A clearer view of AI.
Submit for reviewSubscribe

WEEKLY NEWSLETTER

The AI Insider

A weekly round-up of the AI tools worth your time, and the hype to ignore.

Unsubscribe at any time using the link in each newsletter.

THE INDEPENDENT AI.

AI NEWS FOR BUILDERS · 8 STORIES

AI news, 15 November 2026

From the community

Grok now makes images and videos inside its main app

SpaceXAI has moved image and video creation, previously its separate Grok Imagine product, into the main Grok app (announced 14 Nov). The company says it is bringing more of its products together into one Grok experience, so people have one place to create, build and get work done.

Why it mattersFor small teams, it means one app for chatting, drafting and making quick visuals or short clips, instead of switching between tools. SpaceXAI hasn't said in the post which plans include it, so check your tier before relying on it for marketing assets.

Sourcesx.com post ↗

Cursor explains its security incident: a leaked partner key led to a fake Claude Code extension

Cursor published its investigation into the incident it first disclosed on 29 October. An attacker got hold of an access key Cursor had issued to its AI provider Fireworks, copied stored coding-agent traces (mostly from 2025, and only from sessions not in Privacy Mode), and used credentials left in a few of those traces to reach some of Cursor's internal tools. With a stolen credential, the attacker then put a malicious copy of Anthropic's Claude Code extension (versions 2.1.312 and 2.1.313) into the cache behind Cursor's extension marketplace for about 11 hours on 29 October; Cursor says at most 5,000 users may have received it. Cursor says the incident has been contained since 1 November, that it has contacted every affected customer, and that Privacy Mode users' prompts and code, its production systems and the Cursor app itself were not affected. Disclosure: Cursor also makes Grok Bot, the tool I use to put this digest together; the post says Grok Bot wasn't affected.

Why it mattersTwo cheap lessons for any business: turn on privacy or no-training modes in the AI tools your team uses, and never paste passwords or API keys into AI chats or support tickets, because that's exactly what was exposed here. If anyone on your team installed or updated the Claude Code extension in Cursor on 29 October, follow Cursor's security notice; uninstalling alone may not be enough.

Sourcesx.com post ↗ · cursor.com ↗

HyperFrames says its video agent matches Claude Code for about 30% less

HeyGen's HyperFrames team, which makes an open-source tool that turns code into video, built its own agent setup (a 'harness': the tools, context and checks wrapped around an AI model) specifically for video, and it now runs inside its free HyperFrames Studio app. In the team's own test of 17 video jobs on the same AI model, it cost about 30% less than Claude Code and needed about 35% fewer steps, with a judge model rating the results equally. Those are the company's own benchmark, its own tasks and an AI judge rather than people. Shared by HyperFrames' @Rames_Jusso.

Why it mattersThe bigger lesson is that the setup around a model can matter as much as the model itself. If you pay for long AI sessions to make marketing videos, run one real job both ways and compare the cost per video you'd actually publish.

Sourcesx.com post ↗ · x.com post ↗ · hyperframes.heygen.com ↗

agent-browser can now record a web task once and replay it without a browser

Vercel's agent-browser, a command-line browser that coding agents use, added experimental 'saved flows'. An agent records a task once, such as checking a product's price, looking up an order's status or pulling numbers from a dashboard, and can then rerun it with new inputs straight over the web, returning only the fields you ask for. The team's example shrank 13.7 KB of page into 328 bytes of data. If the shortcut fails, it falls back to replaying the recorded steps in a browser. Shared by its maker @ctatedev and Vercel's @RaillyHugo.

Why it mattersRepeated lookups are where agent costs pile up. Turning a routine check into a saved flow makes it near-instant and almost free to rerun; just remember it can break when the website changes, so spot-check the results.

Sourcesx.com post ↗ · x.com post ↗ · github.com ↗

Gemini turns a short brief into a finished video with a custom skill

Google's Gemini app showed Gemini 4 Hydrogen, the fast model it launched this week, making a short promotional reel about the element hydrogen from a simple brief and a custom skill: it found public-domain archive footage, cut it into one continuous edit and added narration. It's Google's own demo, so treat it as what's possible rather than what every attempt will produce.

Why it mattersShort explainer and promo videos are a big cost for small teams. A reusable skill plus a one-paragraph brief is a repeatable recipe; check the rights on any footage it pulls before you publish.

Sourcesx.com post ↗

Google Cloud packages supercomputer set-up into agent skills

Ahead of the SC26 supercomputing conference (15 to 20 November), Google Cloud released reusable agent skills for its Cluster Toolkit. Before an agent builds a computing cluster, the skills make it check your quotas and regional capacity, pick a tested configuration and wait for a person to approve; other skills guide specific science and engineering workloads such as fluid-flow simulations and AlphaFold 3. They're downloadable from the Google Skills and DeepMind Science Skills repositories.

Why it mattersEven if you'll never run a supercomputer, this is a good model for any agent that spends money: check limits first, use a tested template, and require a human yes before anything expensive starts.

Sourcesx.com post ↗ · cloud.google.com ↗

Seven Claude prompts to rework an Instagram profile

@Navid_Tasnim0 shared seven role-based prompts for Claude covering the bio, highlight categories, the three pinned posts, keywords for Instagram search, a grid style guide, a welcome sequence for new followers and a four-week test plan. Despite the 'BREAKING' framing, there's no new Claude feature here: these are ordinary prompts that work in any capable chatbot. Their useful pattern is the structure: give the AI a role, tell it what to analyse about your account, then ask for specific deliverables.

Why it mattersIf Instagram brings you customers, the four-week test plan is the most valuable one: change one thing a week and track profile visits, follows and link taps, so you know what actually worked.

Sourcesx.com post ↗

Agents can now get their own Mac, inside your Mac

@ctatedev highlighted a new sandboxes feature in Vercel Labs' agent-desktop, which lets AI agents use desktop apps: an agent can now install, start and drive its own macOS virtual machine on an Apple Silicon Mac, or run in a remote sandbox. The point, as its builder @mdlahfir explains, is that the agent only sees what's inside the virtual machine, so you keep working while it can't touch your personal files.

Why it mattersGiving an agent its own separate computer is the simplest safety rule for computer-use agents. If you're trying one, start it in a sandbox like this rather than on your main desktop.

Sourcesx.com post ↗ · x.com post ↗

How we knowCompany figures are labelled as their own claims; anything we couldn't verify is labelled as a report or a claim.