I just published a deep-dive into the 250-hour build behind syntaqlite, a SQLite formatter and LSP I built using AI agents.
AI agents were the only reason built this after 8 years of wanting but there's a psychological toll to AI-assisted engineering.
The post-mortem:
lalitm.com/post/buildin...
Really hopeful that the Lundbeck PACAP med is successful in phase III #migraine #longcovid
Looks liked phase IIb is going well.
www.clinicaltrialsarena.com/news/lundbec...
1) This new paper in the Lancet suggests that the risk of Long Covid has substantially decreased compared to the pandemic period. The excess prevalence of severe symptoms self-attributed to SARS-CoV-2 was only 0.3% after 180 days.
A brief breakdown...
This is the only way PPR and the newer use cache actually work well I’ve found.
Basically nothing can be blocking and memoization is the way around it. Makes a lot of classic auth provider patterns not work, you need to get the status async at the leaf level. Doable, but, makes everything async…
Claude code gets stuck on interactive prompts all the time. It hates drizzle ORM in dev mode! Playwright HTML reporter puts it in a frozen state too.
Constant entertainment. 6 am to 8 pm. Small breaks for TV here and there.
Been following this one. Was maybe going to try and get in phase 3. I take vyepti and it helps a lot but not as much as I’d like.
I’ve read they plan to combine this with their CGRP at some point too since CGRP and PACAP are active at different times/places in a migraine.
Ralph-wiggum
The difference is it’s changing fast, will change again, and in the meantime people using all these can code laps around other devs in many cases. The velocity is insane.
The reality is it’s learning to understand how to work with LLMs, it’s becoming a language expert.
I asked this same question at the next.js conference a they said yes for, use cache:remote at least.
Interesting! I have it caching with dynamic parameters in my very recently released cache-components redis handler. I didn’t even try the behavior in vercel.
github.com/mrjasonroy/c...
Makes sense. Curious though, are you wrapping those random things in suspense and they are using “use”? Is that where the promises resolve?
The model wants you to move that logged in state fetch further down the tree, wherever it is needed. It makes you ask “does the root provider need user login?”
Nexjts Cache Components ("use cache") cache handler for those of us whole self host.
Existing cache handlers don't support nextjs16 cache component patterns
Supports redis, valkey, elasticache and memory
Essential for k8s/pods/clusters
github.com/mrjasonroy/c...
How much money wasted??? No fault on the injury but it wasn’t like it was going to go well unless things were perfect.
In general if it doesn’t run on node honestly I’ll probably skip it unless it’s business critical.
Lots of junior devs in “far off lands” insulated from social interaction and influenced by the flavor of the day because they don’t have experience building things or working with people.
Just maintaining their local plumber and church websites for clients that hate them.
Ha. I just have very similar experience of people I have known - crust punks, ar-15 kind of racist conservatives, weed farmers, crystal healers and off the land folk all converging together to have horrible, horrible converging opinions.
Hmm, I haven’t…. That’s a good idea though. I’ll see what I can setup. We use cloudflare zerotrust so I’ll see what is possible.
I imagine this is an issue for lots of people behind private network APIs not connected to the internet (sometimes behind AWS SSO sessions too)
But none of this works if you need to access things over a VPN or through tunnels… I can’t get web versions to make the requests to internal dev URLs secured through short lived VPN connections. I never fully yolo anyways but get close, letting it modify vault secrets even. Too cumbersome!
It won’t be an alpha/beta feature after Wednesday.
Yep! It’s the return of the PMs and well defined tickets :)
You learn the language over time. Simple, well directed instructions with specific directions where to start an learn from help a lot.
It will change code. We’ll end up with a lot of markdown for LLMs colocated with files.
You know who hates AI coding? Influencers who make videos teaching coding tricks.
Videos that are long, hard to navigate, and made to sell ads or controversy.
The video might be useful but the lesson could have been a git gist.
AI just does that part for you and you learn it in a few minutes.
New Antlers is great! First full length in a while. Peter’s voice and songwriting are as beautiful as always.
music.apple.com/us/album/som...
Exercising fathers, enhanced endurance and metabolic heath in offspring, through sperm microRNAs (as demonstrated in mice)
www.cell.com/cell-metabol...
Let me guess, his family history in Argentina goes back to about 1945…
I’m around 60. It only works well with tests… it is better at writing unit than E2E or integration tests but that makes sense.
I do wonder why affect it will have on languages things are written in. Which are better for AI? Typing seems essential, and popularity of the language is important.
I love that Claude code is horrible at Helm charts and K8s. Just horribly complex design decisions that our LLM buddies make a mess of too.
For us, we have a React Context provider that includes "use" and "useOptimistic." fireEvent and act sometimes caused render errors, though they may be fixed now. We had to use a lot of tricks in react-testing-library to get React not to complain or tests to fail, see my response to @kentcdodds.com.
