34%. Not yet.

An index from 0 to 100, not a literal share of software that is solved. It adds up indirect evidence that AI can build and maintain useful software with little human help, at acceptable cost and reliability, measured as movement since ChatGPT launched in November 2022 across 4 questions and the signals behind them.

The line is recalculated from today's data rather than recorded at the time, and it can jump when a new signal joins.

The goals behind each bar are our judgment. Moving them all easier or harder gives 30% to 40%. Why each goal.

Failures are counted per git push, and AI agents may inflate pushes. Counting them raw instead, with no adjustment, the score would be 32%.

Most bars use a log scale, so they track multiples rather than the share of a goal. Growing 32 times fills half the bar for a goal of 1,000 times.

"Solved." needs 90% overall and at least 75% on every question, so one strong area cannot hide a weak one.

How it worksScore=1n∑i=1nQuestioniQuestioni=∑j=1mwj Signalj∑j=1mwj\begin{aligned} \text{Score} &= \frac{1}{n}\sum_{i=1}^{n} \text{Question}_i \\[6pt] \text{Question}_i &= \frac{\sum_{j=1}^{m} w_j\,\text{Signal}_j}{\sum_{j=1}^{m} w_j} \end{aligned}

Overall score (out of 100) · 2023-01 – 2026-11

2026-11 — Overall score (out of 100): 34
View data table
Period / observationOverall score (out of 100)
2023-011
2023-022
2023-035
2023-046
2023-056
2023-067
2023-077
2023-087
2023-097
2023-106
2023-116
2023-128
2024-0111
2024-028
2024-0311
2024-0411
2024-0512
2024-0614
2024-0714
2024-0815
2024-0915
2024-1015
2024-1115
2024-1217
2025-0116
2025-0220
2025-0320
2025-0422
2025-0523
2025-0625
2025-0725
2025-0825
2025-0926
2025-1026
2025-1127
2025-1227
2026-0129
2026-0231
2026-0334
2026-0434
2026-0534
2026-0633
2026-0734
2026-0834
2026-0935
2026-1034
2026-1134

Can AI do the work?

58%

Is it being used to build?

41%

This measures use, not whether the software works. Pushes and tokens can also grow through retries, longer reasoning or automation, so growth here is the weakest kind of evidence in the score.

Are fewer people needed?

30%

Is software cheaper and more reliable?

8%

Money is compared with the economy and inflation, and failures with how much code is pushed. A rise scores 0%, which means no sign of change yet. AI agents may be inflating the number of pushes, which would make failures look rarer than they are, so each failure signal shows what it would score without that adjustment.