The argument in one line.
Claude, Codex, Google, and Cursor are all converging on the same agentic super-app architecture, but only the companies that understand what they are building will win the race.
Read if. Skip if.
- AI practitioners who want a single weekly briefing across Anthropic, OpenAI, Google, and Cursor instead of following each separately
- Builders using Claude Code, Codex, or Cursor who want early signal on capability changes that affect their workflow
- Founders staying on the frontier of AI agents who want the signal without drowning in Twitter noise
- People who want deep technical analysis — this is a fast weekly briefing format, not an engineering deep dive
- Viewers who prefer primary sources and already follow all four companies directly
- Those not interested in AI agent tooling specifically — this is not a general AI news roundup
The full version, fast.
The major AI agent platforms — Anthropic, OpenAI/Codex, Google, and Cursor — are all converging on the same product category, and this weekly briefing maps what each shipped in a single week. Anthropic launched a multi-task terminal interface for Claude agents and hired Andrej Karpathy. Codex added slash goal mode for multi-day autonomous tasks and a plugin sharing system. Google previewed Spark, their answer to Claude's agent terminal, alongside Gemini 3.5 Flash. Cursor signed a deal with SpaceX and shipped Composer 2.5. The synthesis: the race is clearly toward persistent, long-running AI agents that can operate autonomously across sessions — and the platform that gets multi-day autonomous execution right first will define the next category.
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.

01 · Cold open
Three rapid-fire teasers before the intro — Codex /goal, Karpathy at Anthropic, Google Spark

02 · Series intro
Riley introduces himself and the weekly This Week in Super Apps format

03 · Claude / Anthropic updates
Multi-task terminal demo (claude agents), Karpathy hire and talent wave, desktop app improvements, Cowork vs Claude Code friction

04 · OpenAI / Codex updates
/goal mode for long-running agentic tasks, plugin sharing across teams, full design mode with annotations, AppShots (command+command context capture)

05 · Google / Gemini
Google IO reaction (nothing burger except Spark), Gemini Spark super-app preview, Gemini 3.5 Flash disappointment, verdict: not best at anything

06 · Cursor
SpaceX acquisition at $60B valuation, Composer 2.5 model demo (faster than Codex/Claude for front-end), super-app trajectory confirmed

07 · CTA — chorus.com
Riley's own product: Claude Code and Codex accessible via iMessage/WhatsApp; live demo; $10 free credits offer
Lines worth screenshotting.
- Anthropic is pulling in CTOs from some of the most prestigious tech companies in the world to become individual contributors — not executives.
- The era of the polymathic individual contributor has arrived: experts with rare, deep skill sets are being recruited for hundreds of millions of dollars.
- Google built the research that underlies most of modern AI and currently holds the top spot in none of the categories their products compete in.
- Cursor was acquired by SpaceX for $60 billion with a $10 billion opt-out clause, giving the coding tool immediate access to enormous compute resources.
- A super app is defined by three capabilities in one platform: chat, knowledge work, and coding — plus integrations to everything you already use.
- Codex's slash-goal mode keeps the agent oriented toward an outcome rather than a single command, enabling tasks that run for over a day without losing context.
- The annotation feature in Codex lets you click any UI element, describe what you want changed, and then fire off all annotations in one batch run.
- Pressing both command keys simultaneously in Codex takes a screenshot of your current screen and immediately passes context to the agent.
- Sharing plugins across a Codex workspace means teammates can build once and distribute automations without each person rebuilding the same workflow.
- Google's Gemini Spark has the architecture to be a genuine super app contender but is being diluted by parallel investment in too many overlapping products.
- Claude Code and Codex treat their in-app browsers as full browsers — this is the interface layer that will replace traditional app switching for enterprise work.
- Gemini 3.5 Flash is fast and cheap but not frontier — even Google's own researchers have stopped recommending it for serious agentic work.
Launch the weekly roundup — and own a platform beat.
Riley is not just covering the news — he is franchising a weekly correspondent slot in a category moving too fast for anyone to track alone.
- The weekly update format compounds: viewers subscribe to the cadence, not just any single video.
- Open with 3 rapid-fire teasers before your intro card — frontload the reason to stay.
- Each segment needs a thesis, not just a summary. 'Google is not the best at anything' is watchable; 'here are Google updates' is a changelog.
- Demo-to-talk ratio matters — screen-recording yourself using the tools live is the trust multiplier.
- The whiteboard org chart with columns per company is a low-effort visual anchor that gives structure at a glance.
- Pick one fast-moving category in your lane, commit to weekly coverage, have a POV on each player's trajectory.
- Soft-sell your own product as the final segment — demo it inside the content, not as an ad break.
Terms worth knowing.
- Super app (AI)
- An AI platform that combines chat, knowledge work, and coding in a single interface, connected to external tools and integrations, so users can handle diverse tasks without switching between products.
- Claude agents (terminal)
- A terminal interface mode in Claude Code that lets users fire off multiple concurrent tasks and navigate between them simultaneously, rather than working in a single linear conversation.
- Codex (OpenAI)
- OpenAI's AI-powered super app for coding and knowledge work, offering a unified interface with skills, integrations, plugin sharing, and an in-app browser for agentic tasks.
- Slash goal (/goal)
- A Codex command that sets a high-level objective for the current session, keeping the AI agent oriented toward an end outcome across multiple steps and hours-long tasks rather than responding command by command.
- Plugin (Codex / Claude Code)
- A packaged bundle of skills, document templates, and API integrations inside an AI agent platform that can be installed, reused, and shared with teammates to handle specific workflows.
- AppShots
- A Codex feature that lets users take a screenshot of any open application and immediately send it to the AI agent with a prompt, enabling context-aware tasks without switching windows.
- Browser use
- The ability of an AI agent to autonomously navigate and interact with a real web browser — clicking links, filling forms, reading pages — as part of completing a task.
- Cursor
- An AI-powered code editor and agentic coding platform that includes its own AI model (Composer), an in-app browser, and marketplace integrations, competing directly with Claude Code and Codex.
- Gemini Spark
- A leaked upcoming feature in Google's Gemini app designed to function as an agentic workspace — with virtual cloud computers, app integrations, and long-horizon task execution.
- Computer use
- An AI capability that allows a model to control a computer's interface — moving the cursor, clicking buttons, typing text — in order to operate any application as a human would.
- Cron job
- A scheduled task set to run automatically at specified times or intervals, commonly used to trigger recurring automations in software systems.
- Polymathic individual contributor
- A highly skilled specialist who operates independently rather than in a management role, valued for deep expertise across multiple disciplines rather than organizational authority.
Things they pointed at.
Lines you could clip.
“Nothing that they've created is state of the art.”
“I think Cowork was one of their biggest mistakes.”
“Just give me one single app that can do anything, coding task or knowledge work.”
“It is the era of the polymathic individual contributor.”
Word for word.
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
The bait, then the rug-pull.
Three bullets land before the title card even drops: Codex now lets agents run goals for more than a day, Anthropic hired Andrej Karpathy, and Google is building their own answer to both. Riley Brown calls it the first ever AI super app update — a weekly brief designed to separate signal from changelog noise across the platforms that actually matter.
Named ideas worth stealing.
The Super App Definition
- Can chat with it
- Can do knowledge work with it
- Can do coding tasks with it
- Connects to all existing integrations
- Supports automations
- Has a full in-app browser
Riley defines exactly what a super app means in the AI agent context — a 6-part checklist he uses to grade each platform throughout the episode.
Platform Report Card
- Claude: best chat, best desktop app, Cowork friction
- Codex: best agentic tasks, /goal + AppShots are novel
- Google: not best at anything — fragmented across antigravity / Spark / AI Studio
- Cursor: fastest front-end model, super-app path confirmed post-SpaceX
Implicit framework Riley uses to evaluate each platform — who owns what use case.
How they asked for the click.
“If you guys wanna support me and take a look at the project that my team and I have been working on, we've found a way to add Claude Code and Codex to iMessage.”
Soft and demo-forward — shows the product live in iMessage, names the free credit offer, low-pressure. No urgency language. Works because the product is directly relevant to the audience.









































































