Aby Thannikal Joseph
Menu

AI coding activity

Where my work and hobbies overlap

My interests outside work often mirror what I do professionally: AI, engineering, testing and financial services.

A lot of my personal projects happen in the evenings or on weekends, sometimes with a movie playing in the background, or during my Tube commute while I delegate tasks to Codex or Claude Code.

I enjoy trying different tools and models, understanding what they're good at, learning as I go and building things that interest me.

Everything on this page reflects usage from my personal AI subscriptions, not from AI tools I use at work. The token counts reflect how much I've been using these tools to experiment, learn and build in my own time. I try to use them efficiently too, because cheaper models aren't always more cost-effective. And no, it's not an attempt at token-maxxing :)

Aby Joseph @abytjoseph

Figures to

Refreshed periodically; not live.

Tokens processed

1.94B

Claude Code 1.51B Codex 431.6M

Both tools combined, as each reports it. Models used

Active days

40

in the last 12 months · Claude Code 11, Codex 32

Longest streak

11 days

Codex, 29 Sept – 9 Oct 2026

Code and text written

8.6M

tokens of output from Claude Code

What came out of it

The things I’ve been building and poking at with these tools.

  • Rebuilding this website

    Moving this site off an old Gatsby template onto Astro, and rethinking how it’s put together, hosted and deployed along the way. Claude Code and Codex did much of the implementation, review and testing, with a lot of back and forth.

  • Payments learning app

    A small mobile game where you play an analyst at a fictional bank and handle synthetic payments that go wrong. Each case gives you the messages and clues, asks you to make a call, plays out what happens and then explains it.

    Built and tested, but not public yet.

  • Personal experiments and learning

    Smaller things for seeing what different tools and models can and can’t do. Test-AbyLity tried LLMs on everyday testing tasks; Jev was a look at a classifier model that returns probabilities instead of text. Some of these turn into something useful. Plenty are just a way to learn.

When I build

Each square is a day. Select one to see the tools and models used.

Show activity for

40 active days in the last 12 months, 3 with both tools11 active days · 1.49B tokens32 active days · 384.9M tokens on this laptop

  • LessUnder 1M tokens1M–10M tokens10M–100M tokens100M or more tokensMoretokens per day: under 1M · 1–10M · 10–100M · 100M+
  • Both tools that day: Claude Code top left, Codex bottom right
  • Active, no daily token count

Select a day in the calendar to see its details.

Older Cursor and Codex usage isn’t logged here.

Build rhythm

Work comes in focused sprints around projects rather than a steady drip.

Time range
Show

Tokens per dayRunning total of tokens17 Oct – 23 Oct 2026

  • Claude Code 1.19B
  • Codex 1.5M
Show the 7 days as a table
Tokens per day, 7 days.
DayClaude CodeCodexRunning total
18 Oct 2026542.3M1.5M543.7M
19 Oct 2026120.1M0663.9M
20 Oct 20267.4M0671.3M
21 Oct 2026256.3M0927.6M
22 Oct 2026260.7M01.19B
23 Oct 20266.7M01.20B

Periods with no recorded activity are left out.

Tokens per dayRunning total of tokens24 Sept – 23 Oct 2026

  • Claude Code 1.49B
  • Codex 330.5M
Show the 30 days as a table
Tokens per day, 30 days.
DayClaude CodeCodexRunning total
26 Sept 202609.0M9.0M
27 Sept 2026071.2M80.2M
29 Sept 2026069.7M149.9M
30 Sept 2026050.4M200.3M
1 Oct 20260699.3K201.0M
2 Oct 202606.2M207.2M
3 Oct 2026032.9M240.1M
4 Oct 2026045.9M286.0M
5 Oct 202698.8M2.1M386.8M
6 Oct 2026021.1M408.0M
7 Oct 202606.1M414.1M
8 Oct 20260417.1K414.5M
9 Oct 20260419.8K414.9M
10 Oct 2026193.2M0608.0M
13 Oct 202601.8M609.9M
15 Oct 2026010.7M620.5M
16 Oct 20260436.5K621.0M
18 Oct 2026542.3M1.5M1.16B
19 Oct 2026120.1M01.28B
20 Oct 20267.4M01.29B
21 Oct 2026256.3M01.55B
22 Oct 2026260.7M01.81B
23 Oct 20266.7M01.82B

Periods with no recorded activity are left out.

Tokens per dayRunning total of tokens26 Jul – 23 Oct 2026

  • Claude Code 1.49B
  • Codex 347.8M
Show the 90 days as a table
Tokens per day, 90 days.
DayClaude CodeCodexRunning total
28 Aug 202602.9M2.9M
6 Sept 2026014.4M17.3M
26 Sept 202609.0M26.3M
27 Sept 2026071.2M97.5M
29 Sept 2026069.7M167.2M
30 Sept 2026050.4M217.6M
1 Oct 20260699.3K218.3M
2 Oct 202606.2M224.5M
3 Oct 2026032.9M257.4M
4 Oct 2026045.9M303.3M
5 Oct 202698.8M2.1M404.1M
6 Oct 2026021.1M425.2M
7 Oct 202606.1M431.4M
8 Oct 20260417.1K431.8M
9 Oct 20260419.8K432.2M
10 Oct 2026193.2M0625.3M
13 Oct 202601.8M627.2M
15 Oct 2026010.7M637.8M
16 Oct 20260436.5K638.3M
18 Oct 2026542.3M1.5M1.18B
19 Oct 2026120.1M01.30B
20 Oct 20267.4M01.31B
21 Oct 2026256.3M01.57B
22 Oct 2026260.7M01.83B
23 Oct 20266.7M01.83B

Periods with no recorded activity are left out.

Tokens per weekRunning total of tokens20 Oct 2025 – 23 Oct 2026

  • Claude Code 1.49B
  • Codex 384.9M
Show the 12 months as a table
Tokens per week, 12 months.
Week startingClaude CodeCodexRunning total
10 Nov 20250297.7K297.7K
2 Mar 202601.7M2.0M
9 Mar 2026011.1M13.0M
16 Mar 2026015.3M28.4M
30 Mar 202602.5M30.9M
13 Jul 20266.6M6.2M43.6M
24 Aug 202602.9M46.6M
31 Aug 2026014.4M60.9M
21 Sept 2026080.2M141.1M
28 Sept 20260205.8M346.9M
5 Oct 2026291.9M30.1M669.0M
12 Oct 2026542.3M14.4M1.23B
19 Oct 2026651.3M01.88B

Periods with no recorded activity are left out.

Claude Code and Codex side by side

Figures as each tool reports them.

  • Claude Code: 1.51B
    Tokens processed
    Codex: 431.6M
  • Claude Code: 11
    Active days, last 12 months
    Codex: 32
  • Claude Code: 542.3M18 Oct
    Busiest day
    Codex: 76.1M
  • Claude Code: Opus 5.598.6% of tokens
    Most-used model
    Codex: GPT-6 Luna51.8% of tokens

Models I use

Which models did the work in each tool, by share of tokens.

Claude Code models

Claude Code tokens by model
ModelShareOutput
Opus 5.598.6%8.4M
Sonnet 4.60.6%126.8K
Opus 4.80.4%73.9K
Haiku 4.50.4%13.2K

Codex modelslocal logs

Codex tokens by model, local logs
ModelShareTokens
GPT-6 Luna51.8%199.3M
GPT-6 Sol22.7%87.2M
GPT-5.47%27.0M
GPT-6 Astra5.2%20.0M
8 other models13.3%51.4M

Also used: Auto-review, GPT-5.6 Sol, GPT-5.4 Mini, GPT-6.1 Sol, GPT-5.6 Terra, GPT-5.3 Codex, GPT-5, GPT-5.5. Auto-review is the model Codex uses to check each action before it runs.