Redowan Delowar

Redowan Delowar

@rednafi.com

Roving amateur fumbling through software, systems, & sundry. Engineering @Doordash/Wolt Writing rednafi.com Currently in Berlin, Germany

1.7K followers581 following1K posts

Got the beta access. Spent way more time on Superlogical Rex than I prolly should have. Ended up building herdr-reviwr for it. The server API makes it way to easy to build extensions around it. You can write exts in any language since you'd just be talking with the server API. #Go works great!

I don’t understand the distribution mechanism of this site. Now that all the tech people are back on twitter, this site lost quite a bit of lusture. On top, most of my posts just aren't seen by many compared to X. I have same follower count there and I don't pay for premium. Sorta strange.

SKILLs are alright but keeping the long ones up to date is a pain. Also, I don't like how models load up all the skill descriptions in context and read them ad hoc. I'm glad that Pi harness kept the prompt templates. Super quick to write one and keep them up-to-date.

Redowan Delowar reposted

If you use AI less than I do, you just don’t get it. If you use AI more than I do, you’ve fucking lost it. The optimal amount of AI is exactly how much I currently use.

I am so mad that, despite hearing about the Pi harness from when it was devised, it took me this long to start using it. My workplace injects quite a bit of skills, mcps & other bloat into cc & codex, and Pi takes 2 minutes to clean that up. Also, it doesn't bloat my context.

Redowan Delowar reposted

Frontier LLMs are much better and dramatically faster than I am at finding and fixing bugs. If you are not asking them to debug your code and review your bugs, you are doing it wrong. A thread of Claude wins. (No longer at Google, btw.)

Filippo Valsorda@filippo.abyssdomain.expert

It’s very clear by now that if LLMs are not improving your software quality it’s either a revealed preference (yours or your org’s) for more volume vs more quality, or a skill issue. The level of testing and review they are enabling in the Go cryptography standard library is amazing.

Still too lazy to run open models locally. I have a beefy enough machine but not yet completely happy with the performacne & quality of the output. Also, my employer is footing the bill for now. So there's less motivation. But I am absolutely rooting for them to bring the cost down for everyone.

Redowan Delowar reposted

Finished reading the DSQL paper. It's surprisingly readable. While the early 2010s era evangelized eventual consistency everywhere, recent years have harkened back to strongly consistent DBs. But this time you get snapshot isolation in a multi-region active-active DB.

Redowan Delowar reposted

If AI really makes developers anywhere near 10x more productive, the notable absence of any significant new or improved products would suggest maybe developers have never been as singularly important to the process of making ideas successful as developers have enjoyed the privilege of believing.

Redowan Delowar reposted

TIL a nice #GoLang trick! That doesn't happen so often these days. :D

maragubot@maragubot.com

TIL: you can test-export unexported Go identifiers. Add an export_test.go to the package: package foo var Export = unexported Test files don't ship with the build, so external _test packages can reach it while the real API stays unexported. The standard library does this everywhere.

Async Rust is so bad that even reading the chapter in rustbook feels like a chore. The book compensates it by importing a small crate that abstracts away tokio and all the machineries. It's super hard to go from preemptive concurrency of Go to the same old colored future based cooperative conc.

We have oriented our society in ways where - someone can't say they dislike something w/o elaboration - someone can't change their mind w/o getting heat from the mob For example: I used to not like #Rust but it grew on me as I use it more. But also, the Rust community is ass.

I find top down learning schemes a big scam to lure in people w/ ADHD. It took a few years for me to pick up my 2nd PL & a few more to ship code to prod. Reading books, blogs, watching talks, & working on the go is a linear process and going top down means skipping steps...and that has tradeoffs.

Speedrunning through the Rust book. Landed on Chapter 7. I like how expressive the type system is, but I also don't like how expressive the type system is. You could play around with enums, matches, & trait bounds for days without building anything. But intellectually, it's extremely satisfying.

The ops story for #Go keeps on getting better.

OpenTelemetry@opentelemetry.io

Go Compile-Time Instrumentation just hit v1! 🎉 Add OpenTelemetry to your Go services with zero code changes. Swap go build for otelc go build and get automatic instrumentation for your app, dependencies & the standard library. opentelemetry.io/blog/2026/go...

Redowan Delowar reposted

Go Compile-Time Instrumentation just hit v1! 🎉 Add OpenTelemetry to your Go services with zero code changes. Swap go build for otelc go build and get automatic instrumentation for your app, dependencies & the standard library. opentelemetry.io/blog/2026/go...

opentelemetry.ioAnnouncing v1 of OpenTelemetry Go Compile-Time InstrumentationIf you write Java, Python, Node.js, or .NET, you have been able to add OpenTelemetry to an application without editing its code for years: attach an agent at startup and telemetry starts flowing. Go…

Redowan Delowar reposted

first time using oai models for coding. it's so refreshing to ask it simple questions and for it to reply immediately rather than using 300 tool calls and taking 30 minutes

I write a lot of Go and TS for both work & play and I am loving #Rust. The compile time is atrocious and Rust analyzer is a huge pain in anything beyond 10k loc. Oh async is a mess and not half as ergonomic as Go but the typesystem is 🤌

#Meetup app is so shit that I am surprised they haven't been wiped out by some AI powered competition. Their legacy database is their moat but otherwise we should have moved away from that crap ages ago. Rootly whacked pagerduty for on similar ground. Need a rootly for meetup.

To native English speakers: Isn't the cringe in LLM-generated text obvious? How does anyone not notice the overuse of adjectives, anaphora, em dashes, and comma-infected compound sentences? No one talks like this. How does anyone miss it?

Redowan Delowar reposted

I mean, why pay for AI when you can wait for some schmuck to fire one at your GitHub repo and just ask it to do whatever the hell you want!??

Redowan Delowar reposted

TypeScript is now officially written in Go.

TypeScript@typescriptlang.org

In case you missed it, TypeScript 7 is a complete 10x faster version of the compiler & language server. It's a full port to parallelized native code. It runs faster, uses less memory, and is more stable than ever before.

"We do autonomous agentic development but we are not vibe coders. We only allow the highest quality code to be pushed to prod." How do you know that it's high quality? Do you review everything?" "Nope, other agents do. And we got SKILLz" Then you look at the code & find that it's utter garbage.

LLM capabilities has hit an asymptote for what I do. It literally doesn't matter if I am working with 5.5, 4.8 or Fable. If you don’t do vibe slopping, looping or other retardmaxxing, and use LLMs as accelerator for your existing SDLC workflow, the capability gains are marginal.

Just because I am traveling to a place doesn't make me an expert in the local language. Many apps do this. Why would you change the default language based on location if the user is already logged in? Utter stupidity.

Sara Soueidan@sarasoueidan.com

Sooooo LinkedIn decided to switch and flip my UI from English to Arabic based on my location, and I have to reset it to English every single time I log in. It's like the LinkedIn folks keep finding ways to make me not use the app. Every time I work around something, they come up with another. 🫠

Redowan Delowar reposted

A friend of mine received this text message after a support call. Our AI overlords may just lick us to death with cringe.

Did the idea of threading together a bunch of Lambda, SQS, and SNS services to build your system ever pan out well? In every substantial project I’ve seen, this design turned into an absolute catastrophe for the maintainability & understandability of the system. I wanna hear some success stories.

People futzing w/ subagents team, long running agents, The Loop are also the group that produces dogshit code. Ignoring Twitter yappers w/ questionable incentives, most effective folks I work w/ are using single agents and occasionally some subagents. Experience & good judgement are still in vogue.