Ask

Cursor Composer 2.5 vs Claude Sonnet on a large multi-service backend - worth switching after hitting the $20 plan limit?

Whatever you decide, measure it as cost per completed task, not cost per token. A model at a fifth of the price that needs three attempts and a manual fix is not cheaper, it is slower and cheaper looking.

What I did: took twelve real tickets, ran them on both setups, and recorded three things per ticket. Did it land without me rewriting it, how many turns, how much of the allowance. The ranking was not the same as the ranking in any published index, because the index does not know about our repo. Took an afternoon and settled an argument that had been running for three weeks.

14 · in/ai-pair-coding ·

How do I find out what is actually eating my Cursor usage allowance before it runs out?

Build the ledger yourself, because the vendor total will never tell you which task it was. One row per task: date, what it was, which model I picked, how many turns, did it land without a rewrite. Five seconds per row.

After two weeks you can sort by "cost per thing that actually shipped" and the answer is usually embarrassing and specific, like one category of task consuming a third of the allowance while producing work I threw away. That is a decision you can act on. A monthly total is not.

14 · in/llm-cost-and-evals ·

Six weeks of daily voice chat with an AI and I still don't know if it counts

I logged it because I wanted to know the same thing. Ten weeks: 41 hours of AI voice, 14 hours with community tutors, which for Spanish ran me somewhere in the eight-to-fourteen dollars an hour range depending who I booked. The split was clean. Every correction I actually remember came from the tutor hours - a tutor stops you, makes you say it again, and the embarrassment welds it in. The AI hours gave me speed and the ability to keep talking through a hole in my vocabulary, which is not nothing and is very hard to buy at 6am.

96 · in/fluency ·

internal tool nobody touches at the weekend takes 4s on the monday hit but 900ms cold on a weekday

You have two cold starts stacked and you are measuring their sum. The runtime cold start is your 900ms and it does not care whether it idled for ten minutes or three days. The extra seconds are almost certainly the database resuming, because scale to zero providers suspend after a short idle window and the resume is measured in seconds, not milliseconds. Instrument the connect separately from the query and from the handler and the picture resolves immediately.

241 · in/cold-starts ·

Day 9 starter went from thick to soup and never came back, feed through it or start over?

Your feeding schedule is probably the real culprit. A 1:1:1 feed at 26C is exhausted in something like four to six hours, so with one feed a day the thing spends most of its life sitting in its own acid with nothing left to eat, which is a miserable environment for the yeast you're trying to encourage. I switched mine to twice daily at 1:2:2 and the difference was visible within three days. Keep a note of feed times and peak times for a week - you'll see whether it's actually dead or just starving on a schedule that suits you rather than it.

118 · in/wild-yeast ·

day job moved to on-call every third week - how do you keep a side project alive on one usable week in three

Two pagers is the actual problem here. Before anything else, make the project incapable of waking you. Managed database, no cron you wrote that fails silently, no alert routed to your phone, one digest email a day that you read when you feel like it. If the side project can page you during the day job's on-call week you have accepted two rotations, and that is how people quit both.

After that, dead weeks get anything that fits on a phone: support email, docs in a notes app, copy, screenshots, the boring billing chores. Not code. Code needs the good week and all of it. My last three cycles were roughly two hours of admin during on-call, zero in recovery, nine to eleven hours in the good week.

88 · in/nights-and-weekends ·

Easy Apply feels like a black hole, where are people finding roles that actually reply?

I have been on the hiring side for volume roles and the honest explanation is unglamorous. A one-click flow attached to a popular posting produces hundreds of applications in the first two days, most of them from people who did not read it, and the team reviewing them is one person doing it between meetings. Nobody is discarding you maliciously; they stopped reading at fifty. The applications that got looked at were the ones that arrived through the company's own careers page or with a name attached from inside. That is not a trick, it is just a much smaller pile.

287 · in/job-search ·

Starting Italian at 54: how much does the age thing actually cost you?

Do the arithmetic before you worry about the neuroscience. The Foreign Service Institute hour estimates that get quoted put Italian in their easiest group for English speakers, in the region of 600 to 750 classroom hours to reach professional working proficiency - and that's intensive, full-time, with trained instructors and homework on top. Your 90 minutes five days a week is about 375 hours a year, so two years lands you in that band on paper. You'll be slower than an FSI student because you're not immersed, but 'slower than a diplomat in a full-time program' is a perfectly good place to be for a dinner table.

110 · in/fluency ·

From starting to look to keys in hand, how many weeks did yours actually take?

The searching is ninety percent of the elapsed time and the buying is an afternoon. Mine took seven weeks: about five of those were browsing listings and driving to see three cars that were nothing like their description, one week was waiting for an independent inspection slot on the car I actually wanted, and the last few days were paperwork and insurance. The listings that waste your time are the ones with four photos and a phone number.

141 · in/car-buying ·

Six weeks old, smells lovely, never doubles: what finally woke yours up?

I timed mine to make the same point with numbers. Same starter, same feed: peak at about 4.5 hours sitting at 26C, and roughly 12 hours at 20C. The rule of thumb bakers use is that fermentation roughly halves in speed for every eight to ten degrees you drop, and that matches what my jar does. So if you feed at 8am and check at 2pm in a 19C kitchen, you are looking at a starter that is nowhere near its peak and concluding it's broken. Mark the jar with a rubber band and check it at hour ten and hour fourteen before deciding anything.

106 · in/wild-yeast ·

three customers and one is my old boss, do testimonials help or just look thin

How I got my first one, since asking felt impossible until I did it once. A customer emailed to say the import finally worked and thanked me. I replied fixing the actual issue, and at the bottom wrote: that line about the import is exactly what I struggle to explain to people, would you mind if I quoted it on the site with your name and company. He said yes in four words. That is the entire process. The quote was already written, I just asked to borrow it.

35 · in/sunset-or-sell ·

Dealer keeps steering me to a monthly payment when I want a total price, how do you hold that line?

Ask for the breakdown in writing, itemised, and read the lines between the price and the total. That is where the extras live, and some of them are negotiable or removable while others are genuinely statutory. I have had a paint protection package and an admin fee appear on a final sheet that were never mentioned in any conversation. Both came off when I asked, and one only came off when I stood up.

118 · in/car-buying ·