# Alex Ewerlof Notes Technical Leadership, Reliability Engineering, Growth This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## An experiment with Gemma 4 on a Raspberry Pi explores what an LLM would do with its own computer DevFeed: [An experiment with Gemma 4 on a Raspberry Pi explores what an LLM would do with its own computer]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-09-14T09:09:42Z Content type: opinion Language: en Sources: [Alex Ewerlof Notes]() Topics: [gemma4](), [Large Language Model](), [Raspberry Pi](), [Go Language](), [Simulation]() Tags: [computer](), [data](), [experiment](), [gemma-4](), [go](), [llm](), [patterns](), [raspberry-pi]() ### AI overview The author describes an experiment in which Gemma 4 12B was asked what it would do with its own computer. Running on a Raspberry Pi 1 through a Go harness, the model imagined ingesting large amounts of data, finding patterns, and running large-scale simulations. ### Source excerpt LLM's response ## Optimizing prompt cache DevFeed: [Optimizing prompt cache]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-08-08T14:07:13Z Content type: article Language: en Sources: [Alex Ewerlof Notes]() Topics: [Caching](), [Cache](), [LLMs]() Tags: [cache](), [caching](), [cost](), [token]() ### AI overview This visual article explains prompt caching in LLM APIs and its effect on token cost per task. It describes how requests are tokenized, prefetched into a key-value cache, and decoded, then discusses optimizing the cache to reduce costs. ### Source excerpt One of the best ways to save token cost ## AI Reliability Engineering DevFeed: [AI Reliability Engineering]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-07-12T17:34:19Z Content type: opinion Language: en Sources: [Alex Ewerlof Notes]() Topics: [AI Engineering](), [site-reliability-engineering](), [Large Language Model](), [Generative AI](), [Software Engineering]() Tags: [ai](), [ai-engineering](), [generative-ai](), [llms](), [reliability-engineering](), [software-engineering](), [sre]() ### AI overview This article examines how site reliability engineering practices can be adapted for AI systems and AI-generated black boxes. It focuses on the need to run these systems predictably, securely, and at scale as large language models and their supporting harnesses become more capable. ### Source excerpt Why SRE is a key skill in the age of AI-generated black boxes and how to renovate the traditional toolbox for the new era ## Sampling args in llama-server DevFeed: [Sampling args in llama-server]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-07-01T18:28:35Z Content type: tutorial Language: en Sources: [Alex Ewerlof Notes]() Topics: [llama.cpp](), [Large Language Model](), [Inference](), [quantization](), [Benchmark]() Tags: [benchmark](), [hallucinations](), [inference](), [llama-cpp](), [llm](), [quantization]() ### AI overview A reference guide to llama.cpp sampling and repetition parameters, covering common failure modes in local and quantized language models, parameter ranges and defaults, workload-specific tuning, modern sampling methods, and ways to accelerate experimentation. ### Source excerpt Reducing repetition, hallucinations, degradation, while making inference faster! ## Open source Chrome extension recreates LinkedIn from your data exports DevFeed: [Open source Chrome extension recreates LinkedIn from your data exports]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-06-28T17:41:27Z Content type: article Language: en Sources: [Alex Ewerlof Notes]() Topics: [Extension](), [Chrome extension](), [Open Source](), [App](), [data](), [CSV](), [GitHub](), [Single-page application (SPA)]() Tags: [app](), [chrome](), [chrome-extension](), [code](), [csv](), [extension](), [gdpr](), [github](), [offline](), [open-source]() ### AI overview The article introduces LinkedOut, a free open-source Chrome extension and offline single-page application that lets users inspect and browse data exported from LinkedIn. It describes the author's findings about the breadth of the export and explains how the app processes the data locally without gathering or transmitting it. ### Source excerpt An open source extension to recreate LinkedIn from your data exports ## Using local LLMs for agentic coding DevFeed: [Using local LLMs for agentic coding]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-06-04T09:01:34Z Content type: tutorial Language: en Sources: [Alex Ewerlof Notes]() Topics: [coding](), [agentic-coding](), [Large Language Model](), [GitHub Copilot CLI](), [Linux](), [llama.cpp](), [Ollama](), [NVIDIA RTX]() Tags: [agentic-coding](), [agents](), [ai](), [claude](), [coding](), [github](), [linux](), [llama](), [llama-cpp](), [llms](), [local](), [local-llms](), [microsoft](), [nvidia](), [nvidia-rtx](), [ollama](), [quality](), [rocm](), [state]() ### AI overview A practical guide to using local language models for agentic coding. It covers running local models, configuring coding agents such as Copilot and Pi, and evaluating local models against cloud models in terms of cost, privacy, tooling, and performance. ### Source excerpt AI honeymoon pricing is over, but your work is not ## Reliability Engineering for Air-Gapped Systems DevFeed: [Reliability Engineering for Air-Gapped Systems]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-04-03T21:03:06Z Content type: article Language: en Sources: [Alex Ewerlof Notes]() Topics: [observability](), [systems](), [Resilience](), [on-prem](), [Security](), [Requirements]() Tags: [observability](), [on-prem](), [reliability-engineering](), [resilience](), [security](), [supply-chain-attacks](), [systems](), [work]() ### AI overview This article discusses reliability engineering for air-gapped systems where software teams cannot access runtime data, metrics, or logs. It describes using SLI and SLO practices, alerting, secure operational processes, and on-premises maintenance to manage reliability in systems isolated from the internet. ### Source excerpt Tips and tricks to work around inaccessible observability ## Github Copilot vs Google Antigravity DevFeed: [Github Copilot vs Google Antigravity]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-03-22T23:19:06Z Content type: comparison Language: en Sources: [Alex Ewerlof Notes]() Topics: [GitHub Copilot](), [google-antigravity](), [GitHub Copilot CLI](), [Google](), [Chrome]() Tags: [browser](), [gemini-cli](), [github-copilot](), [google](), [google-antigravity]() ### AI overview A firsthand comparison of GitHub Copilot and Google Antigravity based on the author's experience using both tools. The article describes Antigravity's browser integration and Plan mode, while noting that its newer product maturity is visible compared with Copilot. ### Source excerpt Why Github gets developers and why it's hard to tell who Antigravity is for ## How AI firewalls protect production applications from prompt injection, data leakage, and excessive usage DevFeed: [How AI firewalls protect production applications from prompt injection, data leakage, and excessive usage]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-03-15T23:51:29Z Content type: tutorial Language: en Sources: [Alex Ewerlof Notes]() Topics: [Firewall](), [Security](), [prompt injection](), [LLMs](), [rate-limiting](), [pii]() Tags: [architecture](), [firewall](), [llms](), [pii](), [prompt-injection](), [rate-limiting](), [security]() ### AI overview This tutorial explains AI firewalls, or AI gateways, as reverse proxies with AI-focused inspection between an application backend and an inference provider. It covers ingress attacks such as prompt injection, egress risks including PII and secrets, rate limiting to control costs, latency considerations, implementation trade-offs, and layered defenses. ### Source excerpt How to protect your AI application in production against new classes of attacks ## OWASP Top 10 Agents & AI Vulnerabilities (2026 Cheat Sheet) DevFeed: [OWASP Top 10 Agents & AI Vulnerabilities (2026 Cheat Sheet)]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-03-10T18:18:04Z Content type: tutorial Language: en Sources: [Alex Ewerlof Notes]() Topics: [Vulnerabilities](), [Large Language Model](), [Artificial Intelligence](), [prompt injection](), [Security & Privacy]() Tags: [ai](), [llm](), [prompt-injection](), [recipe](), [security](), [security-privacy](), [techniques](), [vulnerabilities]() ### AI overview A pragmatic engineering guide examines the OWASP Top 10 for LLMs and the OWASP Top 10 for Agents. It explains how mixed instructions and data, unpredictability, agentic attack surfaces, reliability failures, cascading failures, and high LLM costs create security and operational risks, with examples and mitigations. ### Source excerpt A pragmatic engineering guide and cheat sheet for the OWASP Top 10 AI, OWASP Top 10 LLM, and OWASP Top 10 Agents vulnerabilities ## RAG vs SKILL vs MCP vs RLM DevFeed: [RAG vs SKILL vs MCP vs RLM]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-02-25T21:08:33Z Content type: comparison Language: en Sources: [Alex Ewerlof Notes]() Topics: [Retrieval Augmented Generation (RAG)](), [Large Language Model](), [long-context](), [Model Context Protocol]() Tags: [context-window](), [llms](), [mcp](), [rag]() ### AI overview This article compares RAG, SKILL, MCP, and RLM as techniques for making generalist LLMs more reliable on specialized tasks. It explains that RAG and RLM extend usable context, while SKILL and MCP provide external tool access, and discusses implementation mechanics, advantages, disadvantages, and when to use each approach. ### Source excerpt Comparing various techniques to make the models more reliable while working around context window limitation ## Multi-Agent System Reliability DevFeed: [Multi-Agent System Reliability]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-02-19T20:41:37Z Content type: tutorial Language: en Sources: [Alex Ewerlof Notes]() Topics: [Multi Agent Systems](), [LLMs](), [Artificial Intelligence](), [systems]() Tags: [architecture](), [llms](), [multi-agent](), [parallel](), [reliability-engineering](), [systems-engineering]() ### AI overview This article presents four architecture patterns--hierarchy, consensus, adversarial debate, and knock-out--to improve the reliability of multi-agent systems. It explains that underlying LLMs remain prone to hallucination, logical fallacies, and context drift, and that multi-agent architectures can amplify these errors and make debugging harder. ### Source excerpt 4 patterns to tame multi-agent systems for reliability ## A Seven-Level AI Fluency Framework for Knowledge Workers DevFeed: [A Seven-Level AI Fluency Framework for Knowledge Workers]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2026-01-30T18:14:37Z Content type: tutorial Language: en Sources: [Alex Ewerlof Notes]() Topics: [AI Development](), [Prompt Engineering](), [Retrieval Augmented Generation (RAG)](), [context](), [Development]() Tags: [ai](), [ai-development](), [context-engineering](), [guide](), [prompt-engineering](), [rag](), [technical]() ### AI overview This guide presents a seven-level, directional framework for assessing and developing AI fluency among knowledge workers. It describes progression from prompt engineering toward RAG, model ROI, context engineering, and deterministic AI system architecture, with applications for learners, leaders, product teams, consumers, and hiring managers. ### Source excerpt Transition from prompt "engineering" to AI system architecture with this 7-step AI fluency framework. Designed for software engineers and SREs, this guide provides a roadmap for mastering RAG, context engineering, and the critical shift from probabilistic to deterministic AI development. ## Foundation vs. Instruct vs. Thinking Models DevFeed: [Foundation vs. Instruct vs. Thinking Models]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2025-12-24T07:07:00Z Content type: opinion Language: en Sources: [Alex Ewerlof Notes]() Topics: [foundation-models](), [Large Language Model](), [AI-assisted coding]() Tags: [ai](), [code-completion](), [foundation](), [llms](), [models]() ### AI overview This article explains the distinction between base and instruction-oriented language models using a software metaphor. It describes base models as pretrained systems that predict the next token from learned patterns, without an inherent concept of questions, answers, or instructions, and discusses code completion as one use case. ### Source excerpt A Senior Engineer's Mental Model for AI ## Emergent properties DevFeed: [Emergent properties]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2025-12-05T13:23:18Z Content type: tutorial Language: en Sources: [Alex Ewerlof Notes]() Topics: [Resilience](), [systems](), [Latency]() Tags: [api](), [database](), [latency](), [leadership](), [resilience](), [systems](), [systems-engineering](), [technical]() ### AI overview This tutorial explains emergent properties in systems engineering, distinguishing nominal, weak, and strong emergence from resultant properties. It discusses how interactions and dependencies between system parts affect reliability, maintainability, predictability, and cost, with examples for technical leadership and engineering. ### Source excerpt What are nominal, weak and strong emergent properties and how to identify and mitigate their negative impact in system design? ## AI Systems Engineering Patterns DevFeed: [AI Systems Engineering Patterns]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2025-11-30T11:56:00Z Content type: article Language: en Sources: [Alex Ewerlof Notes]() Topics: [Artificial Intelligence](), [AI Engineering](), [systems]() Tags: [ai](), [ai-engineering](), [engineering](), [patterns](), [systems-engineering](), [techniques]() ### AI overview The article presents 30 techniques from conventional systems engineering that can be applied to AI engineering. ### Source excerpt 30 techniques from conventional system engineering to supercharge AI Engineering ## Three Levels of Writing: Thinking, Communication, and Innovation DevFeed: [Three Levels of Writing: Thinking, Communication, and Innovation]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2025-11-21T18:09:39Z Content type: opinion Language: en Sources: [Alex Ewerlof Notes]() Topics: [iteration](), [Artificial Intelligence]() Tags: [ai](), [communication](), [generative-ai](), [innovation](), [iteration](), [writing]() ### AI overview The article presents three levels of writing: organizing thoughts and reaching clarity, communicating ideas, and pushing the boundaries of the unknown. It also discusses how generative AI has changed the author's writing process and argues that writing remains relevant in the age of generative AI. ### Source excerpt A leveling model to use writing for thinking, communication, and innovation ## Fitting parts DevFeed: [Fitting parts]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2025-11-12T14:37:00Z Content type: opinion Language: en Sources: [Alex Ewerlof Notes]() Topics: [distributed-systems](), [systems]() Tags: [distributed-systems](), [pitfalls](), [systems]() ### AI overview The article argues that system performance depends on how parts fit and interact, rather than on optimizing each part separately. It applies this idea to hiring for culture add and to technology choices in complex distributed systems, cautioning against enforcing a single technology stack or relying on a golden-hammer solution. The supplied text ends mid-discussion, so the article's complete recommendations are not available. ### Source excerpt What does it mean? How is it abused? And how to work around the pitfalls? ## AI application topologies: cloud, edge, local, and hybrid inference DevFeed: [AI application topologies: cloud, edge, local, and hybrid inference]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2025-10-24T21:15:00Z Content type: article Language: en Sources: [Alex Ewerlof Notes]() Topics: [Artificial Intelligence](), [Cloud](), [Local AI](), [Inference](), [On-device AI](), [Latency](), [Software as a service]() Tags: [ai](), [cloud](), [edge](), [edge-ai](), [inference](), [latency](), [local-ai](), [on-device-ai](), [privacy](), [saas]() ### AI overview The article categorizes AI application topologies by where inference compute occurs relative to data: cloud, edge, local, and hybrid. It describes trade-offs involving capability, latency, cost, privacy, connectivity, vendor limits, and centralized-service outages. ### Source excerpt Cloud AI, Edge AI, Local AI, and Hybrid AI ## Introducing: Local Browser AI DevFeed: [Introducing: Local Browser AI]() Original publisher: [Read original article]() Author: Alex Ewerlöf Published: 2025-10-13T07:51:29Z Content type: article Language: en Sources: [Alex Ewerlof Notes]() Topics: [browser](), [Artificial Intelligence](), [Browser Extension](), [JavaScript](), [On-device AI]() Tags: [ai](), [browser](), [extension](), [interface](), [javascript](), [on-device-ai](), [open-source]() ### AI overview This technical post explains how the Prompt API enables local language-model chat in the browser and examines the Local Browser AI extension, which is built with plain JavaScript and native browser APIs. It also describes the extension's permissions, privacy properties, open-source MIT license, and supported platforms. ### Source excerpt Using the new Prompt API for local chat in the browser