• Everyday AI
  • Posts
  • Ep 773: New ChatGPT default model, Copilot Cowork goes mobile, Codex heads to Chrome and 7 more AI upgrades worth checking out

Ep 773: New ChatGPT default model, Copilot Cowork goes mobile, Codex heads to Chrome and 7 more AI upgrades worth checking out

OpenAI launched GPT-Realtime-2 and GPT-Realtime-Whisper, Google released Gemini 3.1 Flash-Lite, and Perplexity now runs AI agents locally on Macs, and more.

Author

Everyday AI
May 08, 2026

 

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Codex is heading into Chrome, Microsoft brought Copilot Cowork to mobile, and ChatGPT’s new default model is getting a lot better at memory, personalization, and accuracy. Give today’s show a watch/read/listen.

🕵️‍♂️ Fresh Finds: Microsoft upgraded Copilot with GPT-5.5 Instant, xAI’s Grok coding app is almost here, and Telegram is expanding AI across chats and automation, and more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: OpenAI launched GPT-Realtime-2 and GPT-Realtime-Whisper, Google released Gemini 3.1 Flash-Lite, and Perplexity now runs AI agents locally on Macs, and more. Read on for Byte Sized News.

💪 Leverage AI: Here’s what you missed, how to use it, and why you should pay attention. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: Google just launched a unified AI health app, Adobe turned Acrobat into an AI-powered workspace, OpenAI’s former CTO accused Sam Altman of lying during the Musk trial, and more. Check it here!

Ep 773: New ChatGPT default model, Copilot Cowork goes mobile, Codex heads to Chrome and 7 more AI upgrades worth checking out


Everything got a bit more powerful this week. 💪

Your iPhone got smarter.

Your browser now just does your work.

Your Google docs sounds more like you.

Your computer got simple claw-like features.

All you gotta do? Know what's new, how to use it, and keep up.

That's exactly what we're breaking down in our new Friday Features series, as we dish on 7 new AI upgrades you probably missed but need to pay attention to.

Also on the pod today:

• Codex Chrome extension drops 🚀 
• Explainability: why numbers matter 🧾 
• Sage Copilot’s new agentic powers 🤖 

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 –  Monid is an agent-native router for tool calls, APIEval-20 is A Benchmark for API Test Suite Generation, Fabraix Finds gaps in your AI systems before users do.

GPT-5.5 Instant in Microsoft 365 — Microsoft 365 Copilot now features GPT-5.5 Instant for faster, sharper answers and better image analysis.

New Grok App — Grok Build, xAI’s new coding app, is coming soon for Mac, Windows, and Linux. It’ll have planning mode, plugins, and even a built-in browser

Telegram AI Update — Telegram just dropped a huge AI update, letting bots jump into any chat for search, automation, and creative tasks.

LTX Studio Flows — LTX Studio just launched Flows, a node-based tool that lets you build and scale visual content workflows fast.

Cursor Drops /orchestrate — Cursor just introduced /orchestrate, a new skill that lets planners spawn workers and verifiers to automate coding tasks and self-correct in real time.

Airpods with Cameras — Apple is testing AirPods with built-in cameras that could help Siri AI see what you’re cooking or guide you with directions.

ElevenLabs Studio Agent — ElevenLabs just dropped Studio Agent, an AI co-editor that builds your video and audio edits from a simple prompt.

Anthropic Upgrades — Anthropic's Petri toolbox just got a major upgrade, making it easier and more realistic to test AI models for risky behaviors.

Spotify AI — Now you can turn your daily notes or briefings into personal podcasts, saved right to your Spotify library.

AI Partnerships — Anthropic quietly locked in a massive $1.8 billion cloud deal with Akamai to keep up with demand for its AI models

1. Google Unveils Gemini 3.1 Flash-Lite: Fastest Model Yet ⚡️

Google has rolled out Gemini 3.1 Flash-Lite, touting it as the quickest and most cost-efficient model in its Gemini 3 lineup, now available to developers and enterprises everywhere.

Built for ultra-fast, high-volume tasks, this model is already shaking up industries from software development to finance, promising lightning-fast responses and budget-friendly scalability. Flash-Lite claims to deliver smart, low-latency solutions at a price point that makes sophisticated automation more accessible than ever.

2. Perplexity Brings Local AI Agents to the Mac 🖥️

Perplexity has just launched its new Mac app, introducing Personal Computer—a major upgrade that brings its autonomous AI agents directly onto Macs and Mac minis.

The platform now works seamlessly with local files, native Mac apps, and the open web, letting teams of AI agents run tasks continuously and autonomously from your desktop. This marks a shift away from purely cloud-based operations, allowing users to harness powerful AI directly on the devices where most real work happens.

3. OpenAI Unveils Next-Gen Voice Models for Real-Time AI Apps 📲

OpenAI is shaking up the voice tech world with the release of three cutting-edge audio models, announced today.

The new API lineup includes GPT-Realtime-2 for smarter, more natural conversations, GPT-Realtime-Translate for live multilingual voice translation, and GPT-Realtime-Whisper for near-instant speech-to-text. This marks a big leap in making voice interfaces not just faster but actually useful for complex, real-world tasks, signaling that talking to AI is officially getting practical and intelligent.

4. OpenAI Unleashes GPT-5.5-Cyber for Critical Infrastructure Defenders 👮

OpenAI is rolling out GPT-5.5-Cyber in limited preview, targeting cybersecurity defenders responsible for securing vital infrastructure, as part of a broader push to democratize AI-powered defense.

The move comes hot on the heels of GPT-5.5’s release and the launch of Trusted Access for Cyber, which aims to ensure enhanced cyber tools go only to verified defenders. This initiative tightens safeguards while allowing authorized cyber teams more freedom to validate exploits and patch vulnerabilities, accelerating the security lifecycle.

5. Codex Gets Chrome Extension for Seamless Automation 🖱️

OpenAI has launched a new Chrome extension for Codex, making browser automation smoother and more efficient right from your desktop.

Codex now works quietly in the background across tabs, handling repetitive web tasks by writing and executing code as needed. The extension can mix plugins and browser tools to tackle complex workflows and logged-in sites, expanding Codex’s utility for coding, research, and dashboard management.

6. Cloudflare cuts staff as AI use surges 🪓

Cloudflare announced it is laying off about 20% of its workforce, or roughly 1,100 people, even as it posted its strongest quarterly revenue ever, underscoring how quickly AI is reshaping tech jobs.

According to the company, the cuts are directly tied to massive internal productivity gains from AI tools, not financial distress, despite the company still reporting a quarterly loss. Executives say AI has made teams dramatically more efficient, reducing the need for support roles while accelerating software development and other core work.

Seven AI upgrades dropped this week. All of them flew under the radar.

No new frontier model. No big announcements. Just quietly updated tools your team uses every single day, getting a whole lot more capable.

Every week, some teams pull ahead. This was one of those weeks.

That's what we covered on today's Everyday AI: seven features you can start using right now.

1. Codex Lands Inside Your Actual Chrome Profile 🌐

OpenAI's Codex agent can now operate directly inside the Chrome profile you're already signed into, running in parallel across tabs in the background without taking over your browser.

That last part is the unlock. Claude's computer use takes over your browser. The Claude Chrome extension loses history the moment you close it. Codex does neither.

Available now on Mac OS and Windows, included in any paid ChatGPT plan.

Try This

Install the Codex Chrome plugin today and give it one task that touches a site you're already signed into.

If it runs while you keep working, you just found your new agentic baseline.

2. Microsoft Copilot Cowork Reaches Your Phone 📱

Copilot Cowork just landed on iOS and Android, and here's what makes it different from Anthropic's version: it doesn't require your local machine. It runs in Microsoft's cloud, so the mobile app is just a remote connection to an already-running environment.

Start a task from your commute. Come back to a finished outcome.

Access requires the Microsoft Frontier program plus a Microsoft 365 Copilot premium license. Broader availability targeted around July.

Try This

Delegate one real task from your phone the next time you're away from your desk.

If the outcome is waiting when you get back, start thinking about what else your team is still handling manually.

3. Anthropic Drops 10 Finance Agent Templates 💼

Anthropic released 10 prebuilt agent templates for financial services: pitch builder, meeting preparer, earnings reviewer, model builder, market researcher, valuation reviewer, general ledger reconciler, month end closer, statement auditor, and KYC screener.

Each template packages skills, connectors, and sub-agents together so your team can compress setup from many weeks down to days or hours.

Deploy as plugins inside Claude Cowork or Claude Code. Access through the Anthropic Financial Services Marketplace on GitHub.

Try This

Pull up the Anthropic Financial Services Marketplace on GitHub and look at the KYC screener or month end closer first.

If the prewired logic maps to how your team actually works, you're a lot closer to running than you probably think.

4. Gemini Gets Persistent Instructions in Google Docs 📝

Google added persistent custom instructions to Gemini inside Google Docs. Set your rules once and Gemini applies them to every doc you work in going forward.

Not per doc. Every doc.

Rolling out gradually over the next few weeks. US and English only for now, business and enterprise accounts with Gemini alpha features enabled.

Try This

Set two or three rules you repeat constantly (tone, words to avoid, formatting standards) in Gemini settings inside Google Docs.

If it nails it on the first pass, y'all just cut your editing loop in half.

5. Perplexity Personal Computer Opens to All Paid Mac Users 💻

Perplexity Personal Computer is now available to all paid Mac users. No waitlist. No $200 a month Max plan required.

Quick clarification since the naming is confusing: Perplexity Computer runs online. Personal Computer runs locally on your machine and can actually control your Mac.

The catch? Perplexity chooses the model for you, and you'll burn through your plan limits pretty quickly. If you want full control and don't mind setup time, OpenClaw is the open-source alternative.

Try This

Update the Perplexity Mac app and give Personal Computer one task that requires moving across multiple apps.

If it handles it without you babysitting it, that's a pretty frictionless path to computer agents right there.

6. Copilot Studio Builds Workflows From Plain English 🔧

Microsoft Copilot Studio now lets you describe a workflow in natural language and it builds the nodes, connectors, and conditional logic for you. No manual configuration required.

Think of it as an autonomous agent and a traditional workflow builder combined, living inside your existing Microsoft environment.

Rolling to general availability later this month. Check your Copilot Studio environment now.

Try This

Describe one workflow your team runs manually every week and let Copilot Studio build it.

If it generates a usable flow on the first try, stop building these by hand. Don't look back.

7. GPT 5.5 Instant Becomes the New ChatGPT Default 👑

OpenAI replaced GPT 5.3 instant with GPT 5.5 instant as the default ChatGPT model. If you never touch the model picker, you're already on it, and Sam Altman said last summer that something like 93% of users never do.

OpenAI says it produces 52% fewer hallucinated claims than 5.3, with 37% fewer inaccurate claims in conversations flagged for factual errors. The writing is also noticeably less sycophantic than most default AI outputs right now.

Less "you're absolutely right" energy. More accurate, more useful.

Already live for free, Plus, and Pro users.

Try This

Run a task you'd normally Google through GPT 5.5 instant instead.

If it's cleaner and more accurate than expected, you might have a new daily driver for fast lookups.

Reply

or to participate.