- Everyday AI
- Posts
- Ep 805: Codex Record and Replay: How to Teach an Agent Once Your Most Time-Consuming Workflows
Ep 805: Codex Record and Replay: How to Teach an Agent Once Your Most Time-Consuming Workflows
OpenAI just unveiled its first AI chip, Meta launched a new line of AI glasses, NVIDIA is bringing AI agents into scientific research with BioNeMo and more
Everyday AI
June 24, 2026
👉 Subscribe Here | 🗣 Hire Us To Speak | 🤝 Partner with Us | 🤖 Grow with GenAI
Sup y’all 👋
There’s been a big shift from the web to desktop lately when it comes to AI.
Wondering what AI system you’re using the most today?
What AI system are you using most?🗳️ Vote to see LIVE results 🗳️ |
✌️
Jordan
Outsmart The Future
Today in Everyday AI
8 minute read
🎙 Daily Podcast Episode: OpenAI's Record & Replay is a glimpse at a future where AI learns by watching, not prompting. Give today’s show a watch/read/listen.
🕵️♂️ Fresh Finds: MGX is raising a $50 billion AI fund, NVIDIA powers 81% of top supercomputers, Google is investing $75 million in A24 for AI filmmaking tools and more. Read on for Fresh Finds.
🗞 Byte Sized Daily AI News: OpenAI just unveiled its first AI chip, Meta launched a new line of AI glasses, NVIDIA is bringing AI agents into scientific research with BioNeMo and more. Read on for Byte Sized News.
💪 Leverage AI: OpenAI's Record & Replay could make building AI automations as simple as showing an agent what to do once. Keep reading for that!
↩️ Don’t miss out: Miss our last newsletter? We covered: Claude is expanding deeper into the enterprise, Getty is bringing licensed images into ChatGPT, and Anthropic releases Claude Tag. Check it here!
Ep 805: Codex Record and Replay: How to Teach an Agent Once Your Most Time-Consuming Workflows
Ever wish your agent would just watch you work and copy you? 👯
Oh wait, that's exactly what OpenAI released.
The new 'Record and Replay' feature in Codex allows you to simply record anything you do and turn it into a repeatable workflow for your agents.
Join us LIVE as we show you how. ⏺️
Also on the pod today:
• Codex "record and replay" demo 🖥️
• Editable skill.md files explained 📄
• Prompting vs. demonstration shift ⚖️
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 – Propane is Context-led and powered by your customer data, so product teams build what customers actually want, Rebel is Mindstone’s Fair Source AI workspace for agentic work, Ruby is an AI assistant you hire for a live call.
MGX Funds — MGX just lined up a $50 billion raise, setting the stage for one of the biggest AI funds ever.
NVIDIA Supercomputers — NVIDIA now powers 81% of the TOP500 supercomputers, and its systems also deliver more than 2x the AI training and nearly 3x the inference throughput of every other platform combined.
OpenAI Updates — OpenAI says the last six months brought 30+ API updates, including new GPT-5 variants, realtime models, and a bigger push into agent-building tools.
A24 and AI — Google is putting about $75 million into A24 to build AI filmmaking tools with DeepMind.
The Odyssey AI — An AI-generated Michael Caine is now narrating The Odyssey, which is a pretty wild next step for voice cloning.
Tencent New Release — Tencent is rolling out a new AI agent, Dayuan, for WeCom, its enterprise version of WeChat.
1. OpenAI and Broadcom unveil Jalapeño AI chip 🖥️
OpenAI has just announced Jalapeño, its first in-house “Intelligence Processor,” marking a big step from models and software into custom chip hardware built for AI inference.
The company says the accelerator was designed with Broadcom and Celestica over just nine months, and early tests suggest it could outperform current top systems on performance per watt while being tuned for today’s and future LLMs.
2. Meta launches new AI glasses line 🕶️
Meta has introduced Meta Glasses in partnership with EssilorLuxottica, rolling out a new lineup of AI glasses starting at $299 as the company pushes wearable AI deeper into everyday life.
The glasses come in three frame styles, 26 launch variants, and support prescription lenses, while keeping the core features familiar: hands-free AI, open-ear audio, voice control, photo and video capture, and privacy controls.
3. NVIDIA launches BioNeMo for lab AI 🧪
NVIDIA has just rolled out BioNeMo Agent Toolkit, a new set of tools meant to let AI agents handle scientific work in life sciences instead of merely chatting about it.
The company says the system can speed up research across biology, chemistry, genomics, and drug discovery by giving agents the context to run analyses, interpret results, and choose the next step more reliably.
4. Reflection AI signs $6.3B compute deal with SpaceX 🛰️
Reflection AI has agreed to pay SpaceX about $150 million a month for Nvidia-powered compute at Colossus 2 in Memphis, a multibillion-dollar contract running from July 2026 through 2029, according to CNBC and The Information.
The deal matters because Nvidia sits on both sides of it: it supplies the GB300 chips and has also invested heavily in Reflection, tightening the web of money and hardware around the AI boom.
5. Meta Pauses Employee-Tracking AI Program After Privacy Blowback 🕵️
Meta has paused its Model Capability Initiative after more than 1,600 employees objected to software that tracked keystrokes, mouse clicks, and screen activity to train AI systems. The backlash followed Wired’s report that some of the collected data, including prompts and private conversations, had been broadly accessible inside the company.
The move lands as Mark Zuckerberg is pushing Meta deeper into AI, with massive spending on infrastructure and a message that the best AI learns from top human work.
All that time you spent learning how to talk to AI just became optional. Now you just show it what to do.
Seriously?
OpenAI shipped Record & Replay inside Codex last week. You perform a task on your Mac while the AI watches, and Codex turns that into a skill file your agent runs on repeat, across apps, across workflows, while you do other things.
Most leaders prolly scrolled right past. You know.... "code in the name, not for me."
Biiiig mistake.
Record & Replay might be the most important AI feature shipped all year. Zero technical skill needed.
What's the ACTUAL unlock? You build autonomous AI workers by demonstrating a task once instead of writing a single instruction.
That's what we stress-tested live on today's Everyday AI, plus the three moves every leader should make right now.
Time to hit record.
1. Your next AI hire learns by watching 🔥
Prompt engineering had a good run. Record & Replay just retired it for most business workflows.
You hit record in the Codex plugin, do your task across browsers and desktop apps, then stop. Codex drafts a readable skill file capturing every action you took.
Here's what most people missed about this launch. OpenAI can now understand screen-level video through Record & Replay, a capability only Google Gemini had at scale until now.
So what happens after the recording stops?
You schedule that skill to replay at 6 AM tomorrow, daily, or on any trigger you want, with zero code and zero technical background. The leader who's been waiting for someone technical to build their automations can now do it themselves in minutes.
Try This
List three workflows you repeat weekly across more than two apps and pick the lowest stakes one. Something bouncing between tabs that takes five to 15 minutes is the sweet spot.
Download Codex, install Record & Replay, and record it while avoiding passwords or sensitive data. Read the skill file it generates before replaying.
2. This skill file is your real asset 🧾
The skill file Codex creates is not locked inside Codex. That detail matters way more than the recording itself.
It's editable, readable, and portable, meaning you can tweak any step, add fallback conditions, and share the whole thing across your team without anyone needing to re-record.
Gets wilder. Import that same file into Claude Desktop, swap the Codex-specific calls, and run the same automation on a totally different platform.
One person records a workflow. An entire department runs it.
Why now? Most enterprise AI adoption still stalls at "we paste things into ChatGPT," and skill files turn that scattered usage into repeatable, auditable processes that actually scale across an organization.
Forget feature. This is infrastructure.
Try This
Open the skill file after recording and read every step, looking for anything to adjust like login screens or pages that won't load.
Save that edited version somewhere your team can access and share it with one colleague this week. That's version one of your automation playbook.
3. Stop counting tokens, start counting skills ⚡️
Most companies measure AI by tokens consumed or seats purchased. Both miss the point.
The REAL metric is how many skills your agent runs each week without anyone prompting it.
Record & Replay makes this concrete. Validate a skill, type "run this daily at 6 AM" in Codex, and walk away.
Your agent handles it while you sleep.
What should leaders track? Minutes saved per skill per week, then how many teammates adopt each shared skill across the organization.
That's the shift. AI stops being a chatbox your team opens when they remember and becomes a system compounding value around the clock.
Low-stakes workflows first. Validate before you automate. The compound effect sneaks up fast.
Try This
Pick your most polished skill and schedule it for tomorrow morning, then check the output fresh when you wake up without touching a thing.
If it runs clean, schedule two more by Friday and track the cumulative minutes saved. That number is your proof of concept for leadership.
Reply