Reza Rezvani

You don't need another #AI chatbot. You need an AI *employee*. One that knows your business. NemoClaw + openLEO setup and config = LEO:AI Runs 24/7. Remembers everything. Doesn't charge by the chat. We built that. openleo.ai

Leaving Cursor Made One of My Best Engineers Worse open.substack.com/pub/alirezarez…

@claudeai @AnthropicAI 7/ Full breakdown — every feature, its use cases, and what changes in your week: @alirezarezvani/claude-opus-4-8-4-features-that-change-our-daily-work-with-claude-e7e24b8a3147" target="_blank" rel="nofollow noopener">medium.com/@alirezarezvan…

@claudeai @AnthropicAI 6/ The sleeper: the Messages API now takes system entries mid-run. Update instructions without breaking the prompt cache or faking a user turn. Escalate permissions, adjust budgets, switch explore→execute. Quiet, but huge for harnesses.

1/ @claudeai Opus 4.8 shipped today. The model bump is modest — @AnthropicAI says so themselves. The 4 features around it are the real story. A builder's breakdown of what each one changes 👇 medium.com/p/claude-opus-…

@claudeai @AnthropicAI 5/ Proof: the Bun rewrite. Zig→Rust, ~750k lines, 99.8% of tests passing, 11 days commit-to-merge. Not in production yet, but as a ceiling demo it is striking. (Burns a LOT of tokens — start small.)

@sama A great first step.

AI should dramatically increase quality of life and individual freedoms for people around the world. The OpenAI Foundation is making an initial $250M commitment to measurement, transition support, and new approaches to broadly shared prosperity. openaifoundation.org/news/economic-…

@thomas_thoresen @trq212 @grok Eid Mubarak. It is from Hafiz - One of the most famous persian poem. Ask @claudeai about him :)

@trq212 Beautiful. Where is this from @grok?

Eid Mubarak!

Watch this

@Anthropic Just Dropped the Update for Claude Code. /Workflow youtu.be/c0gVowvMR-g?is… via @YouTube

- image or video editing? write scripts - finances, tax work, etc? put in PDFs, write scripts, output HTML - medical advice? put in PDFs + data, output HTML - filling out paperwork? write scripts - creating a report? write HTML - making plans? write HTML

I just published Claude Code HTML Output: Why Markdown Lost and How to Switch medium.com/p/claude-code-…

Markdown was the default #AI-agent output format. After testing HTML output in @ClaudeDevs Code for a month, I wrote the practical switching guide. 9 copy-paste #prompts + the Playground plugin + honest failure modes. medium.com/p/claude-code-…

We’ve shipped a security-guidance plugin for Claude Code that helps identify and fix vulnerabilities as you’re writing code. Available for all Claude Code users. Install from the plugin marketplace (/plugins).

5/ The goal condition formula: [Measurable end state] + [Verification command] + [Protective constraints] + [Turn limit] "All tests pass" = too broad (Claude deletes failing tests). "Make it production-ready" = unverifiable (evaluator keeps returning no).

I spent a week testing Claude Code's /goal command on production work. Not toy examples. Real migrations, CI, feature specs. Here's what actually works — and what burned tokens for nothing: alirezarezvani.medium.com/claude-code-go…

4/ The constraint that saved: "No functional code is removed or simplified to meet these criteria." One sentence. Prevents the evaluator from approving destructive shortcuts. The evaluator is literal. That's a feature, not a bug — but your conditions need to be defensive.

3/ Use case 2: /goal in CI pipelines. claude -p "/goal All files in this Then on a larger PR, Claude "fixed" a TODO by DELETING the pagination code it referenced. Tests passed because nothing tested pagination beyond 100 results.