Jason Allen

@DougWahl1 They've been saying this for over a week now. Do it already and quit talking about it.

@rawsalerts I'll believe it when I see it.

🚨#BREAKING: President Donald Trump is expected to finalize a deal with Iran by Sunday

@nickgraynews One of the first things when I do whenever I deploy a new site is block China.

I just blocked the entire country of China from my PatronView.com website My site is getting hammered with over 1 million requests in the past 10 hours and a lot of those were passing the Cloudflare Managed Challenge I had to do the same to Vietnam recently too

@Rep_Stansbury Get off of social media and do something about it.

Why is the House not holding Bondi in contempt? Why is Oversight not holding Epstein hearings? Why is Oversight not doing more in this case? Good Question! Republicans have stopped regular committee hearings. And, instead are holding roundtables where official motions (like contempt) cannot be made. And they control the committee and the chamber. That’s why Dems are taking a stand. We demand public hearings. We demand compliance with the law. And, we demand real oversight and justice for the survivors!

@itsjoaki Calling BS on this one. I'm on the 5x plan and I code for 4-6hrs a day and never hit my quota.

Claude Opus 4.7 with 3 prompts on the $100 plan. Opus 4.8 will eat that in one prompt. Opus 5 will eat that with a GM.

@svpino Or you could just use a subagent instead. code.claude.com/docs/en/sub-ag…

Here is a trick to review your AI-generated code: Have two separate Claude Code sessions (open Claude in different terminals). • Use the first session to write the code. • Use the second session to review that code. The reviewer doesn't know anything about the process or shortcuts used by the session that wrote the code. This makes it somewhat unbiased. I also use this to write my tests: • First session writes the test • Second session writes the code that passes that test

Summarized workflow. Fleshed-out feature description in Linear. Pull the issue into CC with linearis. Create an initial plan. Iterate. Write code when plan is baked. Perform manual and automated testing. Fix anything not working. Run various subagents for code review/documentation. Commit when everything is green. Only 1 MCP (context7). I give CC direct access to local database. Use gent-browser for browser interaction. In general, I prefer mature stacks like RoR or Laravel over newer tools like NextJs or TanStack. Models tend to perform much better when they have many years of training data available.

@jasonallen @Shpigford Is it possible to have some tips ??

I have updated, slightly kinder thoughts about Opus 4.7.

@jidefr @pankajkumar_dev One CC instance at a time carefully planning and reviewing all work. Subagents for documentation, code and security reviews. Lots of quality gates along the way. Codebase is about 100k LOC.

@jasonallen @pankajkumar_dev Expand 🙏

why opus 4.5 still clears 4.6 and 4.7 (power users know) - 4.5 is still the most reliable for one-shot coding, you give it a proper prompt and it actually executes end t end without skipping steps or needing constant correction - it follows instructions properly, respects constraints, and doesn't try to be "smart" by changing your approach midway - 4.6 is where things started slipping, it often ignores guardrails and pushes oversimplified or outright wrong fixes just to finish faster - instead of solving properly, 4.6 tends to cut corners, which is frustrating when you’re doing real engineering work - 4.7 makes things worse for power users because of the tokenizer change, same input now consumes way more tokens, so your limits get hit much faster - this basically means you pay more (or get less usage) for similar or sometimes worse output quality - despite being newer and more expensive, 4.7 struggles more with long context and multi-step instruction following compared to 4.5 - in complex tasks (full-stack flows, debugging, system design), it feels less consistent and harder to trust - overall quality drop feels like a tradeoff to handle scale, as demand increases, depth of reasoning and output sharpness seems diluted - because of this, many serious devs move to GPT-5.4 as a more dependable alternative for deep work.

@pankajkumar_dev Cheaper? I can code for 6 hours a day with any model on the 5x plan and never hit my quota. Not sure what you guys are doing to burn through so many tokens.

@jasonallen 4.5 is cheaper than 4.7, and context is better in 4.5, that i feel.

@RonFilipkowski It's telling that he deflected instead of addressing the drinking problem claims head on.

No coincidence that Kash goes on Fox to claim he’s about to make arrests of people who supposedly rigged the 2020 election right after a story breaks that he has a serious drinking problem. A pattern of things like this every time Kash gets in trouble for his audience of one

@haider1 Opus 4.7 is a different beast than 4.6, but after some trial and error, including updating the CLAUDE.md and agent instructions, it's extremely effective.

opus 4.7 is a really confusing release we're on the max ($100) plan, and it's actually burning more credits because it gets lost so often everything takes much longer to fix now before, we could use the full 5 hours efficiently, but now we barely get 1.5-2 hours of real use

@KaiXCreator I just switched to Zed from Cursor and love it. Super lightweight and fast.

As a dev, which is G.O.A.T code editor?

@gnotnuk @Shpigford The coding companion feeling is gone. It's like talking to a robot.

@Shpigford curious, what's the specific dealbreaker for you in 4.7, compared to 4.6 ?

back to opus 4.6. i've done a LOT of model-changes over the past couple of years and never had as many problems as 4.6 → 4.7. absolute dumpster fire of a model.

@specialopsmag Three Kings

What's the most realistic war movie?

@RepRobertGarcia Yeah, so get off social media and make it happen. You post these comments like you expect the general public to do something.

There's a lot the First Lady can do to help our Epstein case. She could demand that her husband release the full files. She could ensure Pam Bondi testifies before Congress. She could tell Todd Blanche that this investigation is not "over," like he says it is. And she should.

@RepRobertGarcia Then do something about it. 🤦‍♂️

There has been no indication that Pam Bondi is actually going to comply with our subpoena, which legally requires her to testify before the Committee. In that case, we have no other choice but to move forward with contempt charges.

@michael_chomsky Add to your Claude settings.json so you don't have to manually specify the options every time you start Claude Code.

If you're not starting every Claude Code session with this command, you're basically donating to Anthropic.

@CleburnW @pcshipp Why not open the terminal inside VS code?

@pcshipp I have my terminal open beside VS code, I like to be able to see and edit whatever I’m having CC work on

Is anyone still using VS Code instead of switching fully to Claude Code or Codex? Or am I the only one 😅