Gemini on your desktop

Plus: The agent revolution

April 17, 2026

Welcome back! Alright, we have to talk about Allbirds, and I’m still trying to wrap my head around how wild this is. The struggling shoemaker announced plans to rebrand as "NewBird AI" and pivot from footwear to … compute infrastructure. 

Shares skyrocketed more than 800% at one point, adding over $100 million to a market cap that had been just $21 million. But by Thursday, the stock had already tumbled more than 20%. 

What do you think about this crazy news? Hit reply and let me know.

Google Launches A Gemini AI App For Mac

Via Google Blog

Google is launching a new Gemini app on Mac that lets you interact with the AI assistant without switching windows on your desktop. Use the Option + Space shortcut to pull up a floating chat bubble, ask questions, and share your current window for context.

How it works: Before sharing your window, you'll need to give Gemini permission to access your system's information. From there, the assistant can pull context from what you're looking at to help answer questions. You can also upload files, photos, or documents from Google Drive, generate images, videos, or music, and revisit previous conversations linked to your Google account.

The competition: The app looks similar to Apple's upgraded Spotlight, which can now perform on-device actions and access models like ChatGPT. But Anthropic's Claude and OpenAI's ChatGPT apps go further—both are leaning into agentic AI and can perform tasks on your behalf directly on your computer.

Why it matters: Every major AI company wants its assistant to be "everywhere"—not confined to a browser tab or dedicated app. The race is on to become the default layer between you and your computer.

OpenAI And Cloudflare Launch New Tools For Building AI Agents

Two major players are giving developers new infrastructure for building AI agents. OpenAI updated its Agents SDK with sandboxing and harness capabilities, while Cloudflare announced Project Think—a full stack for building long-running, durable agents.

What OpenAI is doing: The new SDK lets agents operate in controlled environments with access only to specific files and tools. It also allows developers to deploy and test agents on the most advanced models available. The goal: enable "long-horizon" agents that can handle complex, multi-step tasks safely. Available now via the API with standard pricing.

What Cloudflare is doing: Project Think is a more ambitious infrastructure play. It includes durable execution (agents that survive crashes and resume from checkpoints), sub-agents (isolated child agents with their own storage), persistent sessions (tree-structured conversation history with branching and search), and sandboxed code execution via Dynamic Workers. Cloudflare is betting that agents need to be serverless, durable, and structurally safe—not just smart.

  • Key similarities: Both emphasize sandboxing as critical for safe agent deployment. Both are building toward agents that can run autonomously over longer time horizons without constant supervision.

  • Key differences: OpenAI is focused on making its existing SDK compatible with enterprise sandbox providers, while Cloudflare is building a full platform—compute, storage, execution, and coordination—designed specifically for agents that persist, hibernate, and wake on demand.

The bigger picture: The agent era requires more than good models—it needs infrastructure. Both companies are working to give developers the tools to build agents that act and scale.

AI Learning App Gizmo Raises $22 Million As It Hits 13 Million Users

Gizmo, an AI-powered learning platform that transforms students' notes into interactive study materials, has raised $22 million in Series A funding. The company now has more than 13 million users across 120+ countries.

How it works: Gizmo uses game mechanics to drive engagement—leaderboards, streaks, limited daily lives for incorrect answers, and the ability to challenge friends. It's designed for teens and young adults who already spend hours on TikTok and YouTube.

The bigger picture: Gizmo joins a growing wave of AI-powered learning tools—Anki, Quizlet, Knowt, Yuno, and Google's NotebookLM. With US academic performance at historic lows, the stakes are high. The question is whether gamification can sustain engagement long enough to make a difference.

Remove your personal data from Google and ChatGPT

Have you ever searched for your personal information on Google or ChatGPT? You’d be shocked to find out what people can find out about you.

Your name, phone number, and home address are just the beginning. Anyone deeply researching you can find out about your family members and relationships, SSN, health records, financial accounts, and employment history.

Incogni’s Unlimited plan puts you back in control of your online privacy, keeping you safer from harmful scams, identity theft, financial fraud, and other threats impacting your physical safety.

AI-powered concept search

Via ConceptSeek

ConceptSeek finds ideas, quotes, and exact moments across YouTube videos, transcripts, podcasts, and documents.

How you can use it

  • Search across only the sources you choose, not the entire web

  • Find exact timestamps and transcript passages for quotes

  • Trace arguments and gather verifiable evidence

  • Avoid relying on broad AI summaries

Pricing: Free and paid plans

Open-source workflow engine for AI agents

Via Runsight

Runsight is a self-hosted, YAML-first workflow engine that lets you design, version, and run agent pipelines as Git-native files. 

How you can use it

  • Turn ad-hoc agent code into reproducible, reviewable workflows

  • Set hard budget caps and pause/kill controls

  • Run regression tests to evaluate agent outputs systematically

  • Keep your API keys and models on-premise

Pricing: Free

Jobs, announcements, and big ideas

  • Canva introduces a research preview of its AI 2.0 platform, bringing prompt-driven design and editing into a unified workflow.

  • AI startup Factory reaches a $1.5 billion valuation as it positions itself against competitors like OpenAI, Anthropic, and Cursor.

  • Opera unveils a Browser Connector that integrates ChatGPT and Claude directly into browsing through MCP support.

  • MZLA Technologies releases Thunderbolt, an open-source, self-hosted AI client with built-in Haystack integration.

  • Character.ai rolls out c.ai Books, allowing users to interact with and explore classic literature through AI-driven experiences.

  • Google Gemini adds a Personal Intelligence image generation feature powered by Nano Banana 2 and Google Photos data.

  • Perplexity launches Personal Computer for Mac, designed to coordinate local files, apps, and web-based workflows in one interface.

Behind the scenes: You asked, and I’m answering! Follow along as I share the truth about Future Tools (yes, including how much I make).

That’s a wrap! See you next week for more.

—Matt (FutureTools.io)