Dev

<p>I started building xAgent in April 2025.</p> <p>The original idea was straightforward: build a task-oriented Agent that could run work on its own and turn AI into real automation. Looking back, that sentence sounds simple. Most of what I have done over the past year has been filling in everything hidden inside the words “run work on its own.”</p> <p>The first version used a single Agent. I quickly ran into a problem: once the prompt focused its attention on one kind of work, the Agent could d

<p>I spent a while getting permission levels right. What an agent may read, what it may write, what needs a human. That work was worth doing, and it did not save me here.</p> <p>The gap is simple to state and easy to miss: permission levels answer <em>what may this agent do</em>. They say nothing about <em>what happens when the task itself changes after the agent already has permission</em>.</p> <h2> The situation </h2> <p>When work is handed to an agent, the handoff carries a task definition, a

<p>I’ve been building a side project called <strong>OopsCalorie</strong>, an AI-powered calorie and meal tracking app.</p> <p>The idea sounded simple enough:</p> <ol> <li>User logs or takes a photo of their food.</li> <li>AI identifies the meal.</li> <li>Estimate calories and macros.</li> <li>Save the entry.</li> <li>Done.</li> </ol> <p>Simple, right?</p> <p>Well...</p> <p>Then we started testing it with Filipino food. 😂</p> <h2> AI Meets Filipino Food 🇵🇭 </h2> <p>One of the funniest parts of

<p>SpaceXAI recently introduced Grok Bot, an always-on AI-agent service designed to work more like an autonomous teammate.</p> <p>The agents have their own cloud computer environment and can log into applications, websites and tools to perform multi-step tasks. They can also operate in parallel and coordinate with other agents.</p> <p>The product is entering a market that already includes competing agentic workplace products from OpenAI, Anthropic and Microsoft.</p> <p>Traditional chatbot:<br> <

<p>AI agents can only use tools as reliably as those tools are described.</p> <p>That’s why I built <a href="https://toolready.dev" rel="noopener noreferrer">ToolReady AI</a>—a free tool that reviews MCP and AI-agent tool schemas, identifies reliability problems, and recommends specific fixes.</p> <p>A function might work perfectly when a developer calls it directly, yet still fail when an agent has to decide when to call it, which arguments to provide, and what values are safe. In many cases, t

<p>Last quarter, we ran our production voice AI receptionist — Loquent — across four different TTS engines simultaneously, split-testing real patient calls at dental and healthcare clinics. The results surprised us: the most "natural sounding" engine in demos performed the worst with actual patients.</p> <h2> Why We Ran This Test </h2> <p>At Autor, we've been running Loquent in production for over a year now. It handles thousands of automated calls per month for healthcare and dental clinics acr

<p>Where is SRE going? I've been watching the trends closely. Here's my best guess for the next 5 years.</p> <h2> Trend 1: AI becomes the default copilot </h2> <p>Not replacing SREs. Sitting next to them. Every major incident response tool will have AI built in — not as a gimmick, but as the default way you interact. Log queries written by natural language. Post-mortems drafted automatically. Runbooks generated from historical incidents.</p> <p>The SREs who adapt first get 2-3x productivity gain

<p>Yesterday I stumbled on a <a href="https://lnkd.in/p/eFPzw29C" rel="noopener noreferrer">LinkedIn post</a> about a new open standard for packaging Agent Skills and MCP servers: <a href="https://agent-plugins.org/" rel="noopener noreferrer">Agent Plugins</a>, built with collaboration from players like Vercel, OpenAI, Microsoft, AWS, GitHub and Cursor.</p> <p>I won't lie: at first, I felt that familiar punch in the stomach.</p> <p>A while ago, I had already felt the first signal when I read an

<p>In June I wrote a piece called "Investment outlook for the second half of 2026." Middle East risk versus the AI trade. Three scenarios. Seven things to watch.</p> <p>It got zero likes.</p> <p>I understand why now. There wasn't a single line in it that only I could have written. The oil price outlook, the Fed's stance, the P/E on the Nikkei — I had carefully rearranged that day's news. The numbers were borrowed and the judgment was commentary.</p> <p>But one of those seven items, I remember th

<p>If you build or maintain websites for local businesses, you've probably already noticed organic click-through rates dropping even as rankings hold steady. The reason: <strong>AI Overviews now appear in roughly 68% of local searches</strong> (versus ~39% for the classic three-pack local results), and for "near me" or pricing-style queries, AI-generated answers show up <strong>80–97% of the time</strong>. ChatGPT itself is reportedly the <strong>third most-used source for local recommendations<

<h2> What’s actually happening </h2> <p>Voice-cloning technology has crossed a threshold: audio experts and federal agencies now warn that cloned voices can sound close enough to the real thing that listeners can’t reliably tell the difference by ear alone. Scammers only need a short clip of someone’s voice — often scraped from a public social media video — to generate a convincing fake in that person’s voice, saying whatever the scammer types.</p> <p>The most common version, which the FTC and F

<p>Not a typo. One hundred and twenty-eight versions, published to npm and Homebrew, with CI,<br> contract tests, a documentation site, bilingual docs, a governance policy, and a sponsor<br> recognition policy. Zero external users. Then I did it again — a second product, 442 passing<br> tests, real cost metering, event-sourced audit logs — and stopped two days after the first<br> end-to-end run finally worked.</p> <p>I'm writing this because I went back through both codebases with fresh eyes (an

<p>My client was drowning in support tickets. Same questions, every single day.</p> <p>Their team was spending 20+ hours a week just answering repetitive queries. That's when they came to me — and I built them an AI chatbot that changed everything.</p> <p>The Problem<br> The store was getting ~150 support messages per week. 80% of them were the same 10 questions. This was slow, expensive, and burning out the team.</p> <p>The Stack<br> OpenAI API (GPT-4)<br> Node.js + Express.js<br> Shopify REST

<ul> <li> How Can AI Help Lazy Programmers? <ul> <li>Being a Lazy Programmer</li> <li>The code left by diligent programmers</li> <li>Resources for Lazy Programmers before AI</li> <li>Code Generators</li> <li>Built-in Framework Modules</li> <li>App Generator</li> <li>Model to Code</li> <li>4GL, DSL and 5GL</li> <li>AI Code Agents</li> <li>How AI has been helping me?</li> <li>How Visual Studio and alike with AI has been helping me?</li> <li>How AI compensates my poor memory</li> <li>AI dealing wit

<h1> Response to Ollama Production Queries to the previous post </h1> <p>Excellent questions — you've clearly been in the trenches with Ollama at scale! Let me address your specific concerns about our production setup, now updated with our actual deployment architecture.</p> <h2> On GPU Residency & VRAM Thrashing </h2> <p>You're absolutely right — this is the <strong>single biggest operational challenge</strong> with Ollama in production. We've implemented a hybrid strategy:</p> <h3> Our Approac

<p>I run an AI agent (Claude Code) that produces content and takes it all the way to publishing on external platforms, on its own.</p> <p>I had exactly one rule. <strong>Publishing is irreversible, so right before it posts, it must always get my OK.</strong></p> <p>Anything you publish or send to the outside world lives in a place where deleting it doesn't help — a copy survives. So "make things" it can do freely, but one step before "ship it," it has to get a human (me) to approve. The approval

<p>This week I started the Agentic Engineer Nanodegree through <a href="https://aws.amazon.com/about-aws/our-impact/scholars/" rel="noopener noreferrer">AWS AI & ML Scholars</a>, the next step in what has become my increasingly formal attempt to understand the things I started building before I knew enough to understand all of them.</p> <p>Somewhere in the coursework, ReAct appeared.</p> <p>Record scratch.</p> <p><strong>ReAct?</strong></p> <p>There was an immediate sense of déjà vu, although no

<p>Google has renamed NotebookLM as <a href="https://scalevise.com/resources/notebooklm/" rel="noopener noreferrer"><strong>Gemini Notebook</strong></a>, tying its source-grounded research workspace more closely to the wider Gemini ecosystem. The change is more than a new label: Google says notebooks will sync across the Gemini app and Google Search as the rollout expands, creating a broader home for organizing sources, chats, and AI-assisted analysis.</p> <p>The rebrand provides the clearest co

<p>A new GitHub project titled "Waste" aims to tackle the challenges of running enormous AI models in environments with limited RAM. By leveraging a streaming mechanism to pull activated weights directly from NVMe storage, developers can execute the hefty 2.78-trillion-parameter Kimi K3 model without the need for excessive memory allocation. This method could significantly affect engineering teams dealing with high parameter count models, particularly when physical memory constraints can stymie

<p><em>A journey into DDC/CI — the remote-control protocol that has been hiding inside your display cable for over twenty years, and how it turned my monitor into a KVM.</em></p> <h2> The desk that started it all </h2> <p>My setup looks ordinary: a MacBook Air open on the desk (my primary machine), and an LG 32UN880K — a 32" 4K monitor with two HDMI ports, DisplayPort, and USB-C. The less ordinary part is behind the MacBook: an ASUS gaming laptop running Linux whose internal screen died. It now

<p>In Part 1, I explained why Enola extracts deterministic architectural facts before an AI agent begins reasoning.</p> <p>That leaves the next design question:</p> <p>How should those facts be represented?</p> <p>The obvious answer is to build a graph containing every symbol, dependency, route, service, and repository.</p> <p>But the difficult part is not putting nodes and edges into a graph.</p> <p>It is preserving what those relationships mean.</p> <p>A function call is not the same as a pack

<p>Google's <a href="https://scalevise.com/resources/notebooklm/" rel="noopener noreferrer">Gemini Notebook rollout</a> has expanded across subscriber and account groups, but Google's published information describes a <strong>staged, multi-phase release</strong>, not a confirmed global completion for every Google AI Pro subscriber. That distinction matters for users evaluating access, planning source-based research workflows, or deciding whether a subscription tier will immediately provide the e

<h2> Why Decouple Reasoning from Motor Control </h2> <p>General-purpose robots have to pull off two very different jobs at once. They need to read a cluttered, full-room visual scene, hold a multi-minute plan in memory, and converse with a person — and, in the same instant, close a high-frequency control loop that keeps a balancing humanoid upright and moves a delicate hand without dropping whatever it holds. Cramming both jobs into a single end-to-end network forces uncomfortable trade-offs: th

<p>Two pricing stories dropped this week that are worth a pause if you're building on LLM APIs.</p> <h2> DeepSeek V4 Flash exits preview — and undercuts its own flagship </h2> <p>DeepSeek V4 Flash left preview at <strong>$0.14 / $0.28 per million tokens</strong> (input/output) — and it's beating its own larger Pro model on agentic benchmarks, hitting <strong>82.7% on Terminal-Bench</strong>. That's a smaller, cheaper model outperforming its own bigger sibling on agent tasks.</p> <h2> Claude Sonn

<p>There is a French phrase for it. L'heure entre chien et loup, the hour between dog and wolf. It is that stretch after the sun goes down, when the light has gone soft and a shape coming toward you over the far hill could be your own dog heading home or a wolf that has been watching you. You cannot quite tell. The light is vague enough that, for a little while, the thing you love and the thing you fear wear the same outline.</p> <p>I find myself thinking about it a lot these days. Only what has

<p>Google’s April 2026 Gemini Drop broadens how the Gemini app connects across desktop and mobile experiences. The update adds a native macOS app, integrates NotebookLM into Gemini conversations and introduces new ways to organize chats. It also sits within Google’s wider push toward agent-like task automation, including Gemini Agent for eligible Ultra subscribers.</p> <p>The company’s <a href="https://blog.google/innovation-and-ai/products/gemini-app/gemini-drop-april-2026/" rel="noopener noref

<p>Hello Devs 👋</p> <p>If you've looked at AI code review tools recently, you've probably seen the term <strong>"context-aware code review"</strong> everywhere.</p> <p>Every tool seems to use it.</p> <p>Some talk about repository awareness. Others mention codebase understanding. A few claim they understand architecture and engineering best practices.</p> <p>After spending some time reading docs, testing different tools, and comparing how they review pull requests, I realized something:</p> <p>M

<p>A renewal agent can call the CRM, email, calendar, support, document search, and contract systems. On its first run, it asks every system for everything related to one customer.</p> <p>The result looks thorough: hundreds of CRM fields, years of ticket history, complete email threads, and the same contract attached to several messages. But the agent now has to spend time and tokens deciding which records matter before it can make progress on the renewal.</p> <p>The agent is well connected. It

<p>AI coding agents are strangely lopsided. Hand one a well-scoped task and it writes better code, faster, than I do. Hand it a <em>large</em> feature, the kind that takes days and touches a dozen files across several layers, and the failure mode changes. The context window fills up and quality quietly degrades (AKA <strong>context rot</strong>). A fresh session, missing the history, confidently invents an API that never existed. And even when the code is good, you're left staring at a 3,000-lin

<p>You are a software engineer. Your craft honed through years of careful practice. Then suddenly, there are these chatbots and agents. Overnight, your colleagues got a new title on LinkedIn: "AI engineer". Some are already SENIOR AI engineers. You're curious about this new world, and might want to catch up and become part of it yourself.</p> <p>If this is you, then join me on this tour through the concepts and patterns that make up the field of AI engineering. We will find that AI application d

<p>TL;DR. Bringing AI coding agents into a German operation is not only a technical decision. The moment they meet real systems and real teams, they also meet three things that international AI content almost never mentions: a data processing agreement, the works council, and the question of who owns the generated code. None of this is a footnote. It decides whether your AI project reaches production or stalls in legal review. Here is the layer most vendors forget, and the fast order to clear it

<p>OpenAI has formally outlined a national science initiative designed to connect <a href="https://scalevise.com/resources/openai-frontier-model-access-academic-researchers/" rel="noopener noreferrer">frontier AI models</a> with government research infrastructure, National Laboratories, universities, and working scientists. The program is not a single model launch. Instead, it combines funded access, early product access, scientific campaigns, and an emphasis on fitting advanced AI into real res

<h2> TL;DR </h2> <p>I designed a proposed AWS Transform Custom community skill that prepares Glue 2.0, 3.0, and 4.0 repositories for Glue 5.0. It separates safe mechanical transformations from changes that require human evidence, generates a migration report, and preserves already-compatible files unchanged. Because I didn't have live <code>atx</code> access, the benchmarks in this post are explicitly labeled manually simulated, not agent-executed. The proposal is open as <a href="https://github

<p>If you just want the recommendation: for the ordinary AI-generated image an inference job hands back — a 2 to 8 MB PNG — do one plain object PUT into your S3-compatible storage and stop there, because multipart upload only earns its complexity when a single artifact is big enough that losing a transfer halfway through costs you real money to redo, which for my team starts somewhere north of 100 MB.</p> <p>Everything below is about that threshold, and about the operations bill you pick up the

<p>OpenAI has published a post-mortem examining an unusual pattern in its model testing: recurring references to “goblins” and “gremlins” in model outputs. The company’s <a href="https://openai.com/index/where-the-goblins-came-from/" rel="noopener noreferrer">official post, “Where the goblins came from”</a>, published on April 29, 2026, frames the behavior as an emergent effect of reinforcement learning and human-feedback dynamics, not as a new product feature. Its practical message is more cons

<h1> What Is Temperature in AI? (And How to Stop Getting Poetry When You Asked for a Grocery List) </h1> <p>Remember Magic 8-Balls? Those plastic oracles you'd shake for life advice, only to get "Reply hazy, try again" when you asked if your crush liked you back? Imagine someone added a little dial on the bottom. Turn it all the way to zero and the thing becomes painfully predictable, only ever offering "Yes" or "Most likely." Crank it all the way up and suddenly it's inventing answers that neve

<p>If you've ever built an e-commerce store, you know the drill: storefront, hosting, payment gateway, inventory, shipping, support, security, and analytics.</p> <p>Miss one — and the whole thing breaks.</p> <p>That framework works for human commerce.</p> <p>But what about commerce for AI agents?</p> <p>Over the past few months, I've been building AgentShare — an MCP server that provides Solana DeFi data to autonomous AI agents, with pay-per-call via x402 USDC.</p> <p>Along the way, I realized t

<p>Claude's context window is a game-changer for AI development, but how does it really work? We'll dive into the details and explore what this means for your AI projects.</p> <h2> Introduction to Claude's Context Window </h2> <p>The context window is a critical component of Claude's success, but its inner workings are often misunderstood. To understand why the context window matters, let's first define what it is: a context window is a mechanism that allows Claude to capture and process the con

<h1> The Regulatory Capture of Compute: Why Frontier AI’s Call for Government Intervention is an Infrastructure Play </h1> <p>An unprecedented coalition of researchers and engineers from OpenAI, Anthropic, Google, Meta, Microsoft, and Mistral has signed a joint statement calling for government-coordinated oversight—and potential development slowdowns—of frontier AI systems. Ostensibly framed as a proactive measure against catastrophic risks and the challenges of "automated AI," this collective a

<p>Article Summary</p> <p>Many sales organizations still forecast by asking representatives to enter expected revenue, applying manager judgment, and aggregating the result in a spreadsheet. The number is difficult to explain and even harder to improve.</p> <p>This guide builds a reproducible sales-forecasting workflow for a B2B SaaS company: CRM extraction, data quality, stage calibration, baseline pipeline forecasting, machine-learning win probabilities, close-date distribution, scenarios, Pow