Cryptic Ben

@JEdwardWoody @scorete22 @TheWorthyHouse I believe you. You just need to be creative and combine systems, information, processes, and tools. Some know how to do that, others will never understand.

@scorete22 @TheWorthyHouse Some of u really dont want to believe it's possible to leverage new technology productively 🤔

I don't understand this idea that "AI" can "answer your emails." I sent more than 200,000 emails during the fifteen years I ran a business (and probably 150K more sent when I was a lawyer). Each one required my personal input, tacit knowledge that simply cannot be reduced to an algorithm, involving the balancing of hundreds or thousands of pieces of information internal to me. If I allowed "AI" access to every written piece of data about my business and my prior correspondence, it would not cover 10% of the necessary information to respond correctly to emails, much less the relative weightings and interactions of each of those pieces of data. Whose emails is "AI" supposed to answer? The new "AI" customer response bots are worse than the old ones. Seriously, what emails are we talking about?

@kapilansh_twt Mostly using Claude or GitHub copilot CLI + IntelliJ + GitHub copilot PR review. Works for me.

genuine question are people still using VS Code or has everyone quietly switched to Claude Code or Codex or Cursor or am i the only one ?

@SebAaltonen @simplex_fx @vivoplt Wow, that is the complete opposite of the environment I am working in. We have more repos than developers and thus always jumping around between frameworks and languages.

Also even at HypeHype, an average people maybe pushes 3 times a day. We all work in the same dev branch. People push small things (some behind ifdefs and feature flags) all the time to avoid big merge conflicts. Much easier for testers versus big branch workflow. Out of 50 people we had ~40 pushing to the same monorepo that had: server, social media client, game engine, gfx library, editor and many other independent technologies. All teams on the same repo. 40 people * 3 commits/people/day = 120 commits/day. But most of those commits are of course to other libraries in the same monorepo, so they don't conflict.

Interviewer: Your branch is 200 commits behind main. What will you do? Merge or rebase?

@vivoplt When I know everyone that may have checked out my branch and I can contact them then I prefer rebase. If required, squash commits first so that you have to solve conflicts just once. The force push is nasty but you end up with a clean branch this way.

@RatrektLabs @om_patel5 You cannot run 70b with just 16GB. Try with 9b and use an agent that has a lean system prompt. My combo is goose+llama.cpp+qwen3.5 9b.

@om_patel5 13 minutes is generous. tried running llama 70b on a 16gb macbook and it was borderline unusable for anything beyond simple queries. local makes sense for privacy or edge cases, but for daily dev work the API cost is just the price of doing business.

this guy tried running LLMs locally to save on API costs and waited 13 minutes for a single response lets be honest we've all thought about it "why am i paying for Claude when i can just run an open source model locally for free?" so he tried it and ran Gemma 4 to avoid API costs 13 minutes to get this response: "I am a large language model, trained by Google." tools like Claude Code and OpenClaw have system prompts over 20,000 tokens. so even your first message isn't starting from a clean slate. your local model is choking on context before you even ask it anything the API bill hurts but time is money

@0xPrajwal_ To entertain the models

Interviewer: If AI can write code, why should we hire you ?

@Yuchenj_UW To be honest, I haven't even tried Opus once. Sonnet works perfectly for me. If anything, I am looking at smaller models to run locally. But I am also not vibe coding apps, I am mostly fixing bugs and doing urgent changes, so my tasks are very well defined.

1 year ago, when “vibe coding” was coined, I was like: no real engineer would build serious projects with AI slop anytime soon. 1 year later, everyone is a vibe coder. Today, Claude Mythos looks like a huge leap, while Opus 4.6 is barely 2 months old! Scaling laws aren’t hitting a wall. RL works. AI is accelerating faster than ever. The craziest part: by the end of 2026, we’ll look back at Mythos and laugh: “what a weak model, and they were terrified to release it.”

Agents send a definition of their tools to the model. Many agents use a JSON format and the actual tool invocation is quite close. Often there are just a few missing bits. An RFC could improve compatibility. In fact, I used other agents with Sonnet and other models with Claude. Both is possible.

Porting syntax is easy. Porting behavior is the actual engineering problem. Rewriting Claude Code in Rust doesn't magically make it compatible with every other LLM on the market. People treat these tools like standard software where a Python rewrite behaves exactly like the TypeScript original. But the underlying engine isn't predictable like that. The leaked codebase is tightly coupled to how Claude actually handles context windows and tool calling. Pointing a system built for Anthropic's endpoints at a local Llama model doesn't give you the same results. It just gives you hallucination cascades. To actually build that middle layer so it handles any model successfully requires deeply understanding the behavioral quirks of every single neural net. And honestly if a developer actually managed to build a universal orchestration layer that predicts and wrangles LLM output across the board, they aren't shipping a fun open-source weekend project. They just built a massive enterprise moat. You aren't just translating code at that point. You're trying to normalize alien logic

Devs rewrote the leaked Claude Code in Python and Rust within 48 hours. Before Anthropic could even finish the legal paperwork. Now it works with any AI model you want.

@DThompsonDev When you look at all the agent CLIs out there you will realize that they are very similar. A system prompt, some tools, strategies to manage the context, etc. So you may actually swap out the interface. As long as you keep the model you may get similar results.

After reading a lot of the Claude Code base that was leaked taught me two things 1. I have never been more reassured than I am now. Software Developers will be around for a long time. 2. Claude Code has a hidden command to hide that a contribution was made by a Claude employee

@rezoundous No. I use AI only for cross project full stack work. When I know the code base and it is not a huge refactoring, then I get better and more sustainable results doing the coding myself. But I do use AI for PR reviews as that doesn't hurt and occasionally finds issues.

So it has been about a year since. Is AI writing 100% of your code?

@CodeByNZ Must have been an employee that got fired

Do backend devs really find joy in doing this ?

@Adidotdev Account enumeration

Interviewer: When you click “Forgot password”, The site says : Reset link sent if the email exists. Why not just say : Email not found ? What attack are they preventing?

@r0ktech userId, userid, or user_id, depending on the language.

which one do you prefer ? -userID -user_ID -UserID -userId -user-id

@vivoplt Depends on the protocol. If you are talking about a Bearer token then the only concept you have is token expiration. In the context of OAuth you can also do token revocation. And for OpenID Connect there is an RFC for logout. But hey, just throw these tokens away and pray.

Interviewer: If JWTs are stateless, how do you “log out” a user?

@Manixh02 Basically searches recursively for from and require statements, then extracts the library name and prints all out by sorting and removing duplicates. And yes, I can do that myself and in a way that is easier to understand. Go step by step and then create a script later.

The people who are saying AI will not replace Software Engineers I just want to ask them Can you write this below code? or Can you just explain me what this code do??

@SidJain_80 UUIDs can be beneficial if you ever have to import data from one DB into another and the target DB already contains some data. No clashes.

Simple question for backend engineers: Why are UUIDs preferred over auto-increment IDs in sharded databases? What problem are we actually solving?

@brentwoodmomof3 @aakashgupta Nice examples. That's what makes it a non-trivial challenge to solve.

@be_ncrypted @aakashgupta But consider these scenarios: Person to Chat: I can never please my manager. Help me figure out why. Another one: I want to get a gift for my friend to say thank you for all they have done for me. Etc, etc. you can’t program it to skip words without context.

Sam Altman said people saying “please” and “thank you” to ChatGPT costs OpenAI tens of millions of dollars a year in compute. 67% of Americans do it anyway. Run the math on why. A 2024 Waseda University study tested LLM responses across politeness levels in English, Chinese, and Japanese. Impolite prompts produced measurably worse outputs: more bias, more errors, more refusals. Moderate politeness consistently beat both extremes. The mechanism makes sense once you see it. Polite prompts pattern-match to higher-quality training data. When you write “Could you help me structure this analysis?”, the model pulls from professional, well-reasoned text. When you write “give me the answer,” it pulls from Reddit. Google DeepMind’s Murray Shanahan explained it simply: the model is role-playing a smart intern. Treat the intern like a colleague, you get colleague-quality work. Bark orders, you get minimum-viable compliance. Now look at the cost side. OpenAI handles over a billion queries daily. Each GPT-4 query uses roughly 2.9 watt-hours, ten times a Google search. But OpenAI just raised $40 billion at a $300 billion valuation. Tens of millions in politeness tokens is a rounding error on a rounding error. 67% of users do it anyway, and 55% of them say it’s because it’s “the right thing to do.” They’re maintaining a behavioral habit that governs every other interaction in their life. The parent who teaches their kid to say please to Alexa isn’t doing it for Alexa. They’re doing it because the alternative is raising someone who learns that being rude gets faster results. Telling 900 million people to stop saying thank you so OpenAI can save 0.01% of operating costs is the most engineer-brained optimization take on the internet. You’re training yourself to treat every interaction as a transaction. And that habit doesn’t stay in the chat window.

@trikcode It's easy, you learn English. You don't need a huge vocabulary either. It is a very bad idea to use your native language anywhere near code.

Honest question. People who English is not their first language… how do they code?? Do Germans code in German? Do Arabs code in Arabic??

@kermithetrader @djcows @allvibesnoskill Earn double, pay double in the shops, pay half of the tax, and save double at the end of the month.

@djcows @allvibesnoskill high income and high rent. It balances itself ( a bit )

i just learned Switzerland has ZERO capital gains tax??