I have exported the Search Console coverage list for my own site every day since 28 July. 46 reports across 45 days, 277 URLs in the sitemap.
Fourteen URLs have carried the verdict "Crawled, currently not indexed" at some point in that window. Twelve of them left it. Two are still there, and one of those two has been on every single report since the first one.
I have done client work for years, and it pays the bills. My own product keeps losing the fight for time, because a client deadline always feels more urgent than something only I care about.
Things I have heard or tried: fixed days for each, a protected block every morning before client messages, or taking fewer clients on purpose. Each one breaks at some point, usually in a busy month.
Tech giants are buying up massive energy sources just to power their AI models, which makes me think compute might not stay cheap forever.
Everyone assumes token costs will keep dropping, but if energy gets tight, API prices could easily plateau or go up. Started tracking cost per user query in my own app and already caught a feature that costs way too much for what it actually gives.
Do you guys budget for rising AI costs or are you assuming model prices will only keep getting cheaper?
A note before we start: What follows is fiction. A thought experiment set in an alternate world that only looks like ours by accident. Any resemblance to real labs, real bosses, or real workplace policies is the reader s imagination doing the heavy lifting. With that out of the way.
For anyone who processes supplier invoices for a living: the part that eats the evening isn't understanding the invoice, it's retyping it. Same fields, ten different layouts, and the totals don't always match the lines. Two rules I ended up with: 1) Never let the tool fill a field it isn't sure about. A guessed total looks exactly like a correct total until the accountant finds it three weeks later. An empty field with a flag costs ten seconds. 2) The biggest number on a receipt isn't necessarily the total. I once saw a reading come back with 190.00 while the register had printed 832.50 190 was one line item with a discount. The fiscal QR code holds the total the register itself signed. Curious how others do it: leave the field blank and flag it, let it guess and fix later, or block the document until a human confirms? And do you reconcile the total against something independent, or trust the extraction?
I ve lost count of how many habit tracking apps I ve installed and eventually deleted over the last few months.
They usually start off feeling super motivating for the first fortnight. But as soon as life gets busy and I inevitably break a 10-day streak the whole experience turns into a guilt trip instead of a helpful tool. The constant daily reminders just end up feeling like spam notifications.
Lately, I ve been thinking simple physical journals might actually work better than phone alerts.
How do you keep track of your daily goals without relying on gamified apps?
English is not my main language, so bear with me :) I used to be a daily user of PH. Not as a contributor, but a lurker. I loved to discover new apps, services, ideas... However, after a couple of years of not using the platform, I returned only to find that an overwhelming majority of the products are just... AI.
Not necessarily saying I'm against AI products. That's a debate in itself, and not the point I'm trying to make. I'm just pointing out that there doesn't seem to be a lot of products or ideas around that:
Text LLM serving is a banquet kitchen: you know the menu early, can batch requests, chunk prefill, and reuse common context through KV caches.
Streaming voice is a roadside dosa stall.
The user changes their order mid-sentence. VAD is separating speech from traffic. STT is revising words live. TTS needs to start speaking in ~50ms or the conversation feels broken.
