Why We Built Our Own AI Coding Time Tracker — trackmy.codes Review: $29/yr Reveals the “Engine Running” vs “Actual Work” Gap

TL;DR: trackmy.codes ($29/yr) passively logs active vs idle time across Claude Code, Cursor, and Codex. Our first month showed 40% of “AI session time” was waiting on agents — we cut estimates by 30% and improved sprint planning. Full review →

The Context

Two-dev team, ~120 hrs/mo combined AI-assisted coding. We suspected “Claude Code running for 4 hours” ≠ 4 hours of productive output, but had no data. Existing tools (WakaTime, RescueTime) count editor focus, not agent activity. trackmy.codes launched June 2026 claiming cross-tool “engine running” vs “actual work” detection.

What We Tested

Tool	Use Case	Verdict	Why
trackmy.codes ($29/yr)	Cross-tool active/idle detection	✅	Detects keystrokes, terminal output, API calls per tool; 40% idle rate revealed
WakaTime	Editor focus time	❌	Counts Cursor/Claude Code UI focus equally; can’t distinguish agent thinking vs human typing
RescueTime	App/window focus	❌	Same limitation; “Claude Code” window open ≠ working
Manual Toggl entries	Self-reported	❌	Unreliable; devs forget to stop/start during context switches

The Pivot Point

Week 1 of trackmy.codes (June 3–9, 2026): My logged “Claude Code sessions” averaged 3.2 hrs/day. trackmy.codes reported 1.9 hrs active (keystrokes + terminal output + API responses), 1.3 hrs idle (agent thinking, waiting for approval, reading output). Same pattern for Cursor (2.1 logged → 1.4 active) and Codex (1.8 logged → 1.2 active). The “engine running” tax is real. We now estimate tasks at 60% of logged AI time.

What We Use Now

trackmy.codes Pro ($29/yr) — Installed via their shell script (curl -fsSL trackmy.codes/install.sh | bash):

• Auto-detects claude-code, cursor, codex processes
• --idle-threshold 30s (configurable): no keystrokes/terminal output/API response = idle
• Daily summary webhook → our #dev-metrics Slack: “Seb: 2.1h active / 0.9h idle (CC: 1.3/0.6, CX: 0.8/0.3)”
• Monthly export → Notion sprint retro template for velocity calibration

Workflow change: Before sprint planning, we pull last 4 weeks’ active hours per dev, multiply by 0.8 safety factor, that’s our “AI-assisted capacity.” No more “I’ll knock this out in 2 hours with Claude” → 4 hours later still waiting on Opus.

When You’d Choose Differently

• Solo dev, low volume: Manual tracking fine; $29 ROI unclear under ~20 hrs/mo AI time.
• Enterprise/team >10: trackmy.codes team dashboard (coming Q3 2026 per their Discord) needed for roll-up reporting.
• Privacy-hardened env: Tool sends process names + timestamps to their cloud; we verified no code/content exfiltration via MITM proxy test, but air-gapped teams should wait for self-hosted option.

Tool Crucible Rating

Dimension	Rating (1–5)	Notes
Overall	4.5	Novel category; actionable data from day 1
Ease of Use	5	One-line install; zero config works
Value	5	$29/yr pays for itself in one corrected estimate
Support	3	Early launch; Discord only; roadmap public but unproven

---

This is part of our AI Coding Tool Evaluation series. See full review: trackmy.codes Deep Dive: The First Cross-Tool AI Coding Time Tracker