AI News — Sunday, March 15, 2026

9
Sourcetable – AI Spreadsheet and Data Platform

Sourcetable introduces an AI-powered spreadsheet and data platform designed to simplify data management and analysis.

Hacker Newsproduct
9
Innate (YC F24) – Home robots as easy to program as AI agents

Innate aims to make home robots more accessible by allowing users to program them as easily as AI agents.

Hacker Newsproduct
9
IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse

This paper proposes IndexCache, a novel method to significantly accelerate sparse attention mechanisms in LLMs by reusing indices across layers.

Hugging Faceresearch
8
Dropbase AI – A Prompt-Based Python Web App Builder

Dropbase AI provides a prompt-based system for building Python web applications, streamlining development for users.

Hacker Newsopen-source
8
Vellum (YC W23) – Dev Platform for LLM Apps

Vellum offers a development platform specifically designed to help engineers build and deploy applications powered by large language models.

Hacker Newsproduct
8
Inkeep (YC W23) – Copilot for Support (think Cursor for help desks)

Inkeep introduces an AI copilot for support teams, aiming to enhance help desk efficiency and user assistance.

Hacker Newsproduct
8
Spatial-TTT: Streaming Visual-based Spatial Intelligence with Test-Time Training

This research introduces a method for real-time visual-based spatial intelligence that continuously adapts through test-time training.

Hugging Faceresearch
8
Strategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document Collections

This research investigates the reasoning strategies employed by both AI agents and humans when navigating and searching through document collections.

Hugging Faceresearch
8
Video-Based Reward Modeling for Computer-Use Agents

Researchers present a method for training computer-use agents using video-based reward modeling, improving their ability to learn from demonstrations.

Hugging Faceresearch
8
ShotVerse: Advancing Cinematic Camera Control for Text-Driven Multi-Shot Video Creation

ShotVerse introduces a framework for generating multi-shot videos from text prompts with advanced cinematic camera control.

Hugging Faceresearch
8
DreamVideo-Omni: Omni-Motion Controlled Multi-Subject Video Customization with Latent Identity Reinforcement Learning

DreamVideo-Omni enables highly customizable multi-subject video generation with precise motion control and identity preservation using latent identity reinforcement learning.

Hugging Faceresearch
7
Dedalus Labs (YC S25) – Vercel for Agents

Dedalus Labs aims to provide a Vercel-like platform for AI agents, simplifying their deployment and management.

Hacker Newsproduct
7
Hyperbrowser MCP Server – Connect AI agents to the web through browsers

Hyperbrowser MCP Server enables AI agents to interact with the web by connecting them through browser environments.

Hacker Newsopen-source
7
Trust Your Critic: Robust Reward Modeling and Reinforcement Learning for Faithful Image Editing and Generation

This paper proposes a robust reward modeling and reinforcement learning approach to achieve more faithful and high-quality image editing and generation.

Hugging Faceresearch
7
XSkill: Continual Learning from Experience and Skills in Multimodal Agents

XSkill presents a framework for multimodal agents to achieve continual learning by integrating new experiences and skills over time.

Hugging Faceresearch