Skip to content
Newsletter

📊 ChatGPT now builds slides your team can edit together

Plus GPT-6 Pro splitting one job across agents, and HubSpot capping AI spend

Three-minute read. In today’s issue: slides your team can edit together inside ChatGPT, GPT-6 Pro splitting one job across several agents, HubSpot capping AI credit spend by default, a free offline ChatGPT-style app, and a model named Haiku finally writing haiku.


PRESENTATIONS

ChatGPT now builds slides your team can edit together

What changed. OpenAI began rolling out ChatGPT Slides on 16 October to eligible Pro, Business and Enterprise users with Space access. Ask ChatGPT in Work for a presentation from a conversation, your files or an outline, or share an example so it copies the style.

What it means for you. The monthly update, the sales pitch and the proposal can start where your team already asks ChatGPT questions. Colleagues edit the same presentation in real time and leave shared comments.

The first step. Open ChatGPT in Work this week, upload last quarter’s numbers and ask for a five-slide update. Export it to PowerPoint, PDF or Google Slides, and check every figure before it goes out.

The limitation. It is still rolling out, needs Space access on the web or desktop app, and is not on Plus or Free. OpenAI gives no date for when every eligible account has it.

OpenAI’s ChatGPT release notes (help.openai.com)

MULTITASKING

GPT-6 Pro can now split one job across several agents

On 16 October OpenAI updated GPT-6 Pro in ChatGPT’s chat tab with faster answers, better use of plugins and tools, and several agents that work on different parts of one task at once. You follow each agent in the conversation and open its finished part to see what it did.

It is rolling out to Pro subscribers; Business and Enterprise workspaces get it only where workspace settings allow. Hand it one job with separate parts, such as comparing three suppliers, and check each agent’s part before you act on it.

OpenAI’s ChatGPT release notes on GPT-6 Pro (help.openai.com)


Just shipped

HubSpot now caps AI credit spending by default: each account gets a limit up to 50% above its monthly credits, which you can change at any time, plus limits per feature.

You can also move budget from seats to credits mid-contract, and pay for extra AI work only once it is done. Check your account’s cap before your next busy month. (hubspot.com)


The rest of the day

🗣️ ChatGPT Voice is getting GPT-Live-1.1 and new voices for French, German, Hindi, Japanese, Korean and Mexican Spanish. Pick one in Settings > Voice and rehearse a call with an overseas customer in their language. (help.openai.com)

🌐 ChatGPT Plus and Pro no longer cap how many Sites you can create; usage and storage limits still apply. Give each offer or event its own page instead of cramming them into one. (help.openai.com)

🔎 Google opened Deep Review to the public on 14 October: upload a long PDF or plain-text document and a team of AI agents checks it for errors, two free reviews a week. Google says to treat it as support for a human reviewer, not a replacement. (research.google)

📦 Shopify stores in the UK can now offer Royal Mail pickup points at checkout, with each location’s address, hours and price shown on a map. Switch it on under Settings, Shipping and delivery, Pickup points. (changelog.shopify.com)


🛠️ Built with AI

@rileybrown had Claude edit his raw footage into a finished video. Opus 5.5 cut his mistakes and added music and sound effects around one Blender scene. He rates it “about 3% as good as it could be”. Try it on your next product demo clip. (x.com)


🧰 The Open Source Toolshed

What it is. Jan is a free ChatGPT-style assistant that runs fully offline on your laptop.

Who it’s for. An owner who wants AI help with client contracts, quotes or staff records without those files leaving the machine.

How hard it is. Install the app for Mac, Windows or Linux and download a model inside it; no account needed.

Open source. (github.com)


📈 The AI Odometer

Epoch AI step chart: record Cyber ECI scores by release date, January 2025 to October 2026, with open-weight models such as GLM-5.3 trailing closed models such as Claude Opus 5.5 by six to seven months

Source: Epoch AI, “Open-weight models trail closed models in cyber capabilities by 6–7 months”, 16 October 2026, Jaeho Lee and Alexander Barry, CC BY 4.0.

This places the open cyber frontier roughly where the closed frontier stood 6–7 months ago.

What it means for you. Free models you can download and run offline, like Jan above, sit about half a year behind the paid ones on Epoch’s security tests. Use an offline model for private drafting, and keep security checks, such as asking whether an email is a phishing attempt, on a paid tool.


Too early

Google now lets adults in the US and some other markets build an Android app in AI Studio and run it privately on their own phone through Google Play, with no developer account. Sharing those apps with a small group of trusted people, such as your team, is coming “in the coming months”, with no date yet. (androidheadlines.com)


🌸 One for the road

Claude Haiku 5.5 wrote twenty haiku, then drew every frame of the animation itself, in JavaScript.

A model named Haiku, finally doing the one job its name promised. (x.com)

Join the daily email.

Curated AI news for small businesses and their teams. Free, daily. Sent at 7pm SGT. Unsubscribe anytime.