Moacir Braga

Hey, folks! Eligible ChatGPT Pro plans now include matching API credits: $100, $200 or $500/month. Claim: Settings > Billing. First grant is prorated; unused credits expire at the end of the billing cycle. help.openai.com/en/articles/… #OpenAI #SoftwareDevelopment

Hey, full-stack engineers! Have 10+ years of software engineering experience and strong hands-on experience across both frontend and backend? I have MANY long-term remote opportunities with a U.S. company based in Texas, paid in USD. These opportunities are open ONLY to candidates based in Argentina, Brazil, or Colombia. These are some requirements: → Never had any contact with Velozient, the intermediate company for this position, if you had it, I can't refer you. 😬 → 10+ years of professional software engineering experience. → Substantial experience building and shipping both frontend and backend systems. You should understand the fundamentals of both and be able to discuss technical decisions on each side. → Hands-on system design experience. You’ve helped architect a product and/or contributed throughout its full lifecycle, from design and development to production and ongoing evolution. You should be able to explain the architecture, your contribution, and the trade-offs involved. The stack is primarily Go and React, with Python used mainly for AI-related work. Experience with multiple backend languages is a strong plus. Haven’t worked with Go yet? No problem! If you meet the requirements above, we’d still love to hear from you. Interested and meet these requirements? Send me a quick DM. No résumé or long introduction needed, just: → Your LinkedIn account/URL. → Your country and years of experience. → A short list of the frontend and backend technologies you’ve used extensively. → Your email and phone number with country code. A few lines are enough! If your background matches, I’ll pass your contact details on to the recruitment lead, who will guide you through the process. Know someone whose experience fits? Tag them or share this post with them.

Not sure if you all know about it, but if you have a OpenAI subscription, you can apply to have access to Daybreak model which allow you find vulnerabilities on your own code with way less restrictions on your prompts. Just keep in mind you will need a physical security key like the YubiKey ones. #codex #openai #cybersecurity #security #ai #llm #developement

Architect with Obsidian and ChatGPT desktop app with GPT 6.1 Sol high ultrafast, and asking for reviewing the flow of what I am specifying is soooo amazing.

I think the biggest partner OpenAI announced today is Anthropic because it is not measurable how many customers they will probably transfer to the competitor.

Am I the only one who thinks the timing isn’t a coincidence? Less than a month after their CEO’s public conflict with the Department of War drove a massive influx of new customers, Anthropic’s usage limits have gotten noticeably worse. Yet they say they can’t find the issue. Explosive growth plus constrained GPU capacity for inference plus vague communication on token quotas for paid subscribers equals a strong incentive to quietly throttle limits across the board while claiming no problems found. Meanwhile, users report their quotas draining drastically faster. Scaling high-demand LLM services at this pace is never trivial, but transparency would go a long way. #Claude #ClaudeAI #Anthropic #AI #LLM #RateLimits

After paying for more than an year expensive plans such as OpenAI Pro ($200) and Claude Max 20x ($200), I simply gave up of using these models you know are not sustainable for these companies, they always mess up with our workflow when their GPU are too busy for getting a torrent of new customers. To use them safely it is only by API which is very expensive for coding. I simply moved to @opencode using @Kimi_Moonshot Kimi K2.5 Turbo from @FireworksAI_HQ which is possible to pay even by API token usage. Fortunately, Fireworks AI is also providing FirePass which only costs for now $7/week. This model is insanely fast and with a high accuracy, fewer hallucinations, and if you are engineer and know what you are asking step by step, the model does the right job way faster than Opus 4.6 or GPT 5.4. The worse case when I want an extra pair of eyes reviewing the plan or code, I use OpenCode Zen credits to support them and also centralize my budget to access these frontier models.

Step 14: Using GitHub MCP, request Claude Code to create all TODO items as issues on GitHub and for each issue, get its ID and update the TODO items with the issue number and link. Describe the template for the issue so the AI can create all issues following a same pattern. My pattern is: Tasks, Acceptance Criteria, Related Documentation.

Step 13: With all documentation and guidelines in place, ask Gemini 2.5 to create to create a TODO list in markdown with the [ ] in the beginning for non completed tasks and the [x] for completed ones. Explain this TODO will be the implementation so it needs to follow an order capable of be executed considering its dependencies.

Step 12: Based on all previous work, your security expertise and the tech stack, ask Gemini 2.5 Pro to generate a comprehensive security guideline.

Step 10: Ask the AI (Sonnet 3.5 or 3.7) to analyze each page and add links to the relevant API endpoints it will connect to. Step 11: Create a markdown containing the entire tech stack to be used and some tech guidelines.

Step 8: Review and refine all page documents for clarity and completeness. Step 9: Repeat the process for the API, this time ensuring the documentation takes the pages into account.

Step 7: Using the Prisma schema, features, and acceptance criteria, ask the AI to create a folder structure with a markdown file for each page your project will include. Make sure the output is technology-agnostic (but web-oriented), and that each page includes required permissions, its goal, and the user flow.

Step 6: Review the generated Prisma schema and continue refining it with the AI (using Sonnet 3.5 or 3.7 as needed) until it’s correct.

Step 5: Using Windsurf or Cursor, ask Sonnet 3.7 Thinking to generate a Prisma model for a project using SQLite, explicitly requesting no enums. Instruct it to base the schema entirely on the features and acceptance criteria markdowns.

Step 4: Carefully review each item in the acceptance criteria, iterating with the AI until everything is exactly as you want.

Step 3: In Gemini 2.5 Pro, prompt it to create an acceptance criteria markdown using only sections and bullet points. Each bullet should use keywords like MUST, MUST NOT, SHOULD, SHOULD NOT, and MAY.

Step 1: In Gemini, ask it to generate a markdown document listing all the features you want for your project. Be as detailed as possible to ensure nothing gets missed. Step 2: Review and refine the generated features documentation to make sure it matches your vision.