The revealing moment in Last Week in AI’s history arrives when its founder describes how much there used to be. Articles. A website. A newsletter. Eventually, a podcast. Andrey Kurenkov had set out to explain artificial intelligence to people beyond the research community. Then he encountered a problem familiar to anyone who has ever promised to publish regularly: every additional format needs another piece of somebody’s week.
- A researcher-led AI publication combining text briefings and audio discussion.
- Free content, paid subscriptions and sponsor placements support the operation.
- Long episodes come with timestamps, so listeners can choose their own route.
- Its history offers a practical lesson in reducing an editorial workload.
“It turns out doing all this is a lot,” he told the SuperDataScience podcast in 2024. The sentence has none of the glamour of a funding announcement. It explains something more useful. A publication’s ambitions can exceed the hours available to deliver them. Last Week in AI is interesting because its founder was willing to describe the arithmetic.
The operation that got smaller
Kurenkov’s initial irritation was with sensational AI coverage. He wanted explanations that made technical sense and reached a broader audience. Skynet Today launched in 2018; he dates its early newsletter to May that year. The digest later developed a separate identity, while the wider article operation wound down. The newsletter and podcast became the surviving focus.
That origin helps explain where this company sits. It is a media publication about AI, with a researcher’s perspective behind the selection. Its product is the act of choosing and explaining. Readers receive a route through a crowded subject: tools, research, corporate developments, safety and policy. They can use that route to decide what deserves further attention.
His personal biography describes doctoral work in Stanford’s Vision and Learning Lab, at the intersection of robotics and machine learning, before a role at game-creation startup Astrocade. Regular podcast co-host Jeremie Harris brings another perspective through his work at Gladstone AI. These are people with professional connections to the field they discuss, something listeners should understand when weighing their commentary.
A briefing with an escape hatch
The useful detail is in the episode notes. A long conversation becomes an indexed conversation. Topics are grouped, original material is linked, and timestamps mark the turns. A reader looking for a research paper can go directly to that discussion. Someone following a product launch can enter elsewhere. The table of contents does part of the work.
Consider the contrast between two published episodes. Number 221 runs 47 minutes. Number 258 runs one hour, 43 minutes and 29 seconds. “Weekly summary” therefore describes the job rather than a fixed serving size. On a busy week, the audio can become a considerable commitment. Navigation lets the listener decide how much of it to accept.
Across those notes, the subjects include tools and apps, applications and business, open-source projects, research, and policy. The mix matters. A model announcement tells one story; the surrounding evidence, commercial decisions and regulation tell others. Putting them together gives readers a chance to see connections that a feed of isolated launch posts makes easy to miss.
The comparison with other AI briefings is about use. A reader could choose The Batch, TheSequence or The Decoder, or follow another news podcast. Last Week in AI’s particular arrangement pairs a curated text product with a discussion you can hear or watch. It suits people who want an editorial entry point and enjoy listening to the reasoning around the news.
Free to start. Five dollars to support.
The commercial machinery is recognizable. There is free content, paid access through Substack, and sponsorship. Publicly configured dollar plans are $5 a month or $50 a year. Some posts are gated. The publication’s About page also invites sponsor inquiries and thanks founding-member supporters, keeping audience support visible alongside the regular editorial output.
Advertising is visible too. October 2026 posts carry sponsor placements for ODSC AI and Langfuse; podcast acknowledgments also name Box, Notion and Factor. Those are commercial relationships around the publication. A reader can see the messages and decide how to treat them. For someone studying small media businesses, the lesson is concrete: the same editorial beat can serve both subscribers and advertisers.
The weekly promise meets the calendar
Then comes the awkwardly informative update. On August 25, 2026, issue 342 announced the text newsletter’s return after a hiatus. Its subtitle was “The newsletter is finally back!” Kurenkov acknowledged the delay and his intention to resume it alongside the podcast. That announcement belongs in the story because it shows the operating constraint surviving the earlier reduction in scope.
“The newsletter is finally back!”Issue #342 / August 2026
The October 3 podcast post offers another small clue: episode 258 had been recorded on September 26 and arrived with a note about being late. Meanwhile, the official signup page lists over 186,000 subscribers. The audience and the publishing schedule are different measurements. Having a large readership does not automatically produce the time required to serve it.
By October 9, the text archive had reached issue 346. The publication was active again, covering announcements, research disputes and industry developments. The practical condition is clear: a weekly briefing works when a reader can tolerate some delay. Anyone needing immediate operational alerts will need another channel. Anyone making a technical decision must still inspect the underlying evidence.
Take one story further
A useful routine is modest. Scan the text edition. Choose a subject relevant to your work. Use an episode’s timestamps if you want the discussion, then open the linked paper or reporting. Treat the briefing as a beginning for investigation. It can help a curious newcomer find vocabulary and help a busy practitioner notice developments outside a narrow specialty.
For a publisher, the copyable lesson is equally modest: define a beat, make the material easy to navigate, and retire formats that exceed your capacity. Last Week in AI’s history makes those choices visible. The next issue still has to be made by somebody. That is a useful fact to remember while the rest of the industry discusses what machines will do next.