grinding works building a b2c product means failing every day but once in a while you get a breakthrough, and when it happens it's freaking magical. for weeks we've been trying to increase the number of shares on the platform. playing around with placements, sizes, options, and so many things with no sucess. but when we launched the snapshot feature, it finally clicked and number of shares spiked by x2.5!

my llm generated code is better than yours we use the same model and type into the same box and we still get different code. just like two engineers use the same ide don't write the same code. when i prompt claude, i already have so much context of edge cases, history of decision making, common pitfalls, and this matters a lot as it helps me write better prompt. more importantly i know what not to say. so yes, we both use opus 5.5. I've been practicing software engineering for 20 years that's all

Stop chasing. Discover what's next delightfully. @dailydotdev youtu.be/6i4Bd7vyoeQ

Coding agents are just a fancy autocomplete! One of the most used arguments against AI and the most revealing. It shows that you never gave coding agents an honest try and probably stopped somewhere way before Opus 4.6. So if this you here's what you need to do take two weeks, a subscription to claude code or codex, use Opus 5.5 xhigh or Sol 6.1 xhigh, force yourself to not doing anything by hand and asking the agent to do everything. Let's see what you think after.

No more spam! @dailydotdev is now powered by @typesafeai Jev. We scan every signup, post, and comment in real-time to fight bad actors ruining the experience for everyone. The daily dev experience is now the most delighting on the internet, thanks to our ai detection and clickbait shield as well

thanks to @typesafeai jev, we now have real-time spam detection at 90% accuracy like any growing network, @dailydotdev attracts many spammers that tries to abuse our domain ranking. we have a nightly sweep powered by opus to hunt those and now thanks to jev we can do it efficiently in real-time. the nightly sweep is still on and it reviews the decisions made by jev and potential misses. glad we can use AI to make everyone's experience much better on the internet

There's never been a better time to share this Game development is so hyped right now and you're probably curious to see the game dev sites ranked by clicks on @dailydotdev If you're curious about game dev, we've got it all.

I used to hate our search! Every time I searched for something I felt it's too slow and I'm getting the wrong results, the worst combo. Last week, I've had enough, asked Fable, let's improve our search. Gave it access to our analytics warehouse, it pulled the data and created an eval. We iterated on the implementation and fine tune the weights of engagement, recency and relevancy but we managed to pull it off. A new search version that is faster and more accurate, it's now being rolled out to production gradually. Btw, the incumbent was designed for a more sophisticated queries than what our users ask

Fable and I shaved 12% of our Google Cloud bill in less than a day I prompted Claude Code with this naive prompt: find cost optimization opportunities for our google cloud setup. you can use the browser for missing data and spin bg agents if you need Fable then started investigating our cloud setup, cost, different components, and opportunities using gcloud cli and the browser. It gave me a list of opportunities, mentioned their blast radius and effort. We then tackled each opportunity on a new thread. "spin off a new thread to discuss ABC". Most of the tasks, Fable handled for me, some I did manually and others I delegated to the relevant owners. But we're talking about thousands of dollars per month in less than a day.

we learned from the best @nikitabier users were screenshotting the app with marked text which gave us zero attribution. now when they mark they get a dedicated bar with a branded snapshot. now live on @dailydotdev

Stop writing so many tests Unlike humans, coding agents love to write tests, and many of them in all shapes and colors. They spam our codebase with repetitive tests trying to optimize for coverage as much as possible. This leads to longer CI times and more token spends when they try to read these files, in other words it increases the cost on all front. Here's a prompt to help you cut all the unnecessary tests. I removed 300 tests and over 5k lines of code from only two repos at @dailydotdev. ``` use dynamic workflow to identify all the redundant tests in this codebase and remove them. by redundant i mean they basically test nothing, another test already covers them and they just waste cpu time for ci. check for cross file dups as well. ``` To avoid future bloat, add this to your agents/claude md file: ``` **Same branch is not a new test**: before adding a test that differs from a neighbour only in an input value, confirm the production code actually branches on that difference — a passthrough field, or one guard rejecting every disallowed enum, means the existing test already covers it. ```

You wont believe why I switched from Codex to Claude Code I've been a long time Codex users but what got me back to Claude Code isn't a model, it's a harness feature, dynamic workflows. Claude Code and Codex have their built-in workflow for how to execute a task, it's mainly code oriented, but sometimes isn't enough for the task at hand, especially if it involves scanning many many files meticulously. This is where dynamic workflows shine. Claude will create a new workflow specifically to the given task and then fan it out to multiple agents to work on it in parallel. For example, the Bun migration to Rust have used dynamic workflows. At @dailydotdev, I recently used it to remove redundant tests across tens of files and to find new sources we should aggregate across the web. Claude can decide on its own to use it, but if you can also ask it to use it by saying "use dynamic workflow" and then your task

Software factory is the wrong bottleneck For a small startup velocity is usually not the blocker. The missing analyst, sales ops, product and marketer are exactly those blockers. They prevent you from winning. Agents can do much more than coding, don't sleep on it and act now. daily.dev/blog/software-fact…

Building a software factory? Leveling up your agentic engineering skills? Top reads below. (ranked by @dailydotdev community)

You're spending too many tokens Since Opus 5 claude.md files should be reduced significantly for better results and token efficiency. The claude code team reduced the system prompt by 80%. If you haven't done it yet, here's a prompt to help you out: ``` The new generational models dont need so much context and we can reduce significantly the claude.md files. see this: claude.com/blog/the-new-rule…. find all the claude.md files and keep only the min context we need there based on the doc above. our goal is to avoid mistake, keep the code consistent and be token efficient ```

Juniors and veterans don't read the same way. 90 days of reads on @dailydotdev split by how long people have been coding. While juniors are follow the hype of AI and web development, veterans still focus on the fundamentals and deep expertise.

One Trend a day keeps the FOMO away! Today we're introducing @dailydotdev Trends. Reading the code or not? Claude design is taking over? Usually you would so many tweets about these topics. Now we aggregate them into one coherent post with everything you need to know.

You don't know what you don't know about AI. The field moves too fast for anyone to hold a map of it in their head. So we made the map. The Agentic AI Hub takes you from "what's a token" all the way to running your own agents, one guided path at a time, each starting from a question. Along the way you'll run into ideas from @danshapiro, @simonw, @addyosmani, @thorstenball, @karpathy, and @steipete. We just put them in the right order. It's free, no signup. Now you know where to start.