ChatGPT Agent Can Now Do Your Tasks🤖

Your Front-Row Seat to the Future of Intelligence

Welcome to this week’s AI roundup—where cutting-edge innovation meets real-world impact. From OpenAI’s new personal assistant-style agent, to Stanford’s fully transparent foundation model, and Anthropic’s financial AI gamechanger, we’re covering the tools and breakthroughs redefining how we live and work. Whether you're a developer, researcher, or business leader, these updates signal a powerful shift in how AI integrates into our daily systems.

Let’s dive into what’s shaping the next wave of productivity, transparency, and intelligent decision-making.👇

OpenAI’s ChatGPT Agent Can Now Work Like a Personal Assistant

The ChatGPT agent by OpenAI is a major leap in AI productivity, combining web interaction, deep research, and conversational intelligence into one unified system. It autonomously handles complex, multi-step tasks using its own virtual computer—navigating websites, running code, analyzing data, and delivering editable presentations and spreadsheets. Users can delegate workflows like planning meals, conducting competitive research, or compiling meeting summaries, with the freedom to intervene or refine tasks anytime. Accessible to Pro, Plus, and Team users through the ‘agent mode’ in ChatGPT's tools menu, it leverages a visual browser, text browser, terminal, and APIs, integrating with platforms like Gmail and GitHub. Emphasis is placed on privacy and user control, with safeguards like browser takeover security, confirmation before impactful actions, and prompt injection defense. The agent outperforms prior models in tasks like spreadsheet editing, data science, and investment banking modeling, and while still evolving—especially in presentation formatting—it offers strong real-world utility supported by advanced privacy and safety frameworks.

Stanford’s Marin: The First Fully Open JAX Foundation Model

Stanford’s Marin foundation model, developed by the Center for Research on Foundation Models (CRFM), sets a new standard for transparency in AI by making not just the model but the entire research process fully open, including code, data, methodologies, experiments, and training logs, all under an Apache 2.0 license. Built using JAX and the custom Levanter framework, Marin’s releases—Marin-8B-Base and Marin-8B-Instruct—tackle and solve core challenges of large-scale, reproducible AI training, such as performance on hardware accelerators, ease of scaling across compute clusters, and perfect reproducibility even across restarts and hardware changes. The "open lab" approach means every experiment, adjustment, and outcome is tracked and discussed in public, allowing the broader community to participate in building, scrutinizing, and improving the models. The Marin project’s unprecedented openness provides a powerful, fully reproducible resource for researchers and may accelerate trustworthy, collaborative AI innovation across the field.

Claude Transforms Financial Analysis with AI

Anthropic’s new Claude for Financial Services is an enterprise AI platform designed to unify market feeds and internal data (from providers like Databricks, Snowflake, S&P Global, and FactSet) into a single, powerful interface for financial professionals. By integrating real-time, verified data from major financial sources and supporting secure, private workflows, Claude enables analysts and institutions to modernize trading systems, automate compliance, perform in-depth financial modeling, conduct due diligence, research, and build investment memos and pitch decks—all while maintaining strict data confidentiality. The platform is tailored to streamline traditional processes, offering features like direct Excel and PowerPoint generation, agent-assisted research, and audit trails, and comes with expert onboarding support. Available through enterprise channels including AWS Marketplace, Claude’s Financial Analysis Solution is aimed at accelerating analysis, improving accuracy, and helping institutions gain and maintain a competitive edge in a rapidly evolving industry.

Hand Picked Video

In this video, we’ll look at Olly’s biggest update yet — Auto Commenter now supports Facebook, Instagram, and X! Discover how you can automatically engage with 60+ comments per hour, grow faster across platforms, and save hours of manual work every single day. If you're looking to scale your engagement effortlessly, this update changes the game.

Top AI Products from this week

• Runway - AI applicant ranking tool that adapts to YOUR hiring priorities. Define custom ranking parameters, upload up to 50 resumes, and get applicants ranked by what matters most to your role and company. No more one-size-fits-all screening and basic ATS filters.

• Snack it - End screenshot clutter: Snack it saves any image, auto-generates AI prompts from moodboards, and sends to ChatGPT, Midjourney, Gemini in seconds. Stop losing inspiration. Free, fun, freakishly useful. Install now! See it → Snack it → Prompt it!

• NotebookLM Featured Notebooks - Bookva is a scheduling service from 2025. If Calendly was as beautiful as WeTransfer, as delightful as Lu.ma, and as intelligent as ChatGPT. If you care about aesthetics and design - use Bookva.ai. If you don't - use Cal.com or Calendly. Made in Amsterdam.

• UTCP - Create and edit photos and illustrations, turn them into videos, and add voice with studio-quality results. Engineered for quality and performance.

• Clevr - Meet Clevr. She explains things with voice and frantically scribbles on a canvas. It's like having a one-on-one with a super smart, slightly over caffeinated artist who just wants you to finally get it.

• Textalyz- Textalyz helps you rewrite, fix grammar, and translate your messages with AI right in Chrome and Slack. Perfect for non-native speakers, remote teams, and anyone who wants to express themselves clearly, professionally, and confidently.

This week in AI

• IBM API Agent Now Live - IBM’s API Agent in API Connect is now generally available, helping teams build, govern, and deploy intelligent, production-ready APIs using natural language.

• Amazon Launches AgentCore Preview - Amazon introduces AgentCore, a secure, scalable platform to deploy and manage AI agents with tools for memory, identity, observability, and web automation.

• Thinking Machines AI Coming Soon - Thinking Machines Lab, led by ex-OpenAI exec Mira Murati, will launch its first AI product with a major open-source component after raising $2B at a $12B valuation.

• Copilot Vision Adds Desktop Share - Windows Insiders can now let Copilot view and analyze their entire desktop for real-time help and insights, rolling out with the latest Copilot update

Paper of The Day

The Generative Energy Arena (GEA) is a new platform from Universidad Politécnica de Madrid designed to study how knowledge of energy consumption affects human preferences when evaluating large language models (LLMs). Unlike traditional automated benchmarks or costly human studies, GEA uses a public arena format where users judge the quality of model responses, then are told which model is more energy efficient and asked if this changes their preference. Early results show that energy awareness strongly influences user choices—on average, 46% of users changed their vote in favor of more energy-efficient, smaller LLMs once they learned about energy usage. This suggests that for most tasks, users see little added value in the extra complexity and energy cost of larger models, making energy information a crucial factor for future LLM evaluation and design.

To read the whole paper, go to the official publisher’s site.