Neil Dave

@AnthropicAI @scaling01 Competitors are collaborating, crazy scenes ahead. Look fwd to it.

We're expanding our collaboration with Amazon to secure up to 5 gigawatts of compute for training and deploying Claude. Capacity begins coming online this quarter, with nearly 1 gigawatt expected by the end of 2026.

@ajambrosino @grok can you verify!

excited to welcome Tim Cook to the Codex team

@BlyndLemons @amit_code On same boat buddy!

@amit_code Been grinding replies for weeks now, but followers are still playing hard to get 😅 Any tips on making them actually stick?

X algorithm is easy. If you want more engagement. Engage with others. More replies = More followers.

I am following tech folks to keep myself updated : @karpathy - No intro required @bcherny @trq212 - for Claude code @thsottiaux - OpenAI @LLMJunky - codex @aakashgupta - PM specific @ziwenxu_ - openclaw Whom do you follow to keep you updated?

If you’re using Claude Code (or want to 10x your AI coding game), stop everything and read this. 📷claude-code.program.repair Tag a dev who lives in Claude Code. RT if you’re saving this for later. #ClaudeCode #AICoding #Anthropic #AI #Programming

🚨 103-PAGE CLAUDE CODE DEEP DIVE JUST DROPPED (and it’s FREE)Columbia students went nuclear: • Context Management • Tools & Agents • The full Agent Loop This isn’t another fluffy tutorial. It’s the actual blueprint of how Anthropic’s Claude Code actually thinks and works.

@aakashgupta To make more money you just need to find solution and create scarcity to book profit from the opportunity. This is same play book.

Microsoft sold every spare CPU it had to Anthropic and OpenAI. Amazon tripled its CPU buys year over year and still can't keep up. Two of AWS's biggest customers asked Andy Jassy if they could buy the entire 2026 production run of Graviton chips. He said no. The ratio inside an AI datacenter used to be 100 megawatts of GPUs to 1 megawatt of CPUs. CPUs handled storage, checkpointing, pre-processing. Light work. GPUs did the actual training and inference. Then OpenAI shipped o1-preview in September 2024. RL post-training went from "check the model output with a regex" to "run classifiers" to "compile the code and run the unit tests" to "spin up a sandbox, call three databases, run a physics simulation, verify the answer." Every rollout now needs a CPU-backed environment to verify against. Codex 5.4 runs agentically for 6-7 hours at a time. Each database call, each cron job, each scraped URL is CPU work. Coding agent revenue went from a couple billion to north of $10B in six months. That compute is sitting on CPUs. The CPU to GPU ratio is now approaching 1:1. The entire global cloud was built for 1:8. That's why GitHub has been unstable for weeks. Nvidia and Arm both announced they're entering the server CPU market in March. TSMC will only meet 80% of server CPU wafer demand this year. High-end server CPU prices are already up 50%. When the GPU king and the IP licensor both pivot to CPUs in the same month, the boring chip isn't boring anymore.

Agentic AI feels like the real shift in 2026 — from chatbots to actual execution in workflows. In India, this could transform GCCs and startups massively. Which agent tool are you experimenting with right now? #AI

@rxhit05 Completely agree

The reason why 90% of Vibe coded SaaS dies after launch: - No input sanitisation - No error boundaries - Hardcoded API keys -Tokens in localStorage - Sessions never expire - No Stripe webhook verification - Reset links never expire - Sync email sending - No CDN for images - No env validation - No health checks - No rate limiting - No pagination - No DB indexing - No CORS policy - No DB pooling - No role checks - No logging - No backups - No TypeScript

@lunarfq Hello

No account should be under 500 followers Say Hello, we follow you asap 😊🥁

@Steezehuman Hii

Are you a small account Just drop hey We’ll boost you instantly

Hiring a Twitter Ghostwriter ✍️ If you know how to write scroll-stopping hooks and engaging threads, this is for you. Paid opportunity. Full remote. Apply directly: hello@theneildave.in #hiring

@TechLayoffLover Its time to go for manufacturing business or independent AI consultant to safe yourself

**NVIDIA JUST TERMINATED 800 SOFTWARE ENGINEERS WHILE JENSEN HUANG COLLECTED $7.8 BILLION IN STOCK WEALTH AND ANNOUNCED THEY'RE REPLACING THEM WITH THEIR OWN AI CODING AGENTS** The same company selling AI chips to automate everyone else just automated their own goddamn engineers 800 software engineers. Gone. The people who built CUDA, who optimized the drivers, who made the AI revolution possible in the first place Jensen's personal net worth jumped $7.8 billion in Q1 alone while he walked out the humans who created his empire But here's the truly sick part: sources inside are saying NVIDIA's new "CodeGen AI" agents wrote 73% of their latest driver updates The engineers getting terminated? They spent the last 18 months training those exact systems "Workforce optimization through internal AI deployment" is corporate speak for "we taught machines to replace ourselves and now we're fired" I'm hearing the remaining teams are 90% prompt engineers managing AI workflows with skeleton crews of senior architects The company that's powering every other tech layoff just ate its own children Stock hit $1,847 after the announcement because Wall Street knows exactly what this means If NVIDIA's own engineers aren't safe from their AI, what the fuck makes you think you are Every line of code you write is training data for your replacement

@ClaudeDevs Dev is only burning more tokens then previous it is indirect tax to all user bcz of tokeniser

Here’s what developers should know to get the most out of Opus 4.7:

Another beast out! How is your experience? Tell me. I will share my experience where it is beating old version.

@agentshield_ai Yeah exactly that's where the blast radius comes and stop execution if it doesnt have necessary items.

Most people building AI agents are winging the architecture. Here are 17 design patterns every serious agentic system needs 🧵 🔴 Agent Circuit Breaker — Prevents cascading failures by stopping agent execution. Your agents WILL fail. Make sure one failure doesn't take down the whole system. 🟡 Orchestrator vs Choreography — Defines the control flow of agent interactions. Know whether you need a conductor or a jazz band. 🔵 Confidence Threshold Gate — Ensures agents only act when confident in their decisions. Low confidence? Stop. Ask. Don't hallucinate forward. 🟣 Idempotent Tool Calls — Ensures tool calls can be repeated without side effects. Because your agent WILL retry. Make sure retrying twice doesn't charge a customer twice. ⚪ LLM Gateway Pattern — Manages LLM interactions and provides a single interface. Centralize your model calls. One place for rate limits, logging, and fallbacks. 🟠 Human Escalation Protocol — Provides a mechanism for human intervention when needed. The best agentic systems know when to tap out and hand control back. 🟢 Replanning Loop — Allows agents to adapt and replan their actions. The real world doesn't follow your prompt. Build agents that can course-correct. 🔵 Agentic Observability Tracing — Tracks agent behavior and performance for debugging. You can't fix what you can't see. Treat agent traces like production logs. 🟠 Blast Radius Limiter — Restricts the impact of failures to a specific scope. Blast radius isn't just for microservices. Contain the damage before it spreads. 🟢 Tool Invocation Timeout — Prevents agents from getting stuck waiting for tools. An agent waiting forever on a hung API call is a frozen agent. Always set timeouts. 🔵 Context Window Checkpointing — Saves agent progress to resume from where they left off. Long-running agents need memory across steps. Don't restart from zero. 🟣 Dead Letter Queue for Agents — Stores failed agent tasks for later analysis. Failed tasks are gold. Capture them. Replay them. Learn from them. ⚪ Semantic Caching — Stores frequently used LLM responses to improve efficiency. "What's the weather in Paris?" shouldn't cost you an API call every single time. 🟣 Multi-Agent State Sync — Ensures consistent state across multiple agents. Distributed agents need a single source of truth. Don't let them contradict each other. 🟡 Canary Agent Deployment — Tests new agent versions with a small subset of users. Don't YOLO a new agent to prod. Shadow test it. Gate it. Ship it safely. The gap between a demo agent and a production agent is these 15 patterns. Save this. You'll need it. 🔖

Unpopular opinion: if you're not spending $1,000+/month on AI — you're not actually using it. This is 7-day Claude Code bill: $1,453.84 14,220 calls. 2,131 sessions. 100% cache hit. And you would have pay it again tomorrow. Here's what serious AI usage actually looks like: → Opus 4.6 dominates at $1,409 of the spend You don't use a Ferrari in second gear → $815 on Conversation. $311 on Coding. The thinking IS the work → Bash called 1,736 times in 7 days This isn't prompting. This is infrastructure. → 2.3 BILLION cached tokens Cache hit rate: 100%. The system learns you. → One project = 97% of total spend Deep work > shallow sprints. Every time. Most people have a $20/month ChatGPT subscription and wonder why AI isn't changing their life. The ROI isn't in the tool. It's in the intensity of use. Stop treating AI like a search engine. Start treating it like a co-founder. Drop your weekly spend below 👇 (no shame, all signal) Unpopular opinion: if you're not spending $1,000+/month on AI — you're not actually using it. This is my 7-day Claude Code bill: $1,453.84 14,220 calls. 2,131 sessions. 100% cache hit. And I'd pay it again tomorrow. Here's what serious AI usage actually looks like: → Opus 4.6 dominates at $1,409 of the spend You don't use a Ferrari in second gear → $815 on Conversation. $311 on Coding. The thinking IS the work → Bash called 1,736 times in 7 days This isn't prompting. This is infrastructure. → 2.3 BILLION cached tokens Cache hit rate: 100%. The system learns you. → One project = 97% of total spend Deep work > shallow sprints. Every time. Most people have a $20/month ChatGPT subscription and wonder why AI isn't changing their life. The ROI isn't in the tool. It's in the intensity of use. Stop treating AI like a search engine. Start treating it like a co-founder. Want to try the same for you? one line install: npx codeburn Drop your weekly spend below 👇 (no shame, all signal)

@vatsal_sanghvi DM'ed you.

ugc & ad agencies, we are coming for you dm me if you'd like early access

Talk is cheap, show me the code. Code is cheap, show me the prompt.