Exploring AI Agents, Prompt Engineering, and the Future of Automationhttps://chriskarani.xyz/en-ushttps://chriskarani.xyz/posts/the-command-you-approve-is-not-the-command-that-runs/https://chriskarani.xyz/posts/the-command-you-approve-is-not-the-command-that-runs/Why approved commands run as something else: shell expansion rewrites what your agent's safety check read. A case for guarding at execution.Tue, 25 Aug 2026 00:00:00 GMTengineeringswiftai-agentssecurityChris Karanihttps://chriskarani.xyz/posts/wax-one-file-survives-hard-kills-and-context-resets/https://chriskarani.xyz/posts/wax-one-file-survives-hard-kills-and-context-resets/I built Wax, a single-file crash-safe memory engine for AI agents on Apple Silicon: WAL durability, hybrid RAG, bitemporal facts, no cloud dependency.Thu, 28 May 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/swarm-websearch-tool-4k-ondevice/https://chriskarani.xyz/posts/swarm-websearch-tool-4k-ondevice/I built Swarm's Swift web tool for 4,096-token on-device LLMs. Here's the tiered envelope, the Wax-without-embeddings bet, and fourteen decisions behind it.Wed, 22 Apr 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/deep-research-3b-4k-foundation-models/https://chriskarani.xyz/posts/deep-research-3b-4k-foundation-models/A 3B on-device model with a 4K context window produced a 2,336-word grounded research report using live web search. Here's the architecture that made it work.Thu, 09 Apr 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/edgerunner-beats-llamacpp-swift-metal-llm/https://chriskarani.xyz/posts/edgerunner-beats-llamacpp-swift-metal-llm/Built EdgeRunner in pure Swift and Metal from scratch over a weekend. On Qwen3-0.6B Q8_0, it hits 212 tok/s on M3 Max, beating llama.cpp by 16%. Here's what wTue, 31 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/throughput-breakthrough-program-without-lying-to-ourselves/https://chriskarani.xyz/posts/throughput-breakthrough-program-without-lying-to-ourselves/A throughput research postmortem from Espresso: what we tested, what failed, what stayed slower than baseline, and where the real architectural blockers are.Sat, 28 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/conduit-unified-swift-llm-sdk/https://chriskarani.xyz/posts/conduit-unified-swift-llm-sdk/If you've shipped AI features in an iOS app recently, you know the drill. Each provider has its own SDK. Switching means rewriting everything. Conduit fixes tWed, 25 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/contextcore-gpu-memory/https://chriskarani.xyz/posts/contextcore-gpu-memory/We built ContextCore to manage conversation context on Apple Silicon without cloud or battery drain. Four-tier memory, Metal shaders, 63M chunks/sec, sub-5ms Wed, 25 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/edgeunner-architecture-deep-dive/https://chriskarani.xyz/posts/edgeunner-architecture-deep-dive/We fully documented EdgeRunner, a Metal LLM inference engine for Apple Silicon. Here's what we found that was actually interesting.Wed, 25 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/on-device-observability/https://chriskarani.xyz/posts/on-device-observability/How local telemetry transforms AI reliability, privacy, and performance for applications running on edge devices.Wed, 25 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/colony-apple-agent-runtime/https://chriskarani.xyz/posts/colony-apple-agent-runtime/Colony is a Swift framework that orchestrates AI agent loops on iOS and macOS using Apple's Foundation Models. Here's why it exists and how it works.Tue, 24 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/swift-for-coding-agents/https://chriskarani.xyz/posts/swift-for-coding-agents/After building a multi-agent framework in Swift, I have thoughts on why it might actually be a better fit than Python or TypeScript for this kind of work.Tue, 24 Mar 2026 00:00:00 GMTengineeringimportedChris Karanihttps://chriskarani.xyz/posts/teaching-machines-step-by-step/https://chriskarani.xyz/posts/teaching-machines-step-by-step/How chain-of-thought prompting and simple techniques transformed large language models from pattern-matchers into genuine problem-solvers.Sat, 29 Nov 2025 00:00:00 GMTPrompt EngineeringPromptingChain-of-ThoughtAI ReasoningSelf-ConsistencyTree-of-ThoughtsChris Karanihttps://chriskarani.xyz/posts/ai-consciousness-ethics-research/https://chriskarani.xyz/posts/ai-consciousness-ethics-research/Exploring the uncertainty around AI consciousness, ethics implications, and what researchers at Anthropic are learning about model welfare.Fri, 28 Nov 2025 00:00:00 GMTLLMs & ModelsAI EthicsConsciousnessModel WelfareAnthropicPhilosophyChris Karanihttps://chriskarani.xyz/posts/file-system-interface-for-agents/https://chriskarani.xyz/posts/file-system-interface-for-agents/Why the serious agent stacks are converging on files as the primitive for machine behavior, and what that means for the future of AI systems.Thu, 20 Mar 2025 00:00:00 GMTengineeringimportedChris Karani