Hi guys! Big day yesterday, with OpenAI launching all kinds of things at DevDay. I’ve grouped the full list below so you can see what’s new. These are announcements, with a mixture of releases, previews and things still coming. Access depends on your plan, location and workspace settings. 1. Assistants, models and privacy • Dots: an always-on personal assistant with a cloud computer. Rolling out to eligible Pro and Business Premium users; the Enterprise, Edu and Healthcare beta needs an admin to enable it. • GPT-6.1 Sol: stronger coding, computer use and professional work in ChatGPT Work, Codex and the API. Not ordinary Chat yet. Standard API token prices are one-fifth of Astra's, which is separate from subscription pricing. • Ultrafast: up to eight times faster Astra token generation in Codex, using more allowance. Pro 500 and eligible Enterprise access; Sol's version is coming soon. • Private Intelligence: stronger enterprise data protections, with Private Inference previewing this autumn. 2. Building and maintaining software • Codex in the cloud: tasks continue while your laptop sleeps, using reusable development environments. • Refreshed Codex CLI: voice, clearer multi-agent management and improved session workflows in the terminal. • Code Review: a unified review experience plus automatic cloud reviews. GitLab support is in preview. • Codex Security Cloud: repository scans, ongoing commit checks and proposed vulnerability fixes for review. • Decisions API: Luna handles bounded classification and routing decisions. Limited preview. • Agents API with computer use: developers can build agents that operate software through a hosted browser. • AWS Bedrock Managed Agents: OpenAI agent capabilities running within AWS. Preview. 3. Plugins and connected workflows • Plugin extensions: interactive panels, sidebar tools and file viewers inside ChatGPT. • Improved plugin creation, submission and discovery: easier building, clearer submission feedback and better discovery. • Sites can host plugins: shared apps use each person's connected data and permissions.

ChatGPT has had two useful changes this week, especially if you want it to do more than answer questions. OpenAI has added GPT-6 Sol and GPT-6 Luna to ChatGPT Work and Codex. Sol is intended for more demanding professional work, while Luna brings lower-cost access for work you can supervise and check. Astra remains the model for the most important jobs. At the same time, Voice in Work now works on web and mobile for eligible accounts. That means you can talk through a task while you are away from your desk, then have ChatGPT create a document, presentation or spreadsheet, use a connected app or work in a browser. If you stop the voice conversation before the task is done, it can keep going in text. The sensible first test is not "run my business from my phone." Try one contained job. After a client call, say: "Turn these notes into a follow-up email, a list of agreed actions and a one-page brief. Mark anything you are not certain about. Do not send anything." You still review any action that needs approval on screen, but voice makes it easier to get the job moving before you forget the details. Which job would you rather explain out loud than type from scratch? Sources: https://openai.com/index/introducing-gpt-6-sol-and-luna/ and https://help.openai.com/en/articles/6825453-chatgpt-release-notes

An FAQ should not be a dusty page that every business promises to update when things calm down. The useful version comes from the confusion customers are already showing you in emails, sales calls, support tickets and reviews. Give ChatGPT Work a small pile of real material: recent customer questions, a few sales-call notes, support conversations and your current FAQ or service page. Then ask it: "Create an evidence-backed FAQ brief. Group repeated customer questions, show which source each question came from, compare them with our existing answers, and flag any answer that is vague, missing or makes a claim we cannot prove. Draft a short proposed answer for each gap, but keep uncertain points clearly marked for review. Finish with the five questions most likely to help a buyer make a decision." That gives you more than a fresh batch of copy. You get a reviewable list of what customers are actually trying to understand, where your current explanation is falling short, and which answer deserves attention first. Do not publish every draft answer blindly. Check the sources, tighten anything that sounds too confident, then update the page in small batches. A useful FAQ is really a record of the questions your business keeps being asked. Which customer question do you answer so often that it deserves a proper home on your site?

Okay… I had a little holy s#!t moment talking to my friend Abdul the other day. 😂 Me: “Dude… I think Prompt Engineering may be going the way of the DOS prompt.” He looked at me like I was high on caffeine!! 😂 Me: "hear me out…Remember DOS with the C Prompt? || C:\>_ Back then, you actually had to KNOW the commands to make your computer do what you wanted. Then Microsoft Windows came along. Point. Click. Drag. Drop. Windows didn't kill DOS, it just put a much easier and prettier layer on top of DOS soOoo users can do their work effortlessly. You didn't have to think about what was happening underneath. And now I'm wondering… "Are AI Agents becoming the Windows layer for prompting?" Today: You → Clever Prompt → AI → Answer But imagine Tomorrow: You → Instruct Your AI Agent → Agent gets work done You don't necessarily have to figure out the perfect prompt anymore. You just say: > “Hey, build me a marketing campaign for my 3-Day Vibe Code Workshop.” And your Agent figures out the prompts, breaks down the task, uses the tools, does the work and comes back when it needs you.🤯 That's a pretty big shift!! So maybe I'm wrong and Prompt Engineering isn't actually dying. Maybe it's becoming the DOS prompt underneath the new AI Agent layer?? Just like Windows hid the complexity of DOS… AI Agents may hide the complexity of prompting. And if that's where we're heading, maybe the new skill isn't: “How do I write the perfect prompt?” Maybe the new skill is gonna be: “How do I direct an AI Agent to build out the full workflow?" Hmmmm… 🤔 Am I onto something here, or am I high on caffeine??? 😂👇

Alright, we've got a new ChatGPT model to play with. It's called GPT-6 Astra. I know, another model name to remember! But this one does look like a pretty big step forward. The interesting bit isn't all the benchmark scores. It's that Astra is meant to be much better at actually getting on with a job, rather than just telling you how to do it. OpenAI says it can handle multi-step tasks across browsers and everyday business tools. That includes online research, filling in forms, updating records, organising a calendar, and creating documents, spreadsheets, presentations and websites. What it can do will depend on which ChatGPT tools you have access to, of course. It should also be better at knowing when to make a sensible assumption and when it genuinely needs to ask you something. Hopefully that means less babysitting, without it charging off like an unsupervised intern with admin access. One important thing to know is that ordinary ChatGPT customers can't use it just yet. Right now, access is limited to approved organisations that apply for it. OpenAI says Astra should start rolling out to ChatGPT Plus, Pro, Business and Enterprise users over the next few days. So don't waste half an hour hunting through your settings or repeatedly refreshing the model picker. It isn't hiding from you. We just need to wait for the wider rollout. When it arrives, let's give it something more interesting than a trivia question. Give it a real job, the information it needs, and a clear idea of what "done" looks like. Then come back and tell us how it got on. What are you going to try first? Official announcement: https://openai.com/index/gpt-6-astra/