https://samuelagbede.comEssays and notes on technology, work, life, and the lessons I am learning in public.en-gbSun, 12 Jul 2026 23:18:10 GMThttps://samuelagbede.com/posts/borrowed-confidence-is-fragile-in-agentic-systems/https://samuelagbede.com/posts/borrowed-confidence-is-fragile-in-agentic-systems/Production-like evals revealed the retrieval architecture I actually needed and reminded me that confidence in agentic systems has to be earned, not borrowed.Mon, 06 Jul 2026 00:00:00 GMThttps://samuelagbede.com/posts/the-lethal-trifecta-in-ai-agents/https://samuelagbede.com/posts/the-lethal-trifecta-in-ai-agents/When agents can read private data, process untrusted content, and communicate outward, prompt injection becomes a much more serious security problem.Fri, 19 Jun 2026 00:00:00 GMThttps://samuelagbede.com/posts/how-plan-caching-reduces-llm-agent-costs/https://samuelagbede.com/posts/how-plan-caching-reduces-llm-agent-costs/Plan caching reuses planning templates across similar agent tasks, cutting cost and latency without throwing away accuracy.Wed, 17 Jun 2026 00:00:00 GMThttps://samuelagbede.com/posts/stop-filling-your-agents-context-window/https://samuelagbede.com/posts/stop-filling-your-agents-context-window/Bigger context windows do not remove failure modes. They create new ones when we stop being intentional about what goes into an agent's context.Fri, 05 Jun 2026 00:00:00 GMThttps://samuelagbede.com/posts/i-benchmarked-5-embedding-models-across-4-datasets/https://samuelagbede.com/posts/i-benchmarked-5-embedding-models-across-4-datasets/I benchmarked five embedding models across four NanoBEIR datasets and found that bigger embeddings did not always produce better retrieval.Thu, 21 May 2026 00:00:00 GMThttps://samuelagbede.com/posts/why-reranking-matters-with-cross-encoders/https://samuelagbede.com/posts/why-reranking-matters-with-cross-encoders/Bi-encoders make retrieval fast, but cross-encoders expose why reranking matters when meaning depends on the query.Tue, 19 May 2026 00:00:00 GMThttps://samuelagbede.com/posts/beginner-friendly-guide-gguf-model-format/https://samuelagbede.com/posts/beginner-friendly-guide-gguf-model-format/GGUF made local LLM inference feel practical by packaging model weights, vocabulary, hyperparameters, and architecture metadata into one runnable format.Mon, 11 May 2026 00:00:00 GMThttps://samuelagbede.com/posts/semantic-caching-in-production/https://samuelagbede.com/posts/semantic-caching-in-production/Repeated user intents can quietly inflate LLM cost and latency. Semantic caching helps, but production use comes with trade-offs.Fri, 27 Mar 2026 00:00:00 GMT