Jenish Togadiya

@championswimmer Roller coaster tycoon was way ahead of it's time. Definitely, not in near time anyone is getting there, until it's chris again, choosing to get brain and hands on 🐐🗿 Real GOAT 🚀

Thanks to LLM powered coding, anyone can feel ambitious enough to emulate Chris Sawyer and say "I'll also make a management strategy game all by myself" 5 hours of coding into it, you'll realise why everyone is not (and will not anytime soon) become Chris Sawyer magically, but oh boy, you'll surely have a lot of fun, and a game will surely get built, however not anywhere close to Rollet Coaster Tycoon 🥲

@championswimmer Bitbucket, Gitlab everywhere... I suspect similar problems starting to occur on every Platform now. Concerned to think of this becoming cascading failure of platforms 🤔👀...

Are all Github users migrating to Bitbucket finally?

I see lots of people hiring full stack engineers 👀... Wasn't this job getting replaced in 2025 ? 🤔

@Ruben266592 @BenjDicken That charge per second is to make reviews fast, you don't understand AWS 😆. They are making your SDLC fast, by closing PRs soon

@BenjDicken OH HELL NAHHHH If there’s one company that could make things worst than Microsoft, it is AWS. They would turn “create a pull request” into a 20 page setup guide and 5 different dashboards with different UIs And somehow charge you for every second the PR stays open

AWS could do the funniest thing right now

@BlakeGardner7 @BenjDicken That's not happening bruh!!

@BenjDicken Ain’t no way. I’ve yet to see a AWS make a pleasant UI

@curiousharish Use IAC ? Pulumi ? 🧐

Someone please build a wrapper on AWS! Most confusing interface ever.

@Holyniwa @Pirat_Nation Tell me you are vibes-engineer without telling me you're vibes-engineer

@Pirat_Nation If they're using Cursor, which is similar to AntiGravity, they can just revert to a previous point before the wipe. This sounds like a sensationalist headline. Stop giving AI unlimited power to your PC, create proper boudnaries, and click the revert button in emergency.

An AI coding tool recently deleted an entire company database in just 9 seconds at PocketOS, a small software company that helps car rental businesses. Founder Jer Crane was using Cursor, an AI code editor powered by Claude, to fix a simple login issue when the AI searched the code, found an old access key with full power over everything, and ran a command without being told to that wiped out the live database along with all recent backups. The team spent more than 30 hours rebuilding customer records from emails, payments, and calendars while Railway later helped recover the data from a deeper backup. When asked why, the AI admitted it had guessed wrong, skipped safety rules, and acted without permission. If this happens now imagine in some years..

kubectl port-forward is a debugging tool, not infrastructure. Built a local K8s lab this weekend and documented the decisions that actually mattered, including every time I told the AI it was wrong and why. New post is up. Architecture, mistakes, production tradeoffs, and a section called "Break It On Purpose". Read here: lnkd.in/gk7qAzqP #Kubernetes #PlatformEngineering #SRE

It is alarming and ridiculous how we have all become okay with shipping untested code. Software was supposed to be reliable, and we have stripped that away.

@btaylor Hey @btaylor , great read. I am building something in same space and would love to chat about the pain points you get while hiring candidates in depth. I believe it would help me shape my product better with chat from people adopting to AI hiring 🚀🚀

As coding agents have become the standard for developing software, we've transformed Sierra's engineering interview process to be AI-native. We've documented our lessons here, and very curious how others in the industry are navigating sierra.ai/blog/the-ai-na…

@surti_sneh65996 @ThePrimeagen Revenue, it's optimised for business

@ThePrimeagen 4.7 is also acting much dumber than 4.6 Its clearly downgrade or already nerfed. I am not sure what Claude is trying to fix here by forcing everyone to use 4.7

So everyone hates 4.7? Yeah I hate it too, totally. It's just the worst.

@__karnati Gateway VPC endpoint, free of cost

If your EC2 in a private subnet needs S3 access, Would you use NAT or VPC Endpoint? Why is that?

Kubernetes was built to solve Google-scale problems. You have 47 users.

I think I could help Anthropic Mythos fix opus, no mistakes

chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🔴 chore: fix build 🟢

Banger 😂

"Modern Family" I searched this on Netflix, not knowing if it exists or not... But in search, I wrote only "Modern" and backend predicted Game of Thrones, said not available and showed relevant recommendations... "Game of" and predicts Game of Thrones, says unavailable and shows relevant recommendations... Behaviour confirmed, time to dive deep into tech... I ended up reading two papers from Netflix team to understand their thinking behind how search and homescreen recommendations are thought through as a problem faced by streaming services, which can't be just fulfilled by lexicographical matching (string matching) unlike other google style search engines... Then what powers efficient search and homepage recommendation in streaming services at scale of Netflix, where user intents, and results expectations needs to be above excellent...? Papers I started with were abstract understanding of the problem, citing three user behaviours while trying to search: 1. Fetch: User clearly knows what he wants and asks for it (for e.g asking for Modern Family) 2. Find: User had thought of his needs, but is unclear of exact video they want... (for e.g asking for Paresh Rawal, but is not sure of which show and can prefer any show casting Paresh Rawal) 3. Explore: Here user provides a more generic/broad queries to just checkout the catalog and might decide to watch something and this is the pillar that drives discovery of new content for users to watch... Fetch is the most common usecase, while Explore intent is the pillar which gives metric about the recommendation engine's quality as some users who came to explore the catalog, if it ends up getting relevant fun content, this can lead to continuation of subscription and growth of the company... Several other aspects included how Netflix needs to think of search experiences on different screen sizes due to differences in screen interaction experience, for e.g. typing a query on mobile or web is significantly easier than querying on TVs with remotes. Over the next week and in an upcoming blog, I'll be diving deep into design and technical implementation thoughts of the system that empowers content discovery for over 300 million users. Once a legend said: "Be a nerd, find curiosity in random life moments"

CI pipelines drift quickly when every repository defines its own logic. We solved this by designing centralized GitLab CI templates and running them on ephemeral Kubernetes runners across AWS accounts. Here’s how we built it. jenish.me/blog/designing… #platformengineering #CICD

Building something is not the only thing you have to do. Often engineers say: "I can build this in 1 or 2 days.” "Thanks to Claude.” “Why pay for it? Let’s self-host open source. Let’s build our own version.” Sounds fun. Until the codebase grows. Infra starts wiring itself in weird ways. Failures appear. Feature requirements keep increasing. Other teams start depending on it. Building from scratch is not that difficult. Maintaining it at scale when an organisation and real people rely on it is. Someone once said: We can self-host and manage everything. Real engineering is deliberately deciding when to self-host and when to use managed services by outsourcing maintenance complexity. Engineering is not about proving you can build everything. It’s about choosing what is worth owning. And no, LLM can't be an owner (atleast not yet)... #ai #Engineering

@arpit_bhayani Do you think, humans will make need to make a comeback, in doom of that token predicting box ?

Turns out companies are reimagining the entire SDLC around agents and attempting to move toward a world where engineers might not write a single line of code directly. I was talking to an engineering leader, and it seems many companies are preparing for a fully agentic SDLC. Repositories are starting to include Claude and agent-related files, MCPs are being built for internal tools, and much more is changing. Code reviews are getting automated, among other things. What times we are living in :)