• Everyday AI
  • Posts
  • Ep 781: OpenAI’s Codex For Beginners Pt 2: Advanced Agentic Strategies broken down for Newbies

Ep 781: OpenAI’s Codex For Beginners Pt 2: Advanced Agentic Strategies broken down for Newbies

Your complete Codex guide, Google just launched Gemini Spark and major AI Search upgrades at I/O, OpenAI is locking customers into long-term compute deals, and Intuit makes big AI cuts and more.

Author

Everyday AI
May 20, 2026

Sup y’all 👋

Mighta went overboard with this one.

But I’ve been using the Codex desktop app as my main driver since February, and the Codex demand has been silly.

So we put together an insanely in-depth Codex Cookbook: best practices, 50+ ready-to-deploy use cases and more. Listen to today’s show and then put these Codex recipes into place, and you’ll pretty much be from the future and your coworkers/competitors are gonna be caught off guard.

Not gonna charge you a dime for this, though I feel I should. Maybe then I could afford to sleep? lolz

Just repost today’s show on LinkedIn, and I’ll shoot the complete Codex Cookbook over to you.

✌️

Jordan

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Codex is no longer just a coding tool. It’s becoming an always-on workflow system that can connect apps, run tasks, use your computer, and let you control the whole thing from your phone. Give today’s show a watch/read/listen.

🕵️‍♂️ Fresh Finds: Google is pushing Gemini into science, shopping, and smart glasses, Anthropic is expanding Claude’s security and philosophy work, and DeepSeek may be making Codex competitor, and more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Google just launched Gemini Spark and major AI Search upgrades at I/O, OpenAI is locking customers into long-term compute deals, and Intuit makes big AI cuts and more. Read on for Byte Sized News.

💪 Leverage AI: Most companies are still using AI like a chatbot. Meanwhile, tools like Codex are becoming full workflow systems that can actually run work instead of just answering questions. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: Google’s new Gemini 3.5 Flash drop, Musk loses OpenAI lawsuit and OpenAI cofounder joins Anthropic. Check it here!

Ep 781: OpenAI’s Codex For Beginners Pt 2: Advanced Agentic Strategies broken down for Newbies


Every wish on your AI wish list? 🧞‍♂️

It's already here in OpenAI's Codex.

While the masses are months behind just now discovering Claude Cowork, there's a new wave taking over in Codex.

Bad part? It has the word 'code' in the name, so most people assume they're not technical enough.

Good news? If you can use ChatGPT, you can use Codex. (It's pretty much the same thing.)

We already broke down the Codex basics in Part 1 of our AI at Work on Wednesday series, so we're doubling down on the Codex classroom in part 2.

We'll tackle Skills, Plugins, Computer Use, Remote Control and more.

But the big takeaway? Codex can probably already do most of the work you do in front of a computer. We'll show you how.

Also on the pod today:

• Codex: AI’s best-kept secret 🤫 
• Super app for automations 🚀 
• Super app for automations 🚀 

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 –  StoreClaw Grows your store profits with agents that know how to sell, Runtime is Sandboxed coding agents with your company's context, integrations, and guardrails, Owlish turns your knowledge into AI agents that answer, act, and hand off across your website, social channels, and team chat apps.

Deepseek Code — DeepSeek is forming DeepSeek Code, a Beijing team to build a new Code Harness product. Two roles are open now, a Harness Product Manager and an R&D Engineer

Google Eyewear — Google revealed audio-based "intelligent eyewear" running Gemini that gives hands-free help — directions, calls, translations, photos and more

Claude Philosophy — Anthropic is talking with religious, philosophical, and civic groups to shape Claude’s moral character and values, and already testing tools that help the model pause and reflect before key decisions.

Cohere New Model — Cohere just open-sourced Command A+, a sovereign MoE model you can run on-prem for cheaper, powerful multimodal work.

Dont fight AI — HSBC urges staff to embrace AI, warning it will cut some jobs while creating others.

Google Updating Pro Limits — Google cut AI Ultra pricing but quietly tightened AI Pro limits, switching to a credit system that charges by compute and adds five-hour usage windows.

NVIDIA and Google — NVIDIA and Google Cloud are expanding a joint developer community with new JAX and Dynamo labs, Gemma and Nemotron integrations, and resources to build production-ready AI on NVIDIA GPUs.

AI Employees Study — Employees treat personal and work AI the same: 64.5% of “personal” AI use is actually business work, and 45.6% of personal AI minutes happen on employer-paid accounts.

Gemini for Science — Google rolls out "Gemini for Science," a suite of agentic tools that speed hypothesis generation, large-scale computational experiments, and literature synthesis for researchers.

Google Genie Upgrade — Google’s Project Genie can now ground imaginative virtual worlds in real Street View locations in the U.S., letting you remix places like the Golden Gate Bridge into scenes such as underwater or 1920s B&W.

IBM and Anthropic — IBM just bolstered its AI security stack and teamed up with Anthropic on Project Glasswing to harden open-source supply chains.

OpenClaw and Grok — Grok and X Premium subs can now be used inside OpenClaw, so you can chat with agents, generate images/videos, and search X from one place.

Google Search Updates — Google is expanding content provenance and verification across Search, Chrome, Pixel, Gemini and Cloud, using SynthID watermarks and C2PA credentials to show when media is AI-generated or camera-original

AI Powered Shopping Hub — Google just unveiled Universal Cart, an AI-powered shopping hub that tracks price drops, checks compatibility across items, and helps you checkout with Google Pay.

1. Google launches Gemini Spark: a proactive personal agent that auto-manages your data and tasks ✨

Google yesterday unveiled Gemini Spark, a paid, agent-style feature that proactively mines users’ personal data to perform tasks like organizing emails, flagging charges, drafting follow-ups, and automating shopping, with a limited early tester rollout and a beta for subscribers next week.

Spark distinguishes itself from the regular Gemini chatbot by taking action on your behalf without prompts, connecting soon to third-party services and gaining browser and texting controls in future updates.

2. OpenAI launches “Guaranteed Capacity” deals as Sam Altman arrives at Oakland courthouse in Musk lawsuit 🏛️

OpenAI on Tuesday began selling multiyear Guaranteed Capacity contracts that lock customers into one- to three-year compute commitments with rising discounts, signaling a push to secure long-term revenue and capacity as demand for AI compute grows.

The move arrives as CEO Sam Altman appeared at an Oakland federal courthouse the same day in Elon Musk’s lawsuit over OpenAI’s for-profit conversion, underscoring intense industry and legal scrutiny. OpenAI says the program helps customers secure certainty for building AI products while the company reserves capacity for flagship services like ChatGPT and Codex.

3. Google I/O 2026: Google Search gets Gemini 3.5 Flash, AI Search Box, and 24/7 Search agents 🔍

At Google I/O 2026 Google upgraded Search with Gemini 3.5 Flash powering AI Mode and introduced an expanded AI Search Box that accepts natural-language queries plus images, video, files, and tabs to provide richer context.

AI Overviews now support back-and-forth follow-ups so users can probe results like a conversational assistant, and new Search agents will continuously monitor the web for updates and deliver tailored results, with agentic booking and calling features rolling out broadly this summer.

4. Mythos’ hacking scare cools as experts say model speeds discovery but not exploitation 👾

A month after Anthropic’s Mythos raised alarms by reportedly finding thousands of software flaws, cybersecurity practitioners say the immediate danger is overstated and the real news is practical: Mythos meaningfully speeds vulnerability discovery but does not automatically enable large-scale exploit campaigns.

Experts note the bigger challenge is validating, prioritizing and fixing the flood of bugs Mythos can surface, and that substantial compute, tooling and security “harnesses” are still required to use the model effectively.

5. Intuit to cut 17% of workforce, about 3,000 jobs, to reallocate resources toward AI integration 💸

Intuit announced a major restructuring that will eliminate roughly 3,000 positions as the company simplifies its corporate structure and redirects spending into embedding AI across products like TurboTax and QuickBooks, Reuters reports.

The move comes amid strong industry demand for AI even as Intuit’s stock has lagged the S&P 500, and follows a year of widespread tech layoffs tied to similar AI-focused rationales.

The biggest AI advantage in business right now is hiding behind a word you may be scared of: code.

That’s the trap.

Your team hears OpenAI Codex and thinks engineering. Meanwhile, the real play is using it as an always-on work layer that connects apps, remembers context, runs repeatable workflows, controls desktop tasks, drafts deliverables, and lets you steer the whole thing from your phone.

The mythical AI agent that could do everything that the masses have wanted for years actually exist now in Codex, yet hardly no one has noticed. 

And that can be a GREAT thing for your business if you look past that four letter word: code. 

On today’s Everyday AI, we’re breaking down Codex for beginners part two: plugins, skills, computer use, scheduled automations, and remote control. 

Translation? Less “ask AI for help” and more “teach AI the workflow, review the output, and let the machine handle the grind.”

Most teams are still playing chatbot ping-pong.

The ones paying attention are about to turn repeatable work into scheduled agents.

1. Stop treating Codex like code 🔥

The name is doing everyone dirty.

Codex sounds like a developer tool, so nontechnical teams ignore it. That’s the miss, because the real value is turning repeatable business work into something AI can actually operate across apps, files, projects, and tools.

Most companies are still stuck in the weakest version of AI adoption.

Prompt. Copy. Paste. Open another tab. Ask again. Lose context. Pretend the workflow improved because the answer came faster.

Nah.

If the human is still stitching every step together, the workflow didn’t transform. The person just became middleware with a paid ChatGPT account.

Try This

Find one recurring workflow that touches at least three tools and makes a smart person do dumb coordination work.

Don’t ask, “Can AI help with this?” Ask, “Can Codex own the assembly while the human owns the judgment?”

2. Turn messy work into machines ⚡

The guest-prep workflow was the business case hiding inside the demo.

Not because every company books podcast guests. Obviously.

Because every company has its own version of that mess: too many inputs, scattered emails, old forms, rogue spreadsheets, calendar context, analytics, research, approvals, prep docs, follow-ups, and some poor soul trying to make it all coherent before the real work starts.

Codex pulled across Google Calendar, Gmail, Google Drive, Spotify analytics, StreamYard, Canva, Google Slides, Google Docs, and a web app workflow.

But the asset pile wasn’t the point.

The point was the chain: find the work, gather the context, compare the inputs, draft the next step, package the deliverables, keep the human in the loop, and make the whole thing repeatable.

Try This

Pick one workflow your team avoids because it’s fragmented, annoying, or too easy to mess up.

Write the delegation brief like you’re handing it to a sharp operator: where to look, what to compare, what to create, what not to send, and when approval is required. Run it manually twice before you schedule anything.

3. Manage agents before they multiply 🚀

This is where leaders need to get serious.

Plugins connect Codex to real tools. Skills preserve the way you want work done. Computer use lets it click and type. Automations let it run on schedule. ChatGPT mobile turns your phone into the remote control.

All of a sudden, the ChatGPT app can control your entire digital worklife. (Not just one random Dispatch thread like in Claude Cowork. We finally have a true remote controllable super agent y’all!) 

And digital labor needs management. The Gmail example alone should make operators sit up: 24 actions, including 11 write actions and 13 read actions.

Useful? Extremely.

Something to roll out with zero guardrails? Absolutely not.

The winners won’t just be the teams with Codex access. Everyone will get access eventually.

The winners will be the teams that turn workflows into reusable assets, monitor execution, set approval rules, and teach people how to manage agents instead of babysitting chatbots.

Try This

Create an agent permission map before rollout.

For every Codex workflow, define what it can read, what it can write, what stays draft-only, what needs approval, what must be reviewed, and what happens when it fails. Start boring. Make it reliable. Then scale.

That’s the new AI work equation: humans design the system, agents do the grind, humans judge the output.

Everyone else is still asking chatbots for bullet points.

Reply

or to participate.