- Everyday AI
- Posts
- Ep 761: ChatGPT’s SuperApp Preview: Big Codex Updates Explained and the 5 Steps to Get the Most out of It
Ep 761: ChatGPT’s SuperApp Preview: Big Codex Updates Explained and the 5 Steps to Get the Most out of It
OpenAI just launched ChatGPT Images 2.0, Anthropic’s Claude Mythos was briefly accessed in a leak, Google turned Gemini into a full enterprise agent platform, and more.
Everyday AI
April 22, 2026
👉 Subscribe Here | 🗣 Hire Us To Speak | 🤝 Partner with Us | 🤖 Grow with GenAI
Outsmart The Future
Today in Everyday AI
8 minute read
🎙 Daily Podcast Episode: This week, OpenAI previewed its super app through a major Codex update, showing AI can now run tasks on your computer in the background. Give today’s show a watch/read/listen to learn more.
🕵️♂️ Fresh Finds: Google says 75% of its new code is AI-generated, Anthropic removed Claude Code from its Pro signup page, GPT-5.5 around the corner, and more. Read on for Fresh Finds.
🗞 Byte Sized Daily AI News: OpenAI just launched ChatGPT Images 2.0, Anthropic’s Claude Mythos was briefly accessed in a leak, Google turned Gemini into a full enterprise agent platform, and more. Read on for Byte Sized News.
💪 Leverage AI: OpenAI just turned Codex into an agent that can run tasks on your computer, in the background, without interrupting your work. Keep reading for that!
↩️ Don’t miss out: Miss our last newsletter? We covered: Anthropic just signed a $100B deal with Amazon, OpenAI launching GPT-Image-2 in ChatGPT, and Apple’s AI power play with Tim Cook out, and more. Check it here!
Ep 761: ChatGPT’s SuperApp Preview: Big Codex Updates Explained and the 5 Steps to Get the Most out of It
You're waiting for tomorrow's big AI drop? 🔮
It's already hiding in plain sight: OpenAI's recently updated Codex app.
And if you can get over the word code in the name and see past its developer-only roots, you'll notice a juggernaut autonomous AI agent that looks and acts just like ChatGPT.
But better.
(It works alongside you without taking over and disrupting what you're doing. Yes. Really.)
We give you the roadmap and a live demo. What could go wrong?
Also on the pod today:
• Persistent memory learns workflows 🧠
• Desktop agents run in background 🖥️
• Codex’s own independent cursor 🖱️
It’ll be worth your 52 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Here’s our favorite AI finds from across the web:
New AI Tool Spotlight – InstantDB Gives your AI a real backend, Toki 2.0 is Your New Scheduling Agent, Tines Securely scales AI and automation. Integrates agents, teams, and tools with speed and control.
Google Code AI — About 75% of new code at Google is now AI-generated and still reviewed by engineers.
Claude Code Removed — Anthropic briefly removed Claude Code from its $20 Pro signup page, hinting that flat-rate AI plans may need major changes as agent-style features explode usage
Stitch Open Source — Stitch’s DESIGN.md is now open-source, so design rules can be exported and shared across tools.
Google Workspace Updates — Google just turned Workspace into a context engine for enterprise AI, letting Gemini-powered agents pull emails, files, chats, and calendars together to act across Docs, Sheets, Slides, Drive, Gmail, and Chat.
Gemini for Home Memory — Gemini for Home keeps the mic live after one “Hey Google,” so follow-ups feel natural. It now remembers context, works in multiple languages, and cuts down accidental replies.
Google Investments — Google Cloud is investing $750M to help partners build, prototype, and deploy agentic AI with Gemini and embedded engineering support.
GPT-Image-2 Tops Arena Charts — GPT-Image-2 just cleaned up Image Arena, taking #1 in every Text-to-Image category with a record 242-point lead.
AI Tracking — Meta will install software on U.S. employees’ work computers to log clicks, mouse movements and keystrokes to train AI agents.
Youtube Celebrity Deepfakes — Celebrities can now find and remove AI deepfakes of their faces on Youtube. CAA, UTA, WME and Untitled Management back the tool.
Google Cloud Deal — Thinking Machines Lab just signed a single-digit‑billion dollar deal with Google Cloud to run its reinforcement learning workloads on Nvidia GB300 GPUs.
NVIDIA and Google Cloud — NVIDIA and Google Cloud just unleashed Blackwell and Vera Rubin gear to supercharge agentic and physical AI.
Atomic Bot Hermes — Atomic Bot now offers a one-click launcher for Hermes Agent, so you can run a persistent-memory autonomous agent locally without terminal fuss.
GPT-5.5 Leak? — Sleuths spotted that OpenAI’s anticipated GPT-5.5 model was temporarily added to the Codex CLI app.
Meta Program — Meta is launching a free four-week program to train fiber technicians to help staff its expanding data-center buildout.
1. OpenAI launches ChatGPT Images 2.0 🌅
OpenAI yesterday unveiled ChatGPT Images 2.0, a major upgrade that delivers faster, higher-quality image generation and expanded layout options, making it easier to produce horizontal, square, and vertical images directly in ChatGPT.
The update sharpens image fidelity, boosts speed, and adds flexible output modes for creative and practical use, signaling a push to make image creation more integrated into everyday workflows.
2. Unauthorized users accessed Anthropic's Claude Mythos model 🕵️
According to Bloomberg, a small group in a private Discord used a contractor's credentials and leaked data to access Claude Mythos on the day it was announced, then tested it on harmless tasks rather than attacks.
Anthropic, which restricts Mythos to select partners via Project Glasswing because of its potential for misuse, says the breach appears limited to the contractor’s environment and its systems were not compromised. The incident is timely because it underscores how quickly access controls and leaked credentials can undermine guarded rollouts of powerful models.
3. Google folds Vertex AI into Gemini Enterprise as an all-in-one agent platform 🤞
Announced at Google Cloud Next on April 22, 2026, Google expanded Gemini Enterprise into an end-to-end platform that unifies agent development, deployment, and governance for large organizations.
The update folds Vertex AI into the new Gemini Enterprise Agent Platform, adds low-code Agent Studio, long-running runtimes, memory and simulation tools, and access to 200+ models while introducing stronger governance controls like Agent Identity, Model Armor, and an Agent Registry.
4. SpaceX offers $60B buyout or $10B fee for Cursor tie-up 💸
SpaceX confirmed it has struck a cooperative deal with Cursor that gives SpaceX the right to acquire the AI coding platform later this year for $60 billion or instead pay $10 billion for the joint work, a move timed as Musk’s companies prepare for an IPO.
The arrangement signals SpaceX is racing to bolster xAI and coding capabilities against rivals like Anthropic and OpenAI by combining Cursor’s developer-focused product with SpaceX’s massive training compute.
5. Google splits TPU tasks into two specialized chips ✂️
Google today announced that its eighth-generation Tensor Processing Units will come in two specialized chips — one for training and one for inference — available later this year, a timely move as AI workloads shift toward agent-style, low-latency applications.
The company says the new training TPU delivers 2.8 times the performance of the previous generation for the same cost, while the inference-focused TPU 8i uses triple the onboard SRAM to boost latency-sensitive serving by about 80 percent.
6. Google unveils Workspace Intelligence to make AI do the heavy lifting for work apps 📱
Google today announced Workspace Intelligence, a suite of AI features that weaves Gemini reasoning through Gmail, Drive, Docs, Sheets, Slides, and Chat to gather context, prioritize tasks, and automate multi-step work across apps.
The new tools include Ask Gemini in Chat as a command line for work, AI-powered briefs and summaries in Gmail and Drive, Docs and Slides that generate drafts, visuals, and full decks from your workspace context, and Sheets that build complex spreadsheets from plain language.
Anthropic pretty much invented the agentic desktop. Then they got lapped on it.
How?
That’s because OpenAI quietly shipped an update last week to an app called Codex, its unofficial preview of their forthcoming ‘super app’. Most leaders kept scrolling past the news because the name sounds technical. You know…. ‘code’ in the name so not for me.
Biiiig mistake fam.
This sly OpenAI update might signal the biggest shift in how AI actually works for the masses since ChatGPT dropped. We don’t say that lightly.
What’s the big deal with Codex? It’s the first true autonomous agent that works WITH you in real-time and automagically uses your data and apps.
Don’t worry. We’ve got your Codex 101 for the nontechnical leader to get you off and running.
On today's Everyday AI, we showed you with live demos why Codex is a requirement now and the roadmap to the five moves leaders gotta make before everyone else catches on.
You’ve gotta window. Let’s get after it.
1. Being First Might have Cooked Anthropic 🔥
Sometimes being first is actually being stuck.
Anthropic pretty much pioneered the agentic desktop in the mainstream with Claude Code and Cowork, while OpenAI quietly watched and took notes on every complaint. Then OpenAI shipped a unified version while Anthropic's own stuff was still three siloed apps that don't talk to each other.
You know the feeling.
Configure an MCP server in one Claude app, then realize it prolly doesn't show up in the other two.
Codex shipped unified from day one with persistent memory, computer use, 90+ plugins, and an in-app browser under one agentic brain. And, unlike Claude’s desktop version with three silos that don’t talk to each other (Chat/Cowork/Code), Codex is a unified experience.
Whether you wanna code, ‘cowork’ or just chat it just works.
Try This
Audit your AI stack this week and ask one brutal question.
Are your tools sharing context, or are they islands pretending to be a continent?
If your team re-explains the same workflows in three apps daily, you're straight up paying rent on fragmentation.
Pick the tool your team uses most and pilot Codex for one week against it. That reclaimed time IS your AI strategy.
2. Your Coworker Stopped Hijacking Stuff ⚡
Every computer-use agent until last week did the same embarrassing thing. It screen-jacked your setup like a toddler grabbing the remote.
You know the drill. Fire up Claude Cowork to do one small task, then sit there watching your own mouse do things while you stare at the screen doing nothing.
Cool demo. Terrible coworker.
Codex just broke that rule by running at the operating system level with its own cursor, so multiple agents work in parallel in the background while you keep working up front.
This is the first time your AI can actually DO the work while you do yours.
And here's the sneaky part. That unlock just quietly shipped, and barely anyone noticed.
Try This
Find one workflow that eats 30+ minutes of clicking across two or three apps daily, like scanning inboxes or cross-referencing CRMs.
Kick it off in Codex with computer use enabled, then walk away and actually work on something else.
The first time you pivot to real work while your agent handles busywork silently in the background, you'll feel the category shift.
Start small and low-risk.
Persistent memory plus Chronicle plus proactive automations is the stickiest power-user combo we've seen in three years of daily AI coverage.
What does this unlock actually look like?
Codex now schedules agentic workflows while you sleep, watches you repeat manual work, then straight up nudges you to automate it. You wake up at 7 AM and your agent has already scanned industry news, cross-referenced what competitors shipped overnight, and dropped a briefing before your coffee brewed.
This is already shipping to real teams today. Not theoretical.
Chronicle takes it further by building memory from your actual screen context, so the agent learns your stack and tone without you restating anything.
The gap between leaders who schedule and leaders who don't is about to get gnarly wide.
Try This
List every task that's low-risk AND high-frequency, like weekly reports, daily research pulls, or recurring file cleanup.
Pick the lowest-stakes one and iterate the prompt three or four times until outputs come back consistent.
Only THEN set the schedule, and daily at 7 AM tends to be the sweet spot.
Monday morning becomes coffee plus reviewing what your agent shipped overnight. That's the new bar.
Reply