Richard’s Blog - August 20, 2026 12:26

I love it when two news from the same day collide. In this case it's how experience in a domain helps or hinders when using LLMs.

First up, Cory Doctorow in The Nerve (4 August 2026), discussing the positive results from experienced software engineers:

Thanks to their skill and experience, these workers possess discernment, the ability to tell good code from bad [...] They engage with those tools on their own terms, to improve their work in the ways that they, in their expert judgment, consider beneficial.

Sounds right.

Second, Charlotte Blease (of the excellent Dr. Bot fame), also writing on 4 August 2026. Mental health therapists use LLMs in treatment planning, but the public are warned not to:

A clinical qualification does not transform a consumer chatbot into a regulated clinical device. Nor does professional status make its mistakes, biases or privacy risks disappear. If anything, and as readers of this blog will know, the stakes may even be higher when domain experts engage with algorithmic outputs. This is because, as a rich literature shows, experts may be more confident of their own judgment and less judicious than the average Joe, lay public, in appraising AI.

Sounds right, too.

Both could be right, I suppose. Coding has the built-in advantage of compilers and tests. Health and mental health don't. But they do have training and supervision which could help if it acknowledged that LLMs are going to get used.