All insights

Quarterly review19 September 20231 min read

Why we think AI will change how software is built

Our first note as a new company. Language models can already write useful code. The open question is how organisations that answer to auditors can trust what they produce.

We incorporated AIFIN LTD in August with one idea: AI will change how software is built, and the organisations held to the highest standards will need a careful way to use it.

What happened this quarter

The summer was busy. Meta released Llama 2 in July with a licence that allows commercial use, so capable models no longer have to come from a closed API. Anthropic released Claude 2 the same month, and OpenAI launched ChatGPT Enterprise in August with a promise not to train on business data. In the UK, the government confirmed that the first global AI Safety Summit will take place at Bletchley Park in November.

For people who build software, the more important change is quieter. Coding assistants have moved from novelty to habit. Many developers now accept a suggested line or function dozens of times a day.

Our take

Autocomplete is the start, not the end. Models that can suggest a line can, with the right tools around them, draft a test, explain a legacy module or check a page for accessibility problems. The hard part is not getting a model to write code. It is knowing whether the code is right, who checked it and what it was built from.

That is the problem we want to work on. Councils, regulators and banks cannot adopt tools on trust. They need a record of who approved what, tests that prove a requirement is met, and the freedom to change supplier. We plan to spend the next year building a way of working that provides those things, and to write here once a quarter about what we learn.

Keep reading

Related insights