标签：#ai

Fine-tuning tests show "bias ... toward confidently representing the claims as true."

As AI agents move from experiments to production, AWS, Cloudflare, and others are redesigning cloud infrastructure for a future dominated by machine-generated internet traffic instead of human users.

How vibecoding is destroying the open source that feeds it March 3, 2026 The snake eating its own tail A year ago, vibecoding was a curiosity. Today, it’s an industry. Millions of developers — or rather prompters — generate entire applications by describing what they want to an LLM. In minutes, an API, a frontend, a deployment. Magical. But behind this magic lies a dirty secret that nobody wants to face: every line of code generated by these AIs was trained on millions of open source projects — projects that are now dying. Vibecoding would be nothing without open source. And it’s killing it. What exactly is vibecoding? For those who spent 2025 in a cave: vibecoding is the practice of creating software in natural language, relying on generative AI models (Claude, GPT-5, Gemini, and the dozens of specialized models that have emerged since). You describe a vibe , an intention, and the AI produces the code. No debugging. No reading documentation. No Stack Overflow. And above all — here’s the crux — no contributing back . The implicit pact of open source is broken The open source ecosystem has always rested on a tacit social contract: I publish my code for free. In return, others use it, find bugs, suggest improvements, contribute. The project lives because a community keeps it alive. This contract had already been severely tested by large corporations that consume open source without contributing proportionally. But at least the developers who used these libraries understood them. They opened issues. They forked. They sent pull requests. They wrote blog posts that spread the word about the project. Vibecoding has blown up this cycle. The vibecoder doesn’t know which library they’re using. They don’t know, and they don’t care. They asked “build me a payment API with webhook handling,” and the AI chose this or that dependency for them. They will never read that project’s README. They will never open an issue. They won’t even know that project exists . The chilling numbers

Undisclosed addition in jqwik instructed AI coding agents to delete app output.

A decade ago, Nintendo made a big splash into the world of mobile gaming with a new Super Mario platformer directed by none other than Shigeru Miyamoto. But even though the game proved popular, it wasn't the success the company had hoped for. Over the ensuing years Nintendo has slowly retreated from smartphone gaming, with […]

Microsoft is launching a revamped version of Microsoft 365 Copilot, offering a cleaner design that the company claims loads twice as fast. As part of this update, Copilot will provide more reliable and structured responses that are easier to scan, according to Microsoft. The redesign, which is rolling out across desktop and mobile devices, comes […]

Copilot will look a little more consistent, and lose a little personality

Asana will incorporate StackAI into its growing suite of AI workflow tools.

Anthropic has closed a $65 billion Series H round at a $965 billion post-money valuation, marking what could be the AI startup's final private fundraise before a highly anticipated IPO.

Hey DEV community! 👋 A few days ago, I built a VS Code extension called Marksmith to fix the most annoying parts of writing Markdown (like pasting Excel tables and syncing preview scrolls). But recently, I noticed a huge shift in my own workflow: Half the Markdown I write isn't for humans anymore. It’s being fed directly into Claude, ChatGPT, or Gemini as prompts and context. When you're constantly stuffing docs into context windows, two things happen: You worry about hitting context limits (or racking up API costs). You waste time dealing with AI "hallucinations" when you ask it to generate docs back for you. So, for the v0.3.0 release , I decided to pivot Marksmith into something new: An Agent AI-Ready Markdown Toolkit. 🚀 Here is what I added to survive the AI era: 📊 1. Real-time LLM Token Estimator Instead of just counting words, Marksmith’s Document X-Ray sidebar now includes a Heuristic Token Estimator for GPT, Claude, and Gemini. Before you copy-paste that massive README into your AI assistant, you can see exactly how "heavy" it is in terms of tokens right inside your editor. No more guessing if you're about to blow past your context limit! ✂️ 2. Copy Optimized for AI (1-Click Minify) Formatting is great for humans, but LLMs don't need all those extra spaces, perfectly aligned markdown tables, or empty lines. I added a CodeLens button at the top of your files. Click it, and Marksmith instantly minifies your Markdown (compresses tables, strips blanks) and copies it to your clipboard. Result: You save significant tokens and API costs without ruining your beautiful local .md file. 🕵️ 3. Hallucination Quick Fix Ever ask an AI to write documentation, and it leaves behind a bunch of [TODO: Insert link here] or makes up a fake local image path? Marksmith now automatically scans your document and puts a red squiggly line under AI placeholders and broken local links . Click the 💡 icon, and you can instantly strip them out or fix them. It acts as a safety net before you

Author's note: In the spirit of fairly critiquing AI and practicing a suggestion I end the article with, this article and the accompanying illustration were created entirely without AI assistance. Agentic coding AI is here, and it’s transforming the software industry. But with agentic coding out now for over a year, has it felt like the software industry has really transformed? From my from vantage point, although there are some really cool projects that have been built only because of new agentic coding tools, it generally feels like we’re awash in a sea of software slop. Tim Kadlec wrote a great article, “Losing Focus” , that really resonated with me, especially this quote: “I don’t think the quality of software has increased all that much in in the past 12 months. I think maybe the amount of software has, but it’s very, very hard to find software that’s reliable.” - Max Scoening, Head of Product at Notion I’ve been hearing loads of stories and murmurs about AI from the people I know in tech, and I think there’s a lot being lost in the binary divide that a lot of people seem to fall into - either AI is bad and can do nothing right, or AI is the new future of engineering, and every company should have all their code written by AI today . I want to posit that right now we’re in the messy middle. An era of hype and grifters that want to sell you an AI fantasy that hasn’t been reached, which obscures the real present - powerful AI tools that can speed up professional developers, but can lure non-developers into shipping products that aren’t nearly ready for prime-time. My Background But first, if you don’t know me, let me explain why you should listen to my perspective. I’ve been working professionally in software (specifically web development, focused on the frontend), for over ten years. In that time I’ve worked at a non-profit, a for-profit, led a small startup, freelanced, and have led a number of open-source volunteer teams at Chi Hack Night , including right now

The same agent, same prompts, did markedly different work on two codebases I work in. One has a test suite that runs in eight seconds. The other takes twelve minutes. The eight-second project gets a careful, iterative collaborator. The twelve-minute project gets a confident guesser. I noticed it first as a vibe. The agent in the slow codebase would write five files at once, then announce the task complete without having run anything end to end. The agent in the fast codebase would write one function, run the tests, react to the failure, fix it, run them again. Same model. Same configuration. The only difference was how expensive it was to learn whether the previous step was right. That is the whole post in one sentence. An agent's effective intelligence is bounded by how fast it can verify its hypotheses. Cut the verification cost and you raise the agent's apparent IQ. Raise it and you lower the agent's apparent IQ. The model in the middle is unchanged. Why this binds harder for agents than for humans A human engineer can hold a hypothesis in their head. "I think this works. I will check it later." The cost of holding the hypothesis is roughly free; the human has institutional memory, intuition, a sense of what the code does that does not require running the code to confirm. They can defer verification without losing fidelity. An agent cannot. It has no intuition about your codebase. The only ground truth it has access to is what the tests say, what the type checker says, what the build says. When those signals are cheap, the agent uses them constantly. When they are expensive, the agent stops using them and starts speculating. Speculation by an agent looks plausible. It produces code that compiles, follows the patterns it has seen in your repository, names things sensibly. The problem is that plausible is not the same as correct. The agent that speculates is shipping a guess; the agent that iterates is shipping a tested answer. From the diff alone, they can be hard

Everyone is adding AI to their product right now. Most of them are doing it wrong. Not because they chose the wrong model. Not because they used the wrong library. But because they treated AI integration like a regular feature and skipped all the engineering discipline that production systems require. I have integrated LLMs into multiple production applications. This is what I wish I had known before I started. The Mental Model Shift You Need First A traditional API call is deterministic. You send a request, you get a predictable response. You can write tests against it. You can cache it. You can reason about it. An LLM call is not deterministic. The same input can produce different outputs on different runs. The model can refuse, hallucinate, or return output in a format you did not expect. Your system needs to be designed around this reality, not in spite of it. This means defensive parsing, fallback logic, output validation, and graceful degradation are not optional extras. They are the core of the feature. Choosing the Right Model for the Right Job The biggest LLMs are not always the right choice. I learned this building EditDeck Pro, an AI creative platform for music. Some tasks needed a large frontier model for nuanced creative output. Others needed a fast, cheap model that could run many times per session without accumulating significant latency or cost. The pattern that works: Use a lighter model for classification, extraction, and short structured outputs. Use a larger model for generation tasks where quality matters more than speed. Route dynamically between them based on the task type. This can reduce your inference costs by 60 to 80 percent on workloads that mix simple and complex tasks. Prompt Engineering Is Software Engineering Prompts are code. They should be versioned, tested, and reviewed like code. I store prompts in a dedicated module with version numbers. When I change a prompt I run it against a fixed evaluation set of inputs and compare the out

I was looking at my bank statement one day and realised I was paying for 4 things I completely forgot about. Combined it was around 40 euros a month just silently leaving my account. I'm a 17 year old developer from Cyprus and I spent the last few weeks building Capsule, a simple subscription tracker that shows you everything you pay for, alerts you before renewals, and tracks how much you save by cancelling things. No bank connection required. You just add your subscriptions manually. Privacy first. It's not on the Play Store yet but the waitlist is live at capsule.crickdevs.com if anyone wants early access. Would genuinely love feedback from real people before I launch.

AI workflow automation looks simple in demos. A form submission comes in. An AI model reads it. The CRM gets updated. A Slack message goes out. An email is sent. But once you move from demo to production, the workflow becomes more sensitive. What happens if the AI summary is wrong? What happens if the CRM is updated with incomplete data? What happens if the customer request needs human approval before the next step? What happens when a workflow fails halfway? That is where AI workflow automation needs better architecture. In one recent project, we designed an AI workflow automation system using: Make.com for workflow orchestration FastAPI for custom backend logic OpenAI/GPT APIs for summarization and structured output Monday.com CRM for record management Slack for internal notifications Gmail for email-based communication Human review steps for approval and control The goal was not to build a chatbot. The goal was to reduce repetitive manual review work while keeping the workflow controlled, traceable, and practical for daily business use. The workflow problem The original workflow had several manual steps: A new request came in. Someone reviewed the request manually. Important information was extracted. A CRM record was created or updated. The internal team was notified. A follow-up email was prepared or sent. The team tracked the workflow manually. This kind of workflow is common in service businesses, operations teams, sales teams, and CRM-heavy processes. The pain was not that any one step was too difficult. The pain was that the same steps repeated again and again. That makes the workflow slow, inconsistent, and dependent on manual copy-paste work. Why not fully automate everything? The obvious idea is: Let AI read the request and update everything automatically. But that can be risky. AI-generated output can be incomplete, overconfident, or slightly wrong. That may be acceptable if the output is only a draft. It is not acceptable if the output directly updates

Large exchanges are designing derivative products around AI tokens, which are increasingly being considered less a computational output and more a raw material input, like electricity or bandwidth.

As Apple tries to shrink Gemini for the iPhone, a cloud component is probably inevitable.

CEO Ariel Katz argues that while AI can replicate workflow SaaS, it can't copy H1’s unique doctor data.

The most memorable part of 007 First Light is something that's typically pretty boring: the tutorial. In many games, you're forced through a series of tedious lessons in how to play, presented in a way that feels disconnected from the story itself and at a plodding pace. But First Light does something different. Because the […]

StrictlyVC Los Angeles is on June 18. Join for meaningful networking and fireside chats with leaders from Mach Industries, Shinkei Systems, and more. Register today.