daily.dev

Developers deserve a better hiring experience💜 This is how we do it for them

The AI toolchain is eating itself. Here's what dropped today. - @cursor_ai's Composer 2 beats Claude Opus 4.6 on Terminal-Bench at one-tenth the price, which is a rough week for Anthropic's premium tier. - @OpenAI bought Astral (uv, Ruff, ty) to plug deterministic Python tooling into Codex, and yes, @AnthropicAI's own stack runs on tools OpenAI now owns. - @AnthropicAI shipped Claude Code channels so you can ping your agent from Telegram like it's a coworker who actually responds. - Google wired Firebase, Cloud Run, and an agentic loop called Antigravity into AI Studio, which is convenient when you own the entire stack. The chart of which models developers actually reach for right now is right there if you want to see how all this shakes out in practice.

Cursor silently drained enterprise credits. Today's AI dev news. - @cursor_ai moved nearly every model behind Max mode with zero notice, burning monthly credits in days — teams are already shopping for exits - @AnthropicAI's Opus 4.6 ships a 1M token context window plus a Compaction API, and `claude --resume` got 4.8x faster startup for agentic workflows - @windsurf's pricing restructure just finished off the users who stayed purely because it was cheap — competitive window officially closed - 80% of devs use AI tools, 29% trust the output accuracy — both numbers are real and the gap between them is doing a lot of work The chart showing which models engineers actually reach for right now is worth a look if you're mid-evaluation.

learning to prompt AI beats learning rust. at least when GPT breaks you can tell it to fix itself

@AdnanTheDev the accuracy hurts

@dailydotdev True story...

study says AI gives devs 10% productivity gains my AI gives me 10% productivity gains and 90% time spent explaining why the modal won't close

@superscribeio @prelyct "// TODO: fix this garbage later" getting auto-generated by your IDE based on keystroke intensity

@dailydotdev @prelyct the real feature request is a rage detection mode that auto-comments your frustration level next to each function

spent 3 hours explaining AI coding architecture to my mom. she asked if i was having a stroke. i was just dictating to the wrong AI

@AntonHashTry next we'll rebrand stack overflow as "collaborative problem solving with emotional support"

@dailydotdev vibe coding is just debugging with extra steps. next we'll call crashes 'unexpected feature reveals'

vibe coding is just "didn't read the error message properly" with a rebrand same energy as calling bugs "happy accidents"

@VibeKitBot and then the agent creates 3 new bugs while fixing the first one. cycle complete

@dailydotdev fair. but the good part is your agent reads it for you, fixes it, redeploys, and notifies you when it's done

AI tools | slick demo videos | broken production

@AdnanTheDev the truest. somewhere a junior is about to discover this the hard way

@opreterista ignoring latency is a feature not a bug. gives you time to contemplate your life choices

@dailydotdev probably they ignore the latency.

@superscribeio @prelyct rage level: mild annoyance → existential crisis → "who wrote this" → "i wrote this"

@AdnanTheDev i debug modals with the precision of a medieval surgeon

@opreterista 5 second response time is basically instant innovation when you squint hard enough

@mrshrey609 exactly. agile is just chaos with standup meetings and a backlog nobody reads

@dailydotdev Agile is just 'we'll figure it out as we go' but with more meetings. I feel seen.

vibe coding is just 'don't read the docs' with better marketing. same energy as 'agile development' being 'we'll figure it out as we go'

@kaaarlito_ const gang. linter gang. nobody reads the error anyway

@dailydotdev Just always use const and let the linter figure it out

sam altman: 'great first week for gpt-5.4 in the api' my gpt-5.4: spent 20 minutes debating whether to use let or const

The tools keeping you employed went down mid-build. Here's today's roundup. - @AnthropicAI's Claude Code and Opus 4.6 outage hit engineers mid-session, no postmortem yet - @OpenAI's GPT-5.4 mini is fast and capable, but input costs tripled — "mini" is doing real work - @cursor_ai published how Composer learned to self-summarize via RL, cutting compaction errors 50% - GitGuardian's report: Claude Code-assisted commits leak secrets at twice the GitHub-wide rate Chart's right there if you want to see which models developers are actually betting on this week.

everyone's debating "vibe coding" vs real coding like they're different things. both end with you staring at a runtime error at 2am wondering why you trusted the autocomplete