Dashboard
Every number here comes from my own Claude Code usage database — not a survey, not an estimate. This is what 314 active days of agentic engineering actually looks like.
October 29, 2025 → October 15, 2026 ·
Coverage is continuous from December 1, 2025; earlier rows were backfilled from transcripts carried off an older machine and are sparse (October 2025: 3 days; November 2025: 12 days).
Green figures are the change since the previous snapshot, a day earlier (Oct 14). The counters animate from those numbers to these.
Records
The extremes, counted from the same stored rows as the heatmap. They describe the snapshot taken October 15, 2026: a record broken since then is not in here yet.
Lately
The totals above are everything since October 29, 2025. This is the recent end of it: the last week, month and quarter, each set against the stretch just before.
| Window | Sessions | Messages | Messages per day |
|---|---|---|---|
| Today so far Oct 15, still in progress | 15 | 981 | no comparison |
| Last 7 days Oct 8 – Oct 14 | 232 down 51% | 7,516 down 63% | |
| Last 30 days Sep 15 – Oct 14 | 1,407 up 97% | 60,536 up 86% | |
| Last 90 days Jul 17 – Oct 14 | 2,686 up 145% | 122,893 up 125% |
Windows are calendar days ending Oct 14, the last complete day. Each change compares a window with the one of equal length immediately before it. A day with no activity counts as zero and shows as a gap. Every row's bars are scaled to that row's own busiest day, so heights are not comparable between rows.
How I work
This is the panel that matters. Using AI as autocomplete produces an Edit-heavy profile — code being generated into files. Mine is Bash-dominant, with Read at 2.6× Edit. That's not authoring. That's running, inspecting, verifying — operating a system.
275 distinct tools used. Top 14 shown.
What actually got built
GitHub only sees a git push. My work moved to
wrangler deploy
and stopped touching one months ago, so GitHub's graph goes quiet while the
work does not. This is the same period, measured at the source, with GitHub's own count set against it below.
Hover or tap a square, or step with the day buttons
| Month | Days of agent work | Days with a GitHub contribution |
|---|---|---|
| Oct 25 | | |
| Nov 25 | | |
| Dec 25 | | |
| Jan 26 | | |
| Feb 26 | | |
| Mar 26 | | |
| Apr 26 | | |
| May 26 | | |
| Jun 26 | | |
| Jul 26 | | |
| Aug 26 | | |
| Sep 26 | | |
| Oct 26 | | |
days of agent work days with a GitHub contribution
Across the whole window GitHub shows a contribution on 186 of 352 days; the agent worked on 314. The widest gap is January 2026: 10 days on GitHub against 28 of work.
GitHub's figures include private repositories, which a public profile may not display, and GitHub assigns days by its own rules, so a day can sit a day away from the local dates used elsewhere on this page. Counts only: no repository names are kept.
When I work
46.3% of everything happens between 6pm and 2am. Peak hour is 8pm; the quietest is 5am, when I am reliably asleep — the 3am-to-8am window is just 2.6% of everything. The single busiest slot of the week is Saturday at 8pm, with 6,252 messages across the whole window. I didn't decide to be an evening builder; the data just says I am. All times Mountain.
6pm–2am 2am–6pm
By day of week
Busiest day: Thu
Models
12 models across the window, 143.2M output tokens and 46.6B cache-read tokens (both summed per session by the ingest). I move as the frontier moves rather than sitting on a default.
Where the time goes
Personal projects, experiments, and my own tooling. Client work is aggregated and never named. Hours are active time, not wall-clock — see the note at the bottom.
227 further projects (994.2h) are client and private work. They count toward every total on this page but are never named — the dashboard runs an allowlist, so anything not explicitly cleared for publication is anonymous by default.