Ddox
@paraddox
Followers
1K
Following
2K
Media
34
Statuses
1K
Doing Software Engineering from the land of Dracula. But I bleed dry tokens instead of maidens.
Timisoara, Romania
Joined May 2009
People will think you mad if you say this. But truth be told, LLM's will do a better job working in a prod env. Not THE prod env, but a prod-like env.
I think localhost matters much less now. Coding models are good enough that validating in prod-like environments just works. Why grind on localhost when you can iterate async against a real deployed environment?
0
0
1
Easy. Get an orchestrator. Put Claude Opus in it. Give it the parts of Sentry's docs that you need. Tell it to write you the specs for those then implement it. With a proper orchestrator I'd say it's a 3-4 days job.
next up on the list of cancelling subs: @sentry 🤡 best alternative? preferably self hosted via coolify? i don't need the extra 19834813 features just basic error tracking
1
0
0
Software engineering is dead. Long live software engineering. The version where you write every line? Dead. The version where you architect, direct, and verify? Just getting started. Same title. Completely different job. Anthropic isn't contradicting themselves. They're
Can someone explain to me why Anthropic's CEO keeps saying Software Engineering is dead, yet his company is still hiring Software Engineers?
12
6
72
Full-duplex voice + agent orchestration. Think about it: Live status feed from all your agents. Voice control to manage them. No pauses. No "hey assistant" triggers. Just talking to your system while it talks back. We finally have the tech stack for JARVIS.
NVIDIA just dropped PersonaPlex-7B 🤯 A full-duplex voice model that listens and talks at the same time. No pauses. No turn-taking. Real conversation. 100% open source. Free. Voice AI just leveled up. https://t.co/YfzFQfBzMS
0
0
1
This will quietly eliminate jobs. I know people whose entire role is cleaning up Excel files for their boss to present to their boss. The hard ones take 1-2 days. Claude in Excel does it in minutes. We're not ready for how fast this hits.
Claude in Excel is now available on Pro plans. Claude now accepts multiple files via drag and drop, avoids overwriting your existing cells, and handles longer sessions with auto compaction. Get started: https://t.co/cAMDXM1h7r
0
0
2
or just send in a repo of the orchestrator you built because the others you found were lacking :))
the new "technical" interview: - build: a feature with your AI agent of choice on your own time - submit: a PR, a loom of your feature, and the record of your agents sessions - talk: get on the phone and talk through the decisions you made WITH ai, show us how you chat with it
0
0
0
This is going to be so much fun 6months from now.
You can get started instantly—for free!—with our API here. https://t.co/E2QvIySKhD
0
0
0
Same energy. Just built a full CRM for my company this week . Mobile app with OCR included. Deployed to Vercel. Production by Monday. The secret? I wrote specs, not code. My orchestrator + Claude handled the rest. The "buy vs build" debate is over. The job isn't "write code"
Built my own version of @salesforce and @outreach_io. Just doesn't make sense to try to integrate with most external SaaS platforms now. I own all the data in pg and I can build exactly what I need. Yes, I can hear the comments already: "tech debt!" "slop!" "buying is better
7
0
16
Ralph loops aren't overhyped—they're underdeveloped. We have the for loop equivalent. We're missing: → Context management → Requirement drift handling → Domain knowledge injection The foundation is solid. Now let's build on it.
10
2
42
Best debugging prompt I've found for Claude: "Forget everything you did. Go through the plan again without assuming anything." That's it. No fancy framework. No complex chain-of-thought. Just force a clean reset. Works when Claude is stuck in a loop or confidently wrong.
22
3
341
This is bigger than it sounds. Claude Code can now: → Track dependencies between tasks → Coordinate across multiple sessions → Let subagents collaborate on the same project The "unhobbling" era is here. AI agents that can run longer and remember where they left off.
19
25
773
I wouldn't say all. I only got 3 of them being cloned at the moment.
Somewhat hot take: a majority of SaaS companies are screwed and they don't know it yet. Companies will build their own tools with AI at an accelerated rate.
1
0
0
You folks asked for it. Simplest Ralph loop: #!/bin/bash PROMPT="${1:-prompt here}" for i in {1..50}; do echo "=== Run $i/50 ===" claude --dangerously-skip-permissions -p "$PROMPT" echo "" done
10
9
235
AI didn't take your job. It exposed the uncomfortable truth: Writing code was never what made you a software engineer.
2
1
15
35 years of coding. 21 of them professional. You know what most of that time was? Reading docs. Fighting compiler errors. Fixing broken dependencies. Not building. Debugging. AI flipped it. Now I actually build things. Took 35 years to get here.
1
1
17
Or click the Deep Research button on any of those LLMs?
@jasonbosco Did you try running Codex or Claude Code in an environment with Ansible available and having them hack away at the playbook until they figured out how to get it to work?
0
0
0
Personal record on the Glm Max plan. Still have 50% left over. Need to git gud.
16
1
118
Took me a week to recover after I understood Ralph loops the first time.
tbh. i kinda agree. when you discover something new with AI it somewhat space blasts your brain. takes at least four days to recover. during that time is a poor period to make decisions.
1
0
5
Unless you take out the human and put it in a Ralph loop. This way by compounding the non-Deterministic nature of the LLM's you paradoxically get much closer to your stated goal.
AI generated code is not an abstraction like compiler. Compiler outputs are reproducible, deterministic and some are mathematically verified. LLM codegen is a non-deterministic, probabilistic code generator with a human in the loop. The human must understand what it spits out.
1
0
1