OpenAI Introduces Apps, Agents, and More at DevDay 2025

Hey there 👋

We hope you're excited to discover what's new and trending in AI, ML, and data science this week.

Here is your 5-minute pulse...

But first, a quick message from our partner 👇

Startups who switch to Intercom can save up to $12,000/year

Startups who read beehiiv can receive a 90% discount on Intercom's AI-first customer service platform, plus Fin—the #1 AI agent for customer service—free for a full year.

That's like having a full-time human support agent at no cost.

What’s included?

• 6 Advanced Seats

• Fin Copilot for free

• 300 Fin Resolutions per month

Who’s eligible?

Intercom’s program is for high-growth, high-potential companies that are:

• Up to series A (including A)

• Currently not an Intercom customer

• Up to 15 employees

Apply now

print("News & Trends")

OpenAI Introduces Apps, Agents, and More at DevDay 2025

OpenAI’s DevDay 2025 unveiled major upgrades: ChatGPT now supports in-chat app integrations (Canva, Figma, Spotify, Zillow), an Apps SDK for building monetizable apps, and AgentKit for creating custom AI agents with workflows and connectors. GPT-5-Codex gains Slack integration and SDK access, while Sora 2 and cheaper realtime voice models hit the API. OpenAI is effectively turning ChatGPT into an all-in-one AI platform—part assistant, part browser, part automation hub.

Google Introduces Jules Tools: A Command Line Companion for Google’s Async Coding Agent (3 min. read)

Google introduces Jules Tools, a command-line interface for its asynchronous coding agent, Jules. This lightweight CLI allows developers to manage tasks like writing tests, building features, and fixing bugs directly from the terminal, integrating seamlessly with existing workflows. By bringing Jules into the command line, developers gain enhanced control and visibility, enabling real-time task management without leaving their preferred environment.

IBM Granite 4.0: Hyper-efficient, High Performance Hybrid Models for Enterprise (3 min. read)

IBM's latest release, Granite 4.0, introduces a hybrid Mamba/transformer architecture that slashes memory usage while maintaining robust performance. This innovation enables deployment on more affordable GPUs, significantly reducing operational costs. Open-sourced under Apache 2.0, Granite 4.0 is the first open model family to achieve ISO 42001 certification, underscoring its commitment to security and transparency. Available on platforms like IBM watsonx.ai and Hugging Face, these models are tailored for efficient, enterprise-grade AI applications.

OpenAI Introduces AgentKit (5 min. read)

OpenAI has unveiled AgentKit, a comprehensive suite designed to streamline the development, deployment, and optimization of AI agents. This toolkit introduces Agent Builder, a visual interface for crafting multi-agent workflows; ChatKit, which simplifies embedding customizable chat experiences; and enhanced evaluation tools for performance measurement. By consolidating these resources, AgentKit aims to reduce the complexity traditionally associated with building agents, enabling faster iteration and more reliable deployment.

Google Introduces a New Collaborative Approach to Image Generation (7 min. read)

Google Research unveils PASTA, a reinforcement learning agent designed to iteratively refine text-to-image outputs by learning user preferences through interactive feedback. By combining real human interactions with simulated data, PASTA effectively adapts to individual creative intents, producing images that users find more satisfying. This collaborative approach marks a significant advancement in aligning AI-generated visuals with user expectations.

print("Applications & Insights")

Practical Guide to Semantic Layers (5 min. read)
This article delves into the concept of semantic layers, explaining their role in bridging raw data and business insights. It outlines the benefits of implementing a semantic layer, such as improved data consistency and accessibility, and provides practical steps for integrating one into your data architecture. The guide also discusses common challenges and best practices, making it a valuable resource for data professionals aiming to enhance their analytical capabilities.

Docker for Data Scientists (Part 1): A Gentle Introduction (5 min. read)
This article introduces Docker as a game-changer for data scientists, especially given how the field is evolving. It emphasizes its role in creating consistent, reproducible environments. It walks through setting up Docker, crafting Dockerfiles, and managing containers, all tailored to data science workflows. By the end, you'll grasp how Docker can streamline your projects, ensuring they run seamlessly across different systems.

Which Table Format Do LLMs Understand Best? (Results for 11 Formats) (8 min. read)
In a recent study, researchers evaluated 11 data formats to determine which is most effective for LLMs in processing tabular data. The Markdown-KV format led with 60.7% accuracy but required 2.7 times more tokens than the most efficient format, CSV, which had only 44.3% accuracy. This highlights a trade-off between accuracy and token efficiency, suggesting that format choice significantly impacts LLM performance and cost.

Understanding the 4 Main Approaches to LLM Evaluation (From Scratch) (15 min. read)
Sebastian Raschka delves into four primary methods for evaluating large language models: multiple-choice benchmarks, verifiers, leaderboards, and LLM judges. He provides clear explanations and from-scratch code examples, highlighting the strengths and limitations of each approach. This comprehensive guide equips data scientists and ML engineers with the tools to assess LLM performance effectively.

print("Tools & Resources")

TRENDING MODELS

Text Generation
zai-org/GLM-4.6
⇧ 14.6k Downloads
GLM-4.6 is a 357-billion parameter text generation model designed for advanced natural language understanding and generation tasks.

Image-Text-to-Text
ServiceNow-AI/Apriel-1.5-15b-Thinker
⇧ 7.03k Downloads
Apriel-1.5-15b-Thinker is a 15-billion parameter image-text-to-text model optimized for generating descriptive text from images.

Text-to-Speech
neuphonic/neutts-air
⇧ 5.82k Downloads
neutts-air is a 0.7-billion parameter text-to-speech model that converts written text into natural-sounding speech.

Text-Generation
deepseek-ai/DeepSeek-V3.2-Exp
⇧ 20.8k Downloads
DeepSeek-V3.2-Exp is a 685-billion parameter text generation model aimed at providing high-quality language generation capabilities.

Audio-to-Audio
LiquidAI/LFM2-Audio-1.5B
⇧ 775 Downloads
LFM2-Audio-1.5B is a 1-billion parameter audio-to-audio model designed for advanced audio processing tasks.

TRENDING AI TOOLS

• 🚀 Comet: Accelerate machine learning experiments with real-time tracking and collaboration.

• 🛠️ Tinker: Interactive tool for data exploration and visualization by Thinking Machines.

• 🤖 Claude in Slack: Claude integration into Slack for seamless collaboration and productivity.

• 🔍 Sora 2: Advanced tool for efficient data analysis and visualization.

• 🛠️ Agent Workflows: Customize and automate agent tasks with flexible workflows.

print("Everything else")

• Sam Altman announces upcoming changes to Sora, including enhanced character generation controls for rightsholders and a revenue-sharing model for video generation.

• OpenAI and Jony Ive are reportedly facing challenges in designing their AI device.

• ASAPP presents 100 real use cases for generative AI in enhancing customer service agents.

• DeepMind introduces CodeMender, an AI agent designed to enhance code security.

That’s it for today!

Before you go, we’d love to know what you thought of today's newsletter to help us improve the pulse experience for you.

See you soon,

Andres