Your AI week in review
Sep 8, 2026 to Sep 14, 2026 · from Claude and Codex · everything on this page was worked out on your own computer
That is about 212 books of text read and written by AI. A token is a small piece of a word — the unit AI tools count work in. More tokens means more work, and usually more cost.
Where did the effort go?
About 90% of what the AI read was material it had already seen — usually cheaper and faster than reading it fresh.
- Fresh reading · 10% — New text the AI had to read this time.
- Reused from memory · 90% — Text the AI had already seen and kept ready. Usually cheaper and faster.
- Writing · <1% — What the AI produced: answers, code, plans.
Who did the work?
12% of the work happened behind the scenes — helpers, hand-offs and safety checks you never typed a message for.
- Your conversations — 37.3 million tokens. The AI you talk to directly.
- Work handed to Codex — 2.2 million tokens. Bigger jobs one AI tool passed to another (for example Claude to Codex, or Codex to Claude). Matched by wording and timing, so treat as a strong estimate.
- Helpers the AI started — 1.7 million tokens. Smaller sub-tasks an AI started to work in parallel.
- Automatic safety checks — 1.2 million tokens. A second AI that double-checks risky actions before they run.
Which conversations used the most?
Each conversation includes the helpers, hand-offs and safety checks it started. The name is how the conversation began.
Show the top 25 with their IDs
| Conversation | ID | Prompts | Helpers | Tokens |
|---|---|---|---|---|
| Plan the launch checklist for the new mobile app release | demo-c1 | 3 | 12 | 28,149,977 |
| Compare three vendors for the analytics migration | demo-c3 | 2 | 3 | 6,826,874 |
| Fix the flaky login test and explain the root cause | demo-x1 | 1 | 6 | 3,744,206 |
| Draft the quarterly board update from these notes | demo-c2 | 2 | 0 | 1,890,908 |
| Rename the internal config keys across the repo | demo-x2 | 1 | 2 | 853,971 |
| Summarise last week's customer support tickets | demo-c4 | 1 | 0 | 797,347 |
| Translate the onboarding email into Bahasa Indonesia | demo-c5 | 1 | 0 | 97,179 |
To look at just one, pass its ID to --session on any command.
Which tasks were the heaviest?
Your typical task was about 1.8 million tokens. Tap a row to see what happened inside it.
1yes please, go ahead with all of itVery heavy18.1 million tokens
≈ 90 books · 10.2× typical
100 AI steps · 7 helpers involved · held up to ~509 pages in mind at once
- Your conversations: 15.8 million
- Work handed to Codex: 1.5 million across 2 sessions
- Helpers the AI started: 360 thousand across 1 session
- Automatic safety checks: 363 thousand across 4 sessions
Models used: claude-sonnet-5, codex-auto-review, gpt-5.6-sol
2add the rollout risks and ownersHeavy8.0 million tokens
≈ 40 books · 4.5× typical
24 AI steps · 1 helper involved · held up to ~571 pages in mind at once
- Your conversations: 8.0 million
- Automatic safety checks: 55.0 thousand across 1 session
Models used: claude-sonnet-5, codex-auto-review
3Compare three vendors for the analytics migrationHeavy5.1 million tokens
≈ 25 books · 2.9× typical
51 AI steps · 3 helpers involved · held up to ~300 pages in mind at once
- Your conversations: 4.0 million
- Helpers the AI started: 1.1 million across 3 sessions
Models used: claude-sonnet-5
4Fix the flaky login test and explain the root causeHeavy3.7 million tokens
≈ 19 books · 2.1× typical
47 AI steps · 6 helpers involved · held up to ~250 pages in mind at once
- Your conversations: 3.3 million
- Automatic safety checks: 480 thousand across 6 sessions
Models used: codex-auto-review, gpt-5.6-sol
5Plan the launch checklist for the new mobile app releaseTypical2.1 million tokens
≈ 10 books · 1.2× typical
36 AI steps · 4 helpers involved · held up to ~165 pages in mind at once
- Your conversations: 927 thousand
- Work handed to Codex: 680 thousand across 1 session
- Helpers the AI started: 303 thousand across 1 session
- Automatic safety checks: 166 thousand across 2 sessions
Models used: claude-sonnet-5, codex-auto-review, gpt-5.6-sol
6which one would you pick and whyTypical1.8 million tokens
≈ 9 books · 1.0× typical
8 AI steps · 0 helpers involved · held up to ~346 pages in mind at once
- Your conversations: 1.8 million
Models used: claude-sonnet-5
7shorter, and lead with the numbersTypical1.1 million tokens
≈ 1,598 pages · 0.6× typical
12 AI steps · 0 helpers involved · held up to ~153 pages in mind at once
- Your conversations: 1.1 million
Models used: claude-sonnet-5
8Rename the internal config keys across the repoLight854 thousand tokens
≈ 1,280 pages · 0.5× typical
17 AI steps · 2 helpers involved · held up to ~116 pages in mind at once
- Your conversations: 716 thousand
- Automatic safety checks: 138 thousand across 2 sessions
Models used: codex-auto-review, gpt-5.6-sol
9Draft the quarterly board update from these notesLight825 thousand tokens
≈ 1,237 pages · 0.5× typical
16 AI steps · 0 helpers involved · held up to ~107 pages in mind at once
- Your conversations: 825 thousand
Models used: claude-sonnet-5
10Summarise last week's customer support ticketsLight797 thousand tokens
≈ 1,195 pages · 0.5× typical
18 AI steps · 0 helpers involved · held up to ~94 pages in mind at once
- Your conversations: 797 thousand
Models used: claude-sonnet-5
Show the top 25 as a table
| # | Task | Tokens | Steps | Helpers | Size |
|---|---|---|---|---|---|
| 1 | yes please, go ahead with all of it | 18,055,805 | 100 | 7 | Very heavy |
| 2 | add the rollout risks and owners | 8,017,290 | 24 | 1 | Heavy |
| 3 | Compare three vendors for the analytics migration | 5,055,508 | 51 | 3 | Heavy |
| 4 | Fix the flaky login test and explain the root cause | 3,744,206 | 47 | 6 | Heavy |
| 5 | Plan the launch checklist for the new mobile app release | 2,076,882 | 36 | 4 | Typical |
| 6 | which one would you pick and why | 1,771,366 | 8 | 0 | Typical |
| 7 | shorter, and lead with the numbers | 1,065,672 | 12 | 0 | Typical |
| 8 | Rename the internal config keys across the repo | 853,971 | 17 | 2 | Light |
| 9 | Draft the quarterly board update from these notes | 825,236 | 16 | 0 | Light |
| 10 | Summarise last week's customer support tickets | 797,347 | 18 | 0 | Light |
| 11 | Translate the onboarding email into Bahasa Indonesia | 97,179 | 6 | 0 | Light |
When were you busiest?
Tokens per day (M = million, B = billion). Days use your computer's clock.
Show as a table
| Day | Tokens | Roughly |
|---|---|---|
| Tue Sep 8, 2026 | 2,874,229 | 14 books |
| Wed Sep 9, 2026 | 26,073,095 | 130 books |
| Thu Sep 10, 2026 | 5,635,114 | 28 books |
| Fri Sep 11, 2026 | 5,055,508 | 25 books |
| Sat Sep 12, 2026 | 1,771,366 | 9 books |
| Sun Sep 13, 2026 | 853,971 | 1,280 pages |
| Mon Sep 14, 2026 | 97,179 | 146 pages |
Which AI tools and models?
By tool
By model
What this page cannot see
- Only work that was written to log files on this computer. Chats in a web browser or on a phone, and work done on someone else's server, are invisible here.
- Tokens are not dollars and not your plan's allowance. Different AI providers count differently.
- 20 helper/safety sessions were linked to their parent exactly; 3 hand-offs were matched by wording and timing (a strong estimate, not a certainty).
- Read 0 log files.
Words used on this page
- Token
- A small piece of a word. Roughly 3 tokens is 2 words. About 667 tokens fill a page and 200,000 fill a 300-page book (rules of thumb, not exact).
- Fresh reading / reused from memory
- Each step, the AI re-reads the whole conversation so far. Providers remember part of it ("caching"), which is usually cheaper and faster than reading it new.
- Helper, hand-off, safety check
- Modern AI tools often start other AI sessions to do part of the job. Those sessions use tokens too, so they are counted with the task that caused them.
- Light · Typical · Heavy · Very heavy · Extreme
- Size compared with your own typical task (the median): under half is Light, up to 2× Typical, up to 5× Heavy, up to 20× Very heavy, beyond that Extreme.
- Simple · Multi-step · Team effort
- How complicated the job was: Simple is a handful of steps; Team effort means several helpers or 150+ steps.
Generated Sep 25, 2026 11:18. This file may contain snippets of your prompts — check before sharing it.