The number of vibe coders with no connection to software development has grown 137× in a year — OpenAI has released a major study on how AI agents are changing the way we work. 👀
In short:
✨ People are using Codex for longer and more complex tasks. 70.2% of users have assigned it a task lasting more than an hour at least once, and 25.6% have given it tasks that would take a human more than eight hours;
✨ Codex has become the main AI tool across all OpenAI departments: from engineering to legal, finance, and recruiting. The most active OpenAI employees run more than 60 hours of agent work in parallel in a single day;
✨ Agents are growing fastest among people who don’t work in software development at all. Since August, the number of such users has increased 137× among regular customers and 189× among enterprise users.
@AximoLoot
In short:
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM
Shutterstock is turning into an AI powerhouse for creators
🌐 It already includes image and video generation, editing, conversational search, and prompt enhancement. At the same time, the company emphasizes the phrase “human-guided” and promises that creators will still receive royalties when their work is used.
➡️ And a bonus for those who live in chats: back in April, OpenAI integrated Shutterstock into ChatGPT — licensed content can now be searched directly from a conversation
@AximoLoot
The stock platform, which for years simply connected clients with the work of photographers and illustrators, is being relaunched as a unified AI-powered platform. The idea is simple: go from an idea to a finished project in one place without jumping between dozens of tabs.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
Place the object inside a glass cube
📸 Version: Midjourney
Generated on: TrueVideo.ai.
@AximoLoot
photorealistic image of a 3D SUBJECT from image encased in a luxurious transparent box, viewed from an enhanced side angle to better reveal the 3D shape of the SUBJECT from image. The box should be white, exquisitely designed, featuring crystal-clear glass with refined, sharp edgesGenerated on: TrueVideo.ai.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
❤1
Turn any website into a desktop app
⏺ Supports ChatGPT, YouTube, X, Spotify, Discord, and virtually any other website
⏺ Generated apps are up to 20× smaller than Electron-based alternatives
⏺ Each app is only about 10 MB in size
⏺ Built with Rust + Tauri, so it uses minimal system resources
⏺ Supports Windows, macOS, and Linux
@AximoLoot
A developer from China got tired of heavy Electron-based apps and created Pake — a tool that packages websites into lightweight desktop applications.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
AI agents would bankrupt a company if they were put in charge, a new study suggests👀
🗄 By the end of the simulation, most models had gone bankrupt. Only GPT-5.5 finished above the initial $1 million in one out of three runs, while Claude Opus 4.8 delivered the best result, growing the company to $27 million (take notes 👀).
Now here’s the interesting part: other studies show that 62% of CEOs already use AI to make a significant portion of their business decisions.
@AximoLoot
Researchers created CEO-Bench — a simulation where AI models managed a startup for 500 days, starting with $1 million in cash and zero customers. The models had to set prices, manage marketing, budgets, social media, and sales. Just like in the real world, the market was unforgiving: customers left, competitors applied constant pressure, and unexpected challenges kept appearing.
Now here’s the interesting part: other studies show that 62% of CEOs already use AI to make a significant portion of their business decisions.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
Anthropic reportedly embedded tracking mechanisms into Claude Code to flag users from China.
🗄 Researchers claim they found that Claude Code analyzes a user's time zone, proxy, and connection domain, then silently tags requests using hidden Unicode characters. Anthropic says these measures are intended to prevent export-control violations and large-scale model distillation. However, many users on X have criticized what they describe as “hidden spyware” inside a tool they trust.
🗄 What's more, some Chinese users claim that Anthropic also embeds a tracking pixel in account suspension emails. The pixel is triggered when the email is opened and could potentially provide additional information about the recipient's location.
@AximoLoot
The issue has sparked heated debate on social media, with some users calling for people to stop trusting the company. Others argue that if a different group had been targeted, the public reaction would have been even stronger.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
A project audit prompt from OpenAI’s President
Greg Brockman has shared a prompt for Codex that turns the agent into a full-fledged tester and auditor for your application.
With it, the AI agent will:
— Analyze every feature and describe its expected behavior
— Create a single document with the status of all features
— Test all user workflows and scenarios
— Generate a detailed report of every issue it finds
— Fix logic errors, bugs, and UX problems
— Re-run all tests after applying the fixes
The prompt:
@AximoLoot
Greg Brockman has shared a prompt for Codex that turns the agent into a full-fledged tester and auditor for your application.
With it, the AI agent will:
— Analyze every feature and describe its expected behavior
— Create a single document with the status of all features
— Test all user workflows and scenarios
— Generate a detailed report of every issue it finds
— Fix logic errors, bugs, and UX problems
— Re-run all tests after applying the fixes
The prompt:
/goal go over every single feature in this app create a user story with expected behaviour based on the code keep a single canonical spreadsheet tracking the features status
• when done switch loop to testing every user story and documenting all errors
• when done fix every logistical error or ux error
• test every user behaviour again post fix@AximoLoot
❤1
Photoshop is sweating — SnapOtter is here
✍️ Meet SnapOtter — a suite of image tools that runs entirely on your own machine instead of someone else’s cloud.
✍️ One important note: the current stable release (v1.x) only includes image tools. The promised 157 features are planned for v2.0, which is expected to launch soon.
@AximoLoot
• Removes backgrounds, upscales images, and restores even the lowest-quality photos using local AI.
• Can crop, compress, convert formats, clean up images, add watermarks, find duplicates, and vectorize graphics.
• It also lets you chain multiple operations into a single pipeline — launch everything with one command and go grab a coffee.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
This media is not supported in your browser
VIEW IN TELEGRAM
SmallpptAI — a presentation creation tool.
🔨 Smallppt can generate a strong starting point for a professional presentation in just a few minutes.
@AximoLoot
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
This media is not supported in your browser
VIEW IN TELEGRAM
Google turns NotebookLM into a short-form video generator for your documents
🔄 NotebookLM can now create vertical videos up to 60 seconds long based on your uploaded sources.
Think TikTok or Reels — but instead of random content, you get a concise explanation generated from your own documents.
🔨 For now, the feature is available only to Google AI Ultra and Google AI Pro subscribers on both mobile and web. Support for free users will roll out later.
@AximoLoot
Think TikTok or Reels — but instead of random content, you get a concise explanation generated from your own documents.
Previously, the service could generate text summaries or audio podcasts. Now it adds a third option: short videos. Simply upload an article, book, or notes, and NotebookLM will produce a ready-to-watch explainer.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM
This media is not supported in your browser
VIEW IN TELEGRAM
Comfy MCP — an AI agent that builds ComfyUI workflows for you
If you never mastered ComfyUI because of its "spaghetti" pipelines — that’s no longer a problem. Comfy MCP has been released, meaning an AI agent will now find the right nodes and build any workflow for you.
🔄 It connects directly to Claude, Cursor, and other agents. Under the hood, it features the latest models for images, video, 3D, and audio, plus hundreds of ready-to-use workflows.
You just ask in plain English:
And you get a batch of results. No dealing with nodes, downloads, or GPU hassle if you don't want to.
➡️ Plus, every single generation is 100% reproducible — you can reuse your exact workflow a month later with zero pain.
@AximoLoot
If you never mastered ComfyUI because of its "spaghetti" pipelines — that’s no longer a problem. Comfy MCP has been released, meaning an AI agent will now find the right nodes and build any workflow for you.
You just ask in plain English:
"Take one frame with a sneaker and make 20 commercial variations in 4 formats."
And you get a batch of results. No dealing with nodes, downloads, or GPU hassle if you don't want to.
@AximoLoot
Please open Telegram to view this post
VIEW IN TELEGRAM