wtf have I just read
so many things wrong here it's actually impressive:
- no amount of automated checks/tests will make you sure that your deploy won't just straight up kill ur prod
- feature flags save you from nothing when someone merges a migration that drops a column your app references on startup. app literally never boots. what flag
- the automated tests cover what you thought of, dev envs and before-release testing catch what you didn't even know existed, these are completely different things
- your dev/staging envs are insurance, it's the same concept as wearing a seatbelt while driving, doesn't matter if you are Verstappen if some retard runs a red light going 50 over the speed limit
- this might work if your app is a to-do list for you and ur dog, try proposing this to someone handling money, health data, or some air traffic control systems and watch their faces
at the end of the day, some of the practices listed are good and should absolutely be adopted, but you can have good practices AND staging/dev envs, they are not mutually exclusive
@samuelcolvin mostly people juggle a bunch of projects in parallel, they don't have many guardrails/validation layers in place, which causes them to burn way more compared to focused work on a single project, where the person is in full control and doesn't allow agents to wander off
@Aditya_181105 unpopular opinion, but all of those will be almost identical, since code quality hugely depends on your setup and harness settings, i.e., agentsmd, some skills, review process, and such
@ChadNauseam wtf is this article, there is a caffeinate command built in, its literally free, always available, usable when you actually need it:
you need your agent running -> caffeinate -dim
you DON'T need it and "u r on ur bike going home" -> just close the lid, and ur mac sleeps
LMAO, for a second - this guy is a CTO
This by far is my favorite tweet at least through this month, there are so many things wrong with this take that it's extra hilarious ๐คฃ
RESTful APIs may be dead soon. Instead, web services may expose a single POST entry point for a prompt. Internally, an AI agent may decide how to interpret it and what to do with the data and the database.
@rohanpaul_ai I'm still waiting for this take to come true
https://t.co/CNrw1QymXk
Is every engineer typing manually 99% less - yes
Can any responsible engineer leave any agent unattended on an actually big app and let it maintain it - hell nah
Saw this on someone's status.
If your app can't handle 200 users, I don't know what you're doing, man.
Is that even possible? I feel like you have to build it not to be able to handle 200 users intentionally. Even a simple json file as db can scale to 200 users.
Maybe na 200k e wan type sha. Cozzzzz????
this gotta be ragebait, the difference between glm-5.1/kimi-k2.6/deepseek-4-pro and frontier models is way smaller than you are painting it
in most workflows, including heavy coding ones, you probably won't even notice the difference, so if you already HAVE the hardware, there is no point in using subscriptions to get a negligible improvement over the models you can run locally
However, if you don't have the hardware, there is no point in buying it just to run local models over the subscriptions you can stack, unless you are planning to go really heavy, but even then, you are better off renting rather than buying
@eaglebuildz@dreamsofcode_io until your shit crashes Saturday midnight, your agents can't seem to find the issue, there is literally no one who even somewhat knows what is going on inside your codebase, and it takes you a shit ton of time to find and fix the issue.