One big issue is prompt caching - how are you (or any other router) solving for it? The moment caching is gone real agentic workloads would spike like crazy in terms of cost.
1,589 karma · joined November 1, 2011
Contact quan1um dot l3ap at gmail
Based on the env vars seems like they're using anthropic, openai, etc. only?
So I spoke to 15 (remote-friendly/first) companies through my network (could've been more if I were up for onsite work), 4 started the interview process (rest had hiring freeze or wanted to prioritize folks that got laid off from FAANG) out of which:
- 1 ghosted me in the last round
- 1 rejected in the last round without any reason/feedback (although I think I did pretty well)
- 1 rejected in the first round because they wanted me to stick to a specific programming language (in their leetcode round) that I am now rusty with
- 1 rejected me in the last round on fair grounds and were decent enough to give feedback. I believe I made mistakes out of being super desperate to land a job.
One thing I noticed (although small sample set) was that the interview processes are very slow (like 1 round per week). This might mean things like companies are in no hurry to hire and they have abundance of candidates applying for the same positions at different "expected CTCs" making it a buyer's market. I am currently interviewing another company who has scheduled the next round to happen after 2 weeks.
Also no inbounds on Linkedin from recruiters. I used to be swamped till like July/August. Probably means there's decent amount of hiring freeze and most of the recruiters are fired?
Tbh, even though I got rejected in a few places, I did not feel too bad because I did not "feel" like they were doing solid engineering work (I ask lots of questions about the tech stack, challenges, culture, etc. in each round).
I've decided to spend a few months upskilling myself till the situation gets better and take the hiring process slow (rather being desperate). Although lucky to be in a position to be able to do so.
I did try to work around slow network speed (can see in the network_speed_optim branch on github but with the current APIs it's a tough problem to crack.
The current implementation is basic but works well. Suggestions welcome so that I can improve it.
- Custom domains would be nice to have.
- Connecting and pulling info from social presence (twitter, linkedin, etc.) would be a great addition (something like coderwall does?).
- Choose the header image/background from some settings area with a list of stock options. Currently it shows some default one that can't be changed.
With these many of us wouldn't mind paying $x/month, yeah :)