Claude Code IDE extension compromised with persistent SSH tunnel; engineers question agentic trust; AI beats baselines on economic forecasting

Engineering

Martin Fowler

Why are good developers so wary of agentic development?

Rachel Laycock, CTO at Thoughtworks, explains why experienced developers are wary of agentic development. Their main concern is trust, because few teams have run systems built this way. A system can be built and still fail in production. In a one-week test, agents wrote all the code. The team spent most of its time on the checks that guide the agents. Frequent red builds exposed gaps in the delivery pipeline.

Simon Willison

How I Built the Most AI-Pilled Company in the World

Simon Willison shares an essay by Jacob Sansbury about Spawn, a multiplayer game. Autonomous agents build Spawn and ship live code every hour. The company spends 30 times its payroll on artificial intelligence (AI) tokens. The author says only Fable 5 and 5.1 models make this autonomy possible.

seriot.ch

How curl was accidentally Turing-complete for nine months

Nicolas Seriot shows that curl config files were Turing-complete from version 8.12 to 8.16. The proof assumes unlimited string length and unlimited resources. The author maps a two-counter Minsky machine onto curl config files. Curl version 8.17 adds a five-level include cap, which breaks this construction. The author states that this issue has no security impact.

Martin Fowler

LLM Voice

Martin Fowler describes a style of text that seems written by a large language model (LLM). He calls this style LLM Voice. Readers lose interest when they notice LLM Voice. Fowler says that detection tools are not reliable. He tells authors to write their own prose. He argues that when an LLM writes the prose, it also does the thinking.

Social Sciences

Construction Physics

A New Kind of Chemistry

Isotopes are atoms of one element with different neutron counts. Many isotopes are rare, costly, and hard to make. Accelerators and reactors make many isotopes today. Fusion companies now plan to make isotopes too. The author says cost will decide how much these methods matter.

Psyche

'Heroes work here': why I quit being an ER nurse

Anita Lambert worked as a nurse in a Chicago emergency room (ER). She first believed that people must solve their own problems. Her difficult regulars changed that view. She learned that one patient carried grief and hardship. Over time, understaffing and paperwork wore down her compassion. During COVID-19, hospitals gave staff little real support. After she gave a patient a positive test result, he attacked her, and she quit that day.

Marginal Revolution

Can Artificial Intelligence Forecast Treatment Effects in Economics?

A paper by Eva Vivalt tests if large language models (LLMs) can predict economic results. The models predict treatment effects for 33 studies. The studies were pre-registered before their results were published. The best model has a median error of 0.054 standard deviations (SD). The predictions beat two benchmarks: a zero effect and the mean effect of other studies.

Humanities

Aeon

In praise of the dilettante

Jay Miller defends the dilettante, a figure that many people see as a shallow dabbler. Miller argues that a culture of specialisation has damaged the dilettante's reputation. He says that the dilettante's movement between interests is a valid way of life. Dilettantes move between many fields, such as printmaking and powerlifting. Miller admits that this life brings little public recognition and is not for everyone.

JSTOR Daily

London's Railway for the Dead

By the mid-1800s, London churchyards were crowded, and cholera outbreaks made burial a health risk. An 1850 law banned new burials in churchyards. A private company bought land outside London and opened Brookwood Cemetery in 1854. A dedicated railway carried coffins and mourners to the cemetery. The Bishop of London opposed this commercial model. The model followed the garden cemetery movement, which treated cemeteries as places of rest.