The week in startup news: Oct 5-11

1. Open or closed AI: founders' choice

At TechCrunch Disrupt 2026, founders are talking through a decision that used to be simple: pick an AI model and stick with it. Now open models (free for anyone to use and change) keep getting better, closed APIs (ways to plug into a company's AI over the internet) keep advancing too, and some teams are mixing several models in one product instead of betting on one. The panel covers this choice from a few angles, including the chips and plumbing software underneath it all.

2. Gitframes: AI agents that edit video with code

Gitframes is a new npm package (a code library anyone can add to their project) that gives AI coding agents the power of three separate desktop apps at once: Photoshop-style image editing, After Effects-style motion and keyframing, and Blender-style 3D scenes. Instead of installing and paying for a multi-gigabyte creative suite, an agent just writes a bit of code, checks the frames, and renders a finished video file. It is still in beta, so things may change between updates.

3. Minigraf: a tiny database built for AI agents

Minigraf is a single-file embedded graph database (a database that stores how pieces of data connect to each other, built right into your app instead of running on a separate server). It is written in Rust and lets you query relationships and also time travel, meaning you can look back at exactly what the data looked like at any past point. Think of it as SQLite, the small database that ships inside countless apps, but built for connected data with full history.

4. DoorDash launches an AI agent you can text

DoorDash now lets you order food by texting an AI agent through Apple Messages. Say order my usual and it understands your regular order, or describe a craving and it will search nearby restaurants, suggest a cart, and text you photos of the food. It can also handle a group order with different diets and quantities in one go.

5. RobCo crosses a $1 billion valuation

RobCo, a German startup that builds robots to automate factory floor work, has passed a $1 billion valuation (making it a unicorn, startup slang for a company worth that much) in a recent employee share sale. That is double what it was worth in January, when it raised $100 million. It is a fast climb for a company building physical hardware, which usually grows slower than pure software.

6. Meta opens its Muse AI agent to hardware tinkerers

Meta's Muse is a personal AI agent that books travel, fills out forms, and shops for you. The company just launched Muse Gadgets, an open-source project that gives developers free firmware (the low-level software that runs a device) and a Linux software development kit (a toolkit for building on a platform), plus starter ideas like adding an e-ink display or loading Muse onto a stick that plugs into a TV.

7. Someone built Linux containers in 500 lines of code

A developer wrote a minimal version of Linux containers (the technology that lets you run an app in its own isolated box) from scratch, just to understand them properly. The goal was to find the smallest set of restrictions needed to safely run code you do not fully trust, and to figure out which permissions are flat-out unsafe to skip. It is a solid read if your team ships anything in containers and wants to actually understand what is happening under the hood.

8. A free tool deletes 12GB of Apple's AI data

RemoveMacAI is a new open-source command line tool that switches off Apple Intelligence on a Mac, deletes the AI models sitting on your disk, and stops them from being re-downloaded. Apple used to offer one simple switch to turn this off, but on the newest Mac system the option got scattered across different settings menus. The tool's popularity is a small signal that people want a clear, simple off switch for AI features.

9. Five takeaways from Sifted Summit 2026

Sifted Summit brought hundreds of people from Europe's startup world to London for its fifth year. This time, organizers shaped the event around Series A and later conversations, meaning startups that have already raised their first big funding round, so founders at a similar stage could go deeper with their peers.

Why this matters if you're building

Most of this week is about small teams moving fast with limited resources: an agent that edits video without installing paid software, a tiny database you can drop into an app in minutes, an open hardware project with free starter code, and a tool one developer built just to understand containers properly. None of this needs a big budget. It needs curiosity and a willingness to read the source code. That is usually the actual edge a small team has over a bigger one.