Apr 27, 2026 · 1h 4m · startup-ideas
Stop using Claude. Start using Codex?
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this masterclass, Greg Isenberg and Riley Brown explore OpenAI's Codex, demonstrating how its multi-agent graphical workspace, autonomous browser capabilities, custom skills, and programmatic video tools redefine productivity for creators and technical founders.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 18.5% of the talking time here. How this is scored →
Greg as informed peer 3.3 Guest teaching 5.4 Guest disagreement 0.9 Greg pushing back 2.3
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 11:34 Riley rejects Anthropic's Cowork product strategy
Riley aggressively dismisses Claude Cowork as restrictive and flawed, directly rejecting the premise that separating business and coding agents makes sense.
Hardest push from Greg ▶ 38:16 Greg drills Riley with critical questions on product failure modesReacting in real-time to AI feedback that his podcast is too soft, Greg drops his agreeable posture and demands hard answers regarding churn, costs, and who should avoid Codex.
Biggest teaching moment ▶ 43:25 Riley reveals running Claude Code inside Codex terminalRiley demonstrates how to bypass interface trade-offs by running Claude Code directly inside the Codex terminal in skipped-permissions mode to reap flat-rate subscription benefits.
Greg holds their own ▶ 50:04 Greg shuts down Sora hype with market realityGreg demonstrates sharp industry knowledge by pointing out that Sora was delisted from app stores and redirects the conversation away from toy generative models toward high-leverage business tooling.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Greg as informed peer | Guest teaching | Guest disagreement | Greg pushing back | Why |
|---|---|---|---|---|---|---|
| Setting the Stage: Claude Loyalist vs. Codex Evangelist | 2 | 3 | 1 | 3 | Greg introduces the premise as a skeptic who has never used Codex and warns he might conclude the hype is just smoke. Riley accepts the challenge collaboratively, outlining the features he plans to showcase. | |
| Codex Interface Tour and Multi-Agent Workspace Architecture | 0 | 6 | 0 | 0 | Riley provides a guided visual walkthrough of the Codex multi-agent architecture and project hierarchy. Greg acts purely as a receptive audience. | |
| The Transition from Terminal Interfaces to Unified Agent GUIs | 1 | 6 | 1 | 0 | Riley explains the industry shift from terminal interfaces to unified GUIs and criticizes Anthropic for splitting Claude Code and Cowork. Greg listens without challenging. | |
| Deep-Dive: Codex Strengths, Complex Coding, and Atlas Web Browser | 3 | 6 | 2 | 2 | Greg mentions his interest in Claude Cowork, prompting Riley to reframe Cowork as overly restrictive and point out Codex's superior integrated Atlas browser and complex coding capabilities. | |
| Programmatic Video Creation with the Remotion Plugin | 4 | 6 | 1 | 4 | When Codex experiences a connection drop mid-demo, Greg playfully calls out the irony of hyping an unstable tool. Greg also contributes domain knowledge about creator growth using Remotion code-based video. | |
| Advanced Computer Control, In-App Browser, and Chronicle Memory | 3 | 6 | 0 | 2 | Riley explains computer control and the new Chronicle background memory feature. Greg raises practical concerns regarding user privacy and security trade-offs. | |
| Deciphering the Ecosystem: Plugins, Custom Skills, and Tool Integrations | 4 | 5 | 1 | 3 | Greg points out the confusing overlap between plugins and custom skills in Codex's interface, while also contributing an onboarding workflow idea using Google Calendar integrations. | |
| Live Execution: Automated Channel Critiques and Recurring Workflows | 2 | 6 | 0 | 1 | Riley demonstrates extracting channel transcripts to generate critique reports and establishing automated recurring Friday tasks. Greg asks how to refine outputs when results miss the mark. | |
| Output Evaluations, Example Databases, and Workflow Recording | 5 | 5 | 0 | 1 | Greg introduces technical terminology regarding automated evals on subjective outputs and identifies market opportunities around screen recording datasets for future models. | |
| Addressing Podcast Critiques: Hard Questions and Codex Target Users | 6 | 4 | 2 | 7 | Prompted by Codex's critique that his interviews are too soft, Greg deliberately pushes Riley with challenging questions on failure modes, costs, and who should avoid Codex. | |
| Live Autonomous Browser Demo: AI Playing Chess Against Itself | 3 | 5 | 0 | 0 | Riley shows Codex controlling an in-app browser to play chess against itself. Greg offers an apt comparison between earlier browser agents feeling like dial-up versus Codex feeling like broadband. | |
| Running Claude Code in Codex Terminal and Model Cost Economics | 5 | 6 | 2 | 4 | Riley reveals how to run Claude Code within Codex terminal to maximize subscription subsidies. Greg pushes on the steep API costs of GPT-5.5, prompting Riley to explain cost per completed task. | |
| Visual Canvas Annotations, GPT Image 2.0, and Video Model Workflows | 5 | 5 | 2 | 5 | When Riley mentions the Sora plugin, Greg dismisses it as a novelty, noting it was delisted from app stores, and refocuses the discussion on practical tools like Remotion and FAL. | |
| State of Generative Video and Notion Workspace Synchronization | 5 | 5 | 1 | 3 | Greg explains the cognitive overload regular knowledge workers feel when managing fragmented AI stacks, comparing native Notion AI workflows with Codex integrations. | |
| Four Beginner Starter Projects for Mastering Codex | 2 | 7 | 0 | 0 | Riley outlines four concrete starter exercises to master Codex and browser use. Greg warmly endorses the experimentation mindset and wraps up the show. |
Statements from this episode (25)
Brown: OpenAI's Codex is the most powerful interface for AI agents
“People are going to learn how to use the best interface to use AI agents, in my opinion. I think Codex by OpenAI, and that means if you have a ChatGPT subscription, you have access to Codex, which I think is the most powerful way to use AI agents”
Brown: Constantly switching between trending AI tools is the wrong strategy
“I think you have, you should pick a stack and you should stick with it. I think moving from hot tool to hot tool is actually not the right strategy.”
Brown: AI toolmakers are converging on folder-and-thread agent interfaces
“All of these companies are deciding that this is kind of the correct interface to interact with AI agents.”
Brown: Codex beats Claude Desktop by unifying coding and coworking tools
“Another way that I like to think about this is Claude code or this codex is like Claude code and co-work in a single product, but it's even better than that. And the reason why I like it more than Claude desktop is Claude made a decision to split up co-work an…”
Brown: 95% of software development is now as easy as making presentations
“I think vibe coding has gotten so easy that like, 95% of the things that you would want to code, it's as easy as creating a presentation.”
Riley Brown: Codex is the only platform combining vibe coding and knowledge work
“It is the only interface that does vibe coding and knowledge work in one platform.”
Riley Brown: GPT 5.4 one-shotted a Replit clone in 80 minutes
“I've one-shotted, the other day I one-shotted an app like Replit, a vibe coding tool, that spun up sandboxes, and you could vibe code web apps on your phone in one shot. It took an hour and 20 minutes to generate the code, and this was on 5.4.”
Riley Brown: OpenAI consolidated efforts to integrate Atlas browser inside Codex
“Well, recently they had like this big red alert at their company, and they decided to kind of push all of their efforts into one single app, which was Codex. And now what they're doing is they're just putting Atlas inside. They're basically putting Atlas insid…”
Anthropic used Remotion to generate early product launch videos
“Anthropic did this when they first started doing their product launches. They were just launching them With ReMotion videos.”
Isenberg: Creators gained 100k Instagram followers posting Remotion videos
“I know multiple people who've gone from zero to a 100,000 Instagram followers. By doing re-motion videos.”
Riley Brown: Execute AI tasks once before turning them into automations
“This is how I normally set up automations. People sometimes want to create an automation before they do it. It's just like do it once and then tell the agent to do that every time on this day. And that's a kind of a good workflow.”
Brown: Collecting output examples is top priority for enterprise AI implementation
“The number one thing that you can do in a company right now, and we're talking to a lot of companies that are trying to implement AI, is just collect as many good examples as possible. This allows your agents to have examples that they can reference.”
Brown: Codex autonomously built and played its own chess game
“And so, in my video, it did this in one prompt. I said, build a chess game, then use browser use to test the game and play yourself until one team wins. And that's what it did in one shot”
Brown: AI will autonomously build and use custom tools within a year
“If you think about where we were even a year ago and then you think about where we will be a year from now, you know it may just create its own Canva a year from now and then use that. It might be like, oh, it'd be way faster for me to just create my own tool …”
Brown: Claude and Codex Subscriptions Provide Thousands in Subsidized Tokens
“Like, in Claude or in Codex is you just save so much money because they're subsidizing, like, you get thousands of dollars of tokens for your a hundred dollar per month or 200 dollar per month plan because you're using their platforms.”
Brown: GPT-5.5 API Costs Double That of GPT-5.4
“Yeah, if you use it via the API, it's twice as expensive as Ah, the recent model, which is, ah, GPT-Five-Point-Four. It's twice as expensive.”
Brown: AI Benchmarks in One Year Will Measure Cost and Time per Task
“That's where I think the benchmark thing will be a year from now. It's not going to be how token efficient is it. It's going to be how much money and how much time does it cost to do a specific task?”
Brown: GPT-5.5 API Costs 20% More Than Claude Opus 4.7
“And then it's also 20% more expensive than Opus 4.7.”
Brown: GPT Image 2 is undeniably the best image model in the world
“Codex has GBT image two, which is the best image model in the world. And again, this is not like up for debate. I'm not, again, I'm really not trying to show anything.”
Brown: Sora is not the best AI video model right now
“Sora's not the best video model right now, so that's why it's not interesting”
Brown: A startup building an AI video editor in Codex will make a killing
“I'm sure there's going to be a company that only builds, like, an AI video editor like ReMotion that shows up inside Codex, and they're going to make a killing, right?”
Brown: AI will control browsers as well as humans within three months
“Cause this is not a year from now. This is like three months away. AI will be able to control any browser like as good of a human.”
Brown: Codex can generate complete working mobile apps from one prompt
“For this, for now, you still need to have a Mac, and you need to have Xcode downloaded, but it can one-shot mobile apps, like full mobile apps.”
Brown: Successful startup founders prioritize tinkering over immediate tool ROI
“What I've noticed, you know, and being in San Francisco I've noticed that the people who have been, like, really successful creating startups are people who just, like, want to experiment and tinker, and they don't have so many people who are, like, type A, wa…”