Taylor Haun

The Attackathon begins October 27. To maximize your chance of success, the XRPL Attackathon Academy is now live. The Academy has walkthroughs, Devnet guides, test environments, and a C++ curriculum to help you prepare. Get started here: immunefi.com/academy/xrpl-ri…

Wild: Anthropic just proved that prompt engineering is dead. Context engineering is the new game, and it changes everything about building AI agents 🤖 For years, we obsessed over the perfect prompt. The right words. The perfect phrasing. The magic sentence that makes GPT-4 finally understand what you want. But Anthropic just dropped a reality check: Prompts are the easy part. Context is the bottleneck. Here's what they found: → LLMs have an "attention budget" like human working memory, but for tokens → More context = worse performance (they call it "context rot") → The transformer architecture creates n² relationships between tokens → Every token you add depletes focus exponentially Translation? Stuffing your AI with information is like trying to focus while 50 people scream at you. So what's the new playbook? ↳ Just-in-time retrieval - Instead of loading everything upfront, agents pull data dynamically using tools - Claude Code does this brilliantly: it uses grep, head, tail to analyze databases without ever loading full files into context ↳ Compaction - Summarize the conversation when you hit limits - Keep architectural decisions, discard redundant tool outputs - Maintain continuity without context bloat ↳ Structured note-taking - Agents write persistent notes outside the context window - Pull them back only when needed - Think: AI maintaining its own NOTES.md file ↳ Sub-agent architectures - Specialized agents handle focused tasks - Return compressed summaries (2k tokens) instead of raw exploration (50k+ tokens) - Separation of concerns at the AI level The shift is profound: Before: "How do I write the perfect prompt?" Now: "What's the minimal high-signal context that drives my desired outcome?" This isn't just academic. Goldman Sachs is using AI to draft S1s. Perplexity is automating financial analysis. Claude Code is refactoring entire codebases. None of this works if you drown the AI in context. The best AI engineers aren't prompt wizards anymore. They're context architects.