The economics of a newsroom that runs on a hundred dollars a month
Automation changed the cost structure of news. We break down what a lean, AI-assisted newsroom actually spends — and what money can't buy.
There is a number in our internal dashboard that would have been unthinkable to a 1990s wire-service manager: the total monthly bill for reading, classifying, clustering, writing and translating our entire output. It is capped — by policy and by code — at one hundred dollars.
Where the money actually goes
Roughly speaking: classification and translation are the expensive habits, embeddings are nearly free, and the writing itself costs less than the coffee consumed while reviewing it. Every model call is logged with its price; when spending approaches the cap, the pipeline slows itself down rather than asking anyone''s permission.
What the cap buys us editorially
Constraints are editorial decisions in disguise. A hard budget forces the system to be selective — to rank stories and spend tokens on the five that matter rather than summarising everything that moved. The cap is not austerity; it is the ranking function''s sharpest critic.
What it does not buy
Original reporting. No language model attends a council meeting or cultivates a source. That layer — the one you are reading right now — still costs what it always cost: someone''s time and judgement. The hundred dollars buys the reading; the writing worth trusting still has a human price.