标签：#Engineering

Originally published on Mohamad Alsabbagh's Blog . AI coding assistants are excellent at compressing known work into fast drafts. That speed is the preface boost : routine implementation arrives almost immediately. The hidden risk is that teams begin treating the model's first plausible answer as architecture. Because LLMs are trained and aligned around historically common patterns, they can pull engineering teams toward Generative Monoculture : less diverse solutions, narrower exploration, and fewer designs shaped by the exceptional constraints of the system in front of them. Give the same prompt to three engineers using the same assistant and you often get the same shape back: a tidy service layer, a familiar API boundary, a conventional retry wrapper, and code that looks clean enough to merge. That answer is useful. It may even be the right answer for ordinary work. The danger is what happens when ordinary work becomes the default posture for extraordinary constraints. Large Language Models are not neutral architecture engines. They are probabilistic systems trained over historical work and tuned toward answers people tend to reward. Used well, that makes them extraordinary accelerators. Used passively, it creates an optimization paradox: teams gain immediate implementation velocity while becoming anchored to a consensus baseline that may be too average for the actual system. 1. The Default Is a Local Optimum Wu, Black, and Chandrasekaran define Generative Monoculture as a narrowing of model output diversity relative to the diversity available in the training data. That matters because software architecture is rarely a search for the most common answer. It is a search for the answer that fits the exact failure modes, latency envelope, team topology, regulatory constraints, and operational reality of a system. The model's default is often a local optimum: a solution that is statistically likely, syntactically polished, and broadly acceptable. That can be excellent

Most engineering reading lists are optimized for knowledge accumulation. Modern engineering rewards bottleneck elimination. Last week, a junior engineer showed me a "Top 10 Books Every Engineer Should Read" list. It looked almost identical to the lists I saw ten years ago. The same classics. The same process books. The same assumption: Read enough books and you'll become a better engineer. That's not how most high-performing teams learn. The best engineers I know don't build learning plans around books. They build learning plans around constraints. The Problem with standard reading lists Most reading lists assume that knowledge is universally valuable. In practice, engineering value is highly contextual. A backend engineer struggling with database contention does not need another chapter on Agile. A team spending thousands of dollars per month on LLM inference does not need a generic software craftsmanship book. A startup fighting latency issues does not need a leadership framework. They need solutions to the bottleneck directly in front of them. Reading lists rarely account for this. They optimize for completeness. Engineering rewards relevance. The Shift Most Engineers Miss The fundamentals still matter. Distributed systems matter. Databases matter. Networking matters. Operating systems matter. They are not obsolete. But they are no longer sufficient. Modern systems introduce constraints that barely existed a few years ago: AI inference costs Context window limitations Agent orchestration Evaluation pipelines Semantic caching Non-deterministic workflows Model routing Human-in-the-loop systems Many traditional reading lists never touch these problems. Yet these are exactly the problems teams are solving every day. The challenge is no longer simply writing correct software. The challenge is building reliable systems on top of components that are inherently probabilistic. What Changed For decades, engineers mostly worked with deterministic systems. Given the same inp

Working with AI Means Thinking More, Not Less Yes, this text is long. Yes, it repeats itself in places. I did not clean that up. A text that sounded too smooth while arguing that AI forces you to think more, not less, would be at least slightly dishonest. This is not fast food for quick consumption. And yes, don’t worry: you won’t hear anything especially new here. That is part of the problem too. There is a popular and very seductive story about AI in software development. Now that the machine can write code, the human gets to think less. You just point it in the right direction, and the model will quickly and cheaply do a significant part of the work on its own. In that picture, AI is primarily an accelerator for code production, and human thinking gradually shifts from necessity to optional extra. I keep feeling more and more strongly that this description is dangerously wrong. A more accurate formula for my own experience right now is this: I’m the tech lead, the AI is the entire team in one body . And if you take that metaphor seriously, the conclusion is the exact opposite of the mainstream narrative. Working with AI is not a way to think less. It is a mode in which you need to think more, not less . Not because the AI is bad. But because it is too good at one very treacherous thing: it confidently and smoothly fills in what was left unsaid. I’m the tech lead, the AI is the team At first this metaphor felt like a neat formulation. Now it feels like a literal description of what is going on. If you treat AI as a very fast and very capable executor, a lot of things become clearer immediately. It really can wipe out months of routine work. It can spin up prototypes quickly, take over test scaffolding, try out alternatives, make local edits, help break a task into parts, and sometimes even suggest a decent direction. On the surface, this really does look like a silver bullet. Especially if the human knows the stack and can read code. The pace becomes so extreme th

Hello, I'm Maneshwar. I'm building git-lrc, a Micro AI code reviewer that runs on every commit. It is...

A working prompt library is the main event, not an appendix. The industry still treats prompts as some half-baked spitball left in a README, or, worse, a plaintext blob stapled to package.json and forgotten. That's a waste of compute and credibility. What powers reliable AI-assisted refactoring, onboarding, or even next-gen code IDEs is not the size of the model but the clarity and context supplied by the actual, shipped prompt set. OTF kits turn this lesson into a repeatable deliverable: every paid template includes 20+ production-tested prompts tied to the real file structure, component API, and product-specific conventions. This is not a suggestion; it's structural. The takeaway: a real prompt library is as important as your component library. Treat it like one. Start with the pain: why blank chat boxes don't scale The web is full of “integrations” that paste a blank chat input over your codebase and call it an “AI coding assistant.” The result: hallucinated function names, invented conventions, broken import paths. Here’s what happens in real life: Dev: "Add a social login button." AI (blank prompt): "Sure! Insert <SocialLoginButton> in your LoginScreen.js." Dev: (There’s no such component. There's not even a LoginScreen.js.) Short: A generic prompt with zero context simply can't know your conventions, files, or patterns. The agent will either fail, hallucinate, or pepper you with clarifying questions you have already answered in your product architecture. Takeaway: Prompting without context is coding without types — fragile guesses instead of structured outcomes. What a first-class prompt library enables When the prompt library ships with the codebase, it looks like this: Every prompt knows the folder structure (e.g., features/auth , screens/Settings/index.tsx ). Conventions are hard-coded: naming, import styles, design token usage. Endpoints and integration points (e.g., “update the Stripe webhook handler in api/webhooks/stripe.ts ”) are spelled out. The promp

Most agents drag their entire past into every turn. A better default: keep a thin index of what was said hot, and fetch only the few turns you actually need — intact, on demand. Code: github.com/NirajPandey05/jit_context There is a quiet assumption baked into how most agents handle memory: that more context is safer than less. If the model might need something, put it in the window. The conversation grows, every prior turn rides along on every new request, and we trust the model to find the part that matters. That assumption breaks twice. It breaks on cost , because an agent loop re-sends its whole window on every step — a hundred stale turns aren't paid for once, they're paid for on turn 101, 102, and every step after. And it breaks on quality , because models don't read a long window evenly. Relevant facts buried in the middle get underweighted; irrelevant bulk competes for attention with the thing that actually answers the question. Past a point, a bigger context produces a worse answer, not just a costlier one. So the interesting question isn't "how do we fit more in?" It's "how do we keep the window small and dense without losing the one old turn that matters?" This post is the design we built around that question — for the specific case of long conversation history — plus the benchmark we used to keep ourselves honest. 01 · The mechanism: a hot index over a cold store The design borrows directly from how computers have always managed memory that doesn't fit: a small fast tier that's always present, a large slow tier that holds the bulk, and a rule for moving things between them. Virtual memory pages between RAM and disk. We page between the context window and an external store — for attention instead of address space. Concretely, there are two tiers. The cold store holds every turn at full fidelity, keyed by id — nothing is thrown away. The hot index holds one compact entry per turn: a short summary, a little metadata (entities, whether the turn recorded a dec

A few months ago, I was overwhelmed by everything happening in AI. Every week there was a new coding assistant, a new workflow, or someone claiming they built an app in just a few hours. It felt like if you weren't keeping up, you'd be left behind. I tried almost everything. Cursor. ChatGPT. Claude Code. Lovable. At first, I kept switching between tools, hoping one of them would magically make me a better developer. It didn't. The biggest lesson I learned wasn't about choosing the best AI tool. It was learning how to work with AI. These days, I don't start by asking AI to write code. I start by explaining the problem. I describe the feature, the business requirements, the edge cases, and what I want the final result to look like. Sometimes I ask ChatGPT to help me plan the implementation first. Once everything is clear, I pass that plan to an agentic coding assistant and start building. That one change made a huge difference. I spend less time writing boilerplate and more time thinking about architecture, user experience, and solving the actual problem. AI still gets things wrong, so I review everything before it goes into production. But instead of writing every single line myself, I'm guiding the process. Looking back, the first few months were the hardest. Now it just feels normal. The tools will keep changing, but I think the real skill is learning how to communicate with AI and use it as part of your development process. That's something worth investing in.

OpenAI’s Bonnie Xu discusses Kepler, an internal AI data analyst agent built to query 600+ petabytes of data. She explains how they overcome context window limits using MCP, automated code crawling, and RAG. Xu also shares how their team leverages scoped semantic memory for self-learning and utilizes AST-based LLM grading to build a robust, regression-free evaluation pipeline. By Bonnie Xu

CircleCI has launched Chunk Sidecars, a new capability designed to bring CI-style validation directly into an AI coding agent's inner development loop By Craig Risi

In a new Windows Developer Blog post titled "Windows platform security for AI agents", Microsoft positions Windows as the trustworthy operating system for autonomous agents and introduces the Microsoft Execution Containers (MXC) SDK as the core of that strategy. The post argues that containment, identity and manageability must be built into the operating system. By Matt Saunders

I’ve been a software engineer for more than 12 years. And like many developers, I’ve been watching AI improve at an incredible speed. Every new model seems smarter than the one before it. Tasks that used to take hours can now be done in minutes. Problems that required deep research can often be solved with a simple prompt. A few years ago, we used to say: Think of AI as a junior developer. That made sense at the time. But today, I don’t think that’s true anymore. AI still makes mistakes. Sometimes very obvious ones. But it also comes up with solutions that surprise me. Sometimes it finds an approach I wouldn’t have thought of immediately. Sometimes it helps me solve a problem much faster than I could on my own. And honestly, that’s both exciting and a little scary. But the biggest thing AI changed wasn’t how I write software. It changed how I think about my work. For most of my career, I thought I loved writing code. I spent years doing it. At work, on side projects, and whenever I had free time. Then AI became part of my daily workflow. In the last month, I’ve built more projects than I normally would in an entire year. Ideas that had been sitting in my notes for years suddenly became possible. And that’s when I realized something important: I don’t actually love writing code. I love building things. I love taking an idea and turning it into something real. I love creating products, solving problems, and seeing something that only existed in my head become something people can use. Code was simply the tool I used to do that. And now AI is another tool. That’s why I don’t hate it. In many ways, AI has helped me build more than ever before. It helped me revisit old ideas that I never had time to work on. It helped me experiment faster. It even encouraged me to explore areas outside software development, like animation and content creation. And this isn’t just happening to programmers. AI is changing design. It’s changing writing. It’s changing marketing. It’s changin

When you talk to ChatGPT about a subject you understand well, you quickly notice something. The first answer is rarely the final answer. You add context. You correct an assumption. You explain what has already been tried. You point out that one proposed solution conflicts with another part of the system. After a few iterations, the answer becomes useful. The same thing happens when AI writes code for real products. The difference is that a slightly incorrect explanation in a chat is usually harmless. Slightly incorrect code can become part of your product, pass a superficial review, and remain there for years. This is why successful AI adoption in software engineering is not primarily about generating more code. It is about context engineering : giving AI enough context, constraints, and feedback to generate code that belongs in your system. The First Answer Is Usually Not Enough AI coding tools are very good at producing plausible solutions. That word matters: plausible. The code may compile. The tests may pass. The implementation may even look clean when reviewed in isolation. But software does not exist in isolation. A change must fit the broader system architecture : the current architecture existing domain rules security requirements operational constraints established conventions previous technical decisions future product direction An AI assistant does not automatically understand those things. It knows the code it can see and the engineering context you provide. Everything outside that window must be inferred. And inference is where divergence begins. If you trust the first response without validating its assumptions, you are usually not accelerating engineering. You are accelerating uncertainty. Lack of Context Creates Duplication One of the first visible effects is duplication. AI does not necessarily know that your application already has: a validation helper for the same domain rule an established authorization pattern a shared API client a retry mechani

Vinay Chella and Akshat Goel discuss the challenges of running traditional CDC across heterogeneous databases during peak order traffic. They explain how Debezium hit limits under high load and share how they built Write-Ahead Intent Log (WAIL) - a custom architecture that utilizes a dumb producer proxy and a smart consumer pattern to cleanly separate the intent from the state payload. By Vinay Chella, Akshat Goel

Microsoft recently introduced at Build 2026 Microsoft Scout, an always-on agent. Scout belongs to a new category of agents Microsoft called Autopilots: always-on agents that work autonomously on a user’s behalf with their own identity, without needing to be prompted each time. Microsoft Scout integrates with Work IQ and is based on the open-source agent framework OpenClaw. By Bruno Couriol

My motorcycle has a Bluetooth instrument cluster. It pairs with the manufacturer's phone app and shows turn-by-turn navigation right on the dash, which sounds great until you actually use it. The nav is routed through a maps provider I don't love, the app is clunky, and there's no way to extend any of it. I kept thinking: it's just my bike talking to my phone over Bluetooth. How locked down can it really be? So one weekend I decided to find out, and a few weeks later I had Google Maps navigation running on the cluster through an app I wrote myself. Here's how that went. There are no docs Of course there aren't. It's a proprietary protocol, and the only reference that exists is the manufacturer's own app, in compiled form. So step one was just watching. I started with a GATT walk on the live bike, which is the Bluetooth equivalent of knocking on every door to see what's there. The cluster exposes one vendor service with two characteristics: one the phone writes to, one the bike sends notifications back on. That's the entire conversation surface. Then I captured the actual bytes going across. Android can log every Bluetooth packet through its HCI snoop log, so I paired the phone with the bike, rode around, and pulled the capture. Now I had real traffic, and absolutely no idea what any of it meant. Reading the app to read the protocol You can stare at hex forever and still guess wrong. The faster path was the app itself. I pulled the APK, ran it through JADX to decompile it, and got something close to readable source. Most of the class names weren't even obfuscated, which was a gift. From there it was cross-referencing: take a message I saw on the wire, find the code that builds it, and work out what each byte is. Frida helped a lot here. It lets you hook a running app and watch functions get called with their real arguments, so I could catch the exact moment the app turned "next turn is a left in 200m" into bytes and shipped them to the bike. Slowly the shape came out

Abstract In modern software engineering, writing code that simply "works" is only the first step. The real challenge lies in designing systems that are maintainable, scalable, and easy to test. This article explores the Dependency Inversion Principle (DIP), the final pillar of the SOLID design principles. Through a practical, real-world example in Kotlin, we will demonstrate how to transition from a tightly coupled architecture to an abstraction-based design. This shift dramatically improves our codebase, facilitates unit testing, and prepares our applications for future growth. Introduction: The Chaos of Coupling As applications grow, it is common to see how a minor change in a database schema or a third-party API triggers a domino effect, breaking unrelated parts of the system. This fragility is a direct consequence of tight coupling. Software design principles, particularly SOLID, were established to prevent this architectural decay. Today, we focus on the "D" in SOLID: the Dependency Inversion Principle (DIP). This principle establishes two core rules: High-level modules should not depend on low-level modules. Both should depend on abstractions (interfaces). Abstractions should not depend on details. Details (concrete implementations) should depend on abstractions. The Scenario: An E-commerce Payment Processor Imagine you are building the billing system for an online store. To process purchases, the system needs to connect to a payment gateway, such as PayPal. The Bad Way: Tight Coupling (Violating DIP) In this initial design, our high-level business logic (OrderProcessor) directly instantiates and depends on the concrete low-level class (PayPalService). // Low-level component (Concrete detail) class PayPalService { fun executePayment(amount: Double) { println("Processing payment of $$amount via PayPal API.") } } // High-level component (Business logic) class OrderProcessor { // Tight coupling: this class depends directly on a concrete implementation private val

How GPU scheduling complexity and MLOps integration are forcing platform teams to rearchitect Kubernetes clusters before operational debt becomes insurmountable. As AI workloads consume roughly 40% of enterprise Kubernetes clusters by 2026, the platform's default scheduler is proving fundamentally mismatched with the topology-aware, gang-scheduled demands of GPU-intensive training and inference. Platform engineering teams that invest now in purpose-built GPU scheduling layers, multi-tenant partitioning, and FinOps-driven autoscaling will separate themselves from organizations drowning in 30-45% GPU utilization rates and mounting infrastructure costs. Why the Default Kubernetes Scheduler Fails GPU Workloads Kubernetes was designed for stateless, CPU-bound services, and its pod-by-pod bin-packing scheduler has no native awareness of GPU topology, NUMA boundaries, or NVLink interconnect bandwidth. This becomes a critical failure point with NVIDIA H100 SXM5 nodes, where achieving full-bandwidth tensor parallelism requires all 8 GPUs on a node to be scheduled as a single atomic unit. The default scheduler cannot guarantee this co-placement, meaning distributed PyTorch FSDP or MPI training jobs frequently land on suboptimal node configurations, wasting expensive NVLink bandwidth and forcing teams to over-provision GPU capacity. Idle GPU memory stranded across partially-utilized nodes is the primary driver behind the 30-45% utilization rates reported in 2025 surveys by Gradient Dissent and Weights and Biases, representing millions of dollars in annual wasted spend for mid-to-large enterprises running mixed AI workloads. Building the GPU Scheduling Stack: Volcano, KAI Scheduler, and MIG Platform teams are converging on a layered scheduling architecture that replaces or augments the default Kubernetes scheduler with GPU-aware primitives. Volcano has become the dominant choice for distributed training workloads, using its PodGroup abstraction to enforce gang scheduling across

A common way to design an application is to begin with its nouns: User Product Order Payment Then each noun receives the standard architectural starter pack: UserController UserService UserRepository The controller receives users, the service services them, and the repository stores them somewhere responsible. This is noun-oriented architecture : treating every important thing in the domain as if it were automatically a useful software boundary. It works for simple CRUD systems. Unfortunately, most applications eventually do something. The noun becomes a drawer Consider a typical UserService : register() findByEmail() resetPassword() changeAddress() disableAccount() mergeAccounts() assignRole() calculateDiscount() These operations all involve a user. That is approximately where their similarity ends. They have different rules, dependencies, side effects, security concerns, owners, and reasons to change. They live together because User was the nearest available noun when the folders were created. As more behaviour accumulates, UserService becomes the official location for anything vaguely user-shaped. Other components depend on it. It gradually depends on authentication, email, permissions, billing, auditing, and several services added during incidents nobody wishes to revisit. The noun becomes both a dependency of everything and a consumer of everything. The folder remains impressively tidy. Name the capability, not the material A better starting question is not: What things exist in this system? It is: What must this system be capable of doing? That leads to components such as: UserRegistrar PasswordResetter AccountMerger OrderPlacer PaymentRefunder SubscriptionCanceller These are agentive names . They name the component responsible for performing a capability. Compare: UserService with: PasswordResetter UserService tells us which noun is nearby. PasswordResetter tells us what the component is for. That difference produces better architectural questions: What rules

Aditya Kumarakrishnan explains how to move past the "amnesia phase" of AI. He shares a blueprint for engineering leaders to build modular agent frameworks using CoALA, leverage decades of process science for scalable workflows, and "terraform" legacy environments into robust, event-sourced artifacts capable of handling unpredictable, cross-functional agent demands. By Aditya Kumarakrishnan

GitHub has introduced the GitHub Copilot app, a desktop control centre for agent-native development that aims to keep engineers in charge while AI agents handle more coding work. Mario Rodriguez writes on the GitHub blog that the recent wave of coding agents has brought faster delivery but also "disjointed workflows, more context switching, and too much time spent reviewing agent-generated code". By Matt Saunders