Deleted 20+ old GitHub repos
  • Built 3 new skills
  • Logged hours, posted weekly updates, did code reviews

One terminal. No app switching.

The Skill

About 400 lines of JavaScript. Parses the .jsonl session files that pi stores in ~/.pi/agent/sessions/, extracts user messages, computes stats: daily breakdown, project distribution, hourly activity, work categories, skill usage, message lengths, first and last message of each day.

# Last 7 days
review.js

# Specific range
review.js --from 2026-02-01 --to 2026-02-16

# JSON for piping
review.js --json

Going to run this weekly now. Partly to see where time actually goes vs where I think it goes. Partly for content.

Available in the skills pack.

Ein ähnliches Problem? Schreib mir.