Annem Sabah

✨👩🏽‍💻 AI & Forward Deployed Engineering | @AWS Community Builder ☁️ | All views my own. | Tech only here folks. 🫡

[Annem Sabah @annemthecoder]

6am.

[Gergely Orosz @GergelyOrosz]

I’ve been saying this. Leetcode was hated, and it will die sooner rather than later. What will come after is no standard interview, more subjectivity, and no definite preparation advice

Eng hiring will turn into the type of hiring PMs, designers etc know all too well!

[WookieeNews @WookieeNews]

Hace 28 años se lanzaba el póster más icónico de la historia.

[mr pussy @mrpxssy]

wake up babe, every version of PHP for the past 17 years has two critical bugs and now 74.5% of the internet is broken lol. there's a patch coming out tomorrow but until then most of the internet has open backdoors that allow unauthorized remote code execution

[Annem Sabah @annemthecoder]

@VoltForTesla @Tesla Yeah, I've thought the efficiency wasn't great but it should settle right.. with the weather getting cold maybe not!

[Annem Sabah @annemthecoder]

A month into @Tesla ownership and (don’t judge) I’m so sold. One-pedal driving is absolutely a thing I can get used to.

Not sure about the efficiency yet, but it’ll settle into an average after a few months. Finally sprung for a charger so it's time to look at tarrifs.

[Thāni @thanipro]

When you interview at a company like Revolut, some egotistical engineers will mark you down for not diving deep enough into multi-region architecture and system resilience. But look at what happened today.

[Annem Sabah @annemthecoder]

Your girl got into AI, ramblings incoming. 👇

https://t.co/IdhhewQ7is

[timo @timo_rf]

leetcoding at 33 years old is a humilation ritual

[Annem Sabah @annemthecoder]

Why I don’t believe in ‘managed ci/cd’ tools anymore. Why pay $$ for an expensive per-user license when any coding agent can write the scripts/yml for a setting up build pipelines?

[Roy @__roycohen]

Shoutout to everyone who did CI/CD before AI.

[tyler hogge @thogge]

OpenAI has signed now.

What a fascinating day of game theory.

[Annem Sabah @annemthecoder]

One from a London AI Meetup last month ~ 💰 Context tax is real: think about whether your repo needs the context of 10 different repos and microservices. Feeding huge amounts of code/logs to an agent depletes quickly tokens because input tokens are processed on every single step

[Zuhayeer Musa @zuhayeer]

The bifurcation of software engineer pay, AI engineers vs everyone else

[Zara Zhang @zarazhangrui]

If I were to hire today, this is how I would structure the interview process:

Round 1: in-person & no AI allowed; test domain expertise and knowledge on the fly

Round 2: a project that must be done with AI (impossible to complete without AI). Candidate assessed not just on the result but also the chat transcript with agents

[Jeremy Bernier @jeremybernier]

Failed a coding interview at prominent tech company.

Question: Code a load balancer. It should distribute sessions evenly across servers. Can assume fixed # of servers.

I first implemented round robin. Then I discussed possibility of using a heap to get the server with the least # of sessions in O(1) time, but I didn't know how to efficiently handle the case when a session ends since heap deletion is O(N). I mentioned hashing, but they wanted a different solution with stronger guarantees. I went back to the heap, but didn't know how to improve on O(N) deletion, and I could sense the interviewer didn't seem to care for the heap approach.

I was stuck, and at one point just flat out told the interviewer so. Normally in my experience interviewers actually do provide some guidance when you're genuinely stuck because collaboration itself is a signal most companies grade you on. This one told me to re-examine the requirements / invariants, but I wasn't getting anywhere.

With only 10 minutes remaining, they told me I could code my heap solution to "have that written down". I knew that if I did that I'd run out of time and they'd fail me because it obviously wasn't the solution they were looking for, so I tried to persist on finding the more performant solution.

Eventually I realized they wanted a max difference of 1 session across servers. I asked if we could transfer sessions across servers (I'd mistakenly assumed we couldn't thinking that they'd correct me if my assumption was wrong), then came up with the two bucket solution they were looking for. Only had time to talk through the different cases rather than implement.

Afterwards checked the question on ChatGPT and looks like you can do lazy deletion with heaps.

Learnings
- Heaps support lazy deletion
- Two bucket approach when delta <= 1
- Don't say "I'm assuming X" and assume your interviewer will correct you if you're making an incorrect assumption about the problem constraints. Ask directly.

I do think the interviewer was indexing too much on the specific esoteric solution they were looking for, but obviously sour grapes. If I ever get this question again though, I will definitely pass (which means that these questions probably aren't the best signal).

[Annem Sabah @annemthecoder]

3 pieces of advice from the last decade that has stuck with me...

When making decisions, pick the harder lane.

Detachment. Remove emotion from any decision

Your first job doesn’t define, and neither does anyone else’s trajectory.

#turning30 🎂

[Nizzy @nizzyabi]

oh my god

[Darth Thunder ⚡ @XDarthThunderX]

This moment will forever be magical.

[Pakistan @Indusland_]

St. Mathew's Church. Nathiagali, Pakistan.

[Vic 📠 @VicVijayakumar]

🤨