The 2026 AI Index, read by a working developer

Most of what we hear about artificial intelligence comes from the people selling it. That is why I read the AI Index each year. It is produced by an independent team at Stanford University, it measures instead of predicting, and it is honest about what it cannot measure. This year's edition is long. These are the parts I keep coming back to.

It is not slowing down

The report's first message is that capability is still accelerating. One example stuck with me: on a benchmark of real computer tasks carried out by AI agents, the success rate rose from about 20 percent in 2025 to 77.3 percent, according to the Index. That is not a polish on last year's models; it is a different class of tool.

But it is uneven in odd ways

The same report notes that robots attempting ordinary household chores succeeded in only about 12 percent of tasks. Stanford's own summary puts it neatly: AI can win a mathematical olympiad but cannot reliably tell the time. For anyone deciding where to use these systems, that unevenness is the most practical finding of all. They are brilliant at some narrow things and unreliable at others that look easier.

People are adopting it faster than the internet

Generative AI reached 53 percent of the population within three years, by the Index's estimate, faster than personal computers or the internet did. Adoption is highest in places such as Singapore and the United Arab Emirates; the United States ranks well down the list. Adoption is not decided by who builds the models.

Money and power are concentrating

Corporate investment in AI reached about 581.7 billion dollars in 2025, the report says, and private investment in the United States alone was more than twenty times that in China. At the same time, the Index's measure of how openly the big model makers disclose how their systems are built fell sharply, from an average of 58 to 40. More money and less openness is a combination worth watching.

Young workers feel it first

Employment of software developers aged 22 to 25 in the United States has fallen by nearly 20 percent since 2024, according to the report. Entry-level work is where AI arrives first, because it is where tasks are most clearly defined.

What I take from it, here

My own reading, and it is only that: the countries that gain most from this decade will not be the ones that build the largest models, but the ones whose people learn to use them well and to check them. That is an opportunity for Kenya, which has a young, mobile-first population and a habit of adopting useful technology quickly. It is also a warning, because the junior roles through which people usually learn a trade are the ones being reshaped first.

If you read one long document about AI this year, make it this one.