Everyone poked at Jev, levelsio made $3,240 on hotels
A demo posted today claims a browser agent found flights in 7 seconds for $0.0039, and by evening half the timeline was running a new small model called Jev on something. Meanwhile levelsio reports his hotel site made $3,240 in two weeks from actual bookings, not tokens.
The big story
Josh Pigford ran over 10,000 Hacker News comments through Jev, says 43% were corrections
Jev is a new model that spent a day going around founder timelines. Josh Pigford used it to analyze over 10,000 Hacker News comments, reported that 43% of them were correcting someone, and then shipped a "well, actually" leaderboard. Yohei tried it for graph extraction, scoring every word for semantic significance and building a graph from the high scorers, and says it needs some cleaning but is not bad. The real story: nobody in the sources has run it in production, and the interest is not about another chat model, it is about the boring middle of a product. Aaron Levie's list is the tell: classification, routing decisions, quick judgment calls on safety. Shreya Shankar's point is the old business truth underneath, people reach for AI functions because they do not want to train and maintain their own classifier.
i had jev analyze over 10,000 hacker news comments. 43% of them were correcting someone.
Being able to process information insanely quickly, at crazy low costs, with high levels of capability is huge for a wide number of enterprise tasks.
AI functions are popular in databases because no one wants to do the MLOps work of training their own classifier
Real numbers
levelsio says Hotelist made $3,240 in its first two weeks of bookings
levelsio posted his first hotel booking revenue: $3,240 in two weeks, which he annualizes to roughly $6,480 a month, and says it is much more than he expected. He is explicit that the appeal is not being another AI wrapper. He also tempered it in the replies: affiliate revenue has almost never worked for him before, and he thinks hotels are different only because the value per sale is higher. He signed up for Stay22 to boost commissions, says it rewrote his affiliate tags to the wrong hotels, and deleted it the same day. Traffic, he says, is mostly X and Google.
$3,240 in first 2 weeks So about $6,480/mo Much more than I expected!
Very very happy to make money with a project that's not reselling AI tokens again!
It almost never made me money, with hotels it might work because the value of sale is much higher
Tibo says Outrank does $300k+ MRR in a market people called dead
Tibo says he started an SEO tool two years ago while the narrative was "SEO is dead", built it with a Ukrainian co-maker he had never met in person, and Outrank now does $300k+ MRR. He points at the rest of the category as evidence: Semrush sold to Adobe for $1.9B, Ahrefs at roughly $150M ARR with zero investors, and Profound raising $180M at a $1.8B valuation this week. In a reply he says they are accepting to shrink margins a bit with a launch meant to provide higher-quality backlinks. Josh Pigford made the unglamorous version of the point: it is exceedingly rare that one launch removes every other similar business's reason to operate.
today Outrank does $300k+ MRR
if a market is crowded, it means it's validated
multiple businesses can (and should!) exist solving the same problem.
Quick hits
Claude Projects now runs parallel threads after you close the laptop
Anthropic began rolling out Projects in Claude Code, where one conversation splits itself into parallel cloud sessions that keep working when you leave. Boris Cherny says he has stopped managing sessions entirely and now does a ton of his coding this way.
dax adds spend limits after Astra usage, warns about revenue per employee
dax says Astra finally pushed him to implement limits on OpenCode, and followed it with a blunt line about token-hungry products needing high revenue per employee. He also described the cost strategy: when a new model ships they eat the price until they can host it themselves or find a cheaper partner.
Jason Fried hired someone who never applied and was never recruited
Rob Zolkos kept building on 37signals products uninvited: exploring Fizzy, writing up how to use its webhooks, then starting a CLI once the API shipped. Fried says they were not hiring, hired him as a contractor first, and made him a co-worker today.
Pe:p Laja says 88% of surveyed marketing leaders research vendors with AI
Pe:p Laja surveyed 101 CMOs and VPs of marketing and says 88% use AI recommendations to discover and research vendors, which makes "contact sales for pricing" mostly self-harm. In his write-up he cites Kyle Poyar running pricing prompts across six AI engines for the Cloud 100, where a company's own pricing page was the first source in about one in eight answers and the rest came from Vendr, Reddit, G2 and teardown sites.