The bait, then the rug-pull.
Riley Brown spends 24 minutes making one argument: the bottleneck on AI agents is not the model, it is the friction between you and the model. Then he names the seven Mac apps he has docked around Codex to remove every gram of that friction.
What the video promised.
stated at 00:16“So here are seven tools that I have downloaded on my Mac that I use to make AI agents 10 times better.”delivered at 23:14
Where the time goes.

01 · Cold open + Codex framing
Hook on the missing peripheral stack, then a 60-second primer on Codex as a Claude-Code/Cowork hybrid that can vibe-code, multitask threads, and produce documents.

02 · Tool 1 - Wispr Flow
Single-keypress voice-to-text on the Mac. Demos firing three parallel Codex agent tasks (notes app, marketing research, Neon Postgres) by voicing prompts. Pricing: free tier + paid above weekly word cap.

03 · Tool 2 - Raycast
Clipboard manager + launcher. Demos copying multiple tweets/images and pasting them into a Codex slide-deck prompt via cmd+m. Month-long clipboard history is the unlock. Free for everything shown.

04 · Tool 3 - CleanShot X
Pinned screenshots + arrow/text markup for giving Codex unambiguous visual instructions. Bonus: cmd+shift+5 video recording drag-straight-into-Twitter workflow. $30 one-time for local.
05 · Tool 4 - Paper
AI-native Figma-alternative with an MCP plug-in for Codex. Demos voice-prompting Codex to generate three design directions live inside Paper canvas pinned to the right of the screen. $16/mo annual.
06 · Tool 5 - Readwise Reader
Bookmark-driven second brain with Chrome extension + MCP. Auto-syncs X bookmarks. Runs a Readwise MCP prompt in Codex that clusters 144 saved items into a topic map and exports to Word/Excel. $9.99/mo.
07 · Tool 6 - Excalidraw + custom Codex skill
Diagramming tool driven by his custom Excalidraw Codex skill - Codex researches a topic and emits a full 12-slide visual presentation. Free for everything shown; $6/mo annual for paid sync.
08 · Tool 7 - Build Your Own
Mindset segment. Pitch: when a tool you want does not exist, prompt Codex to build a single-purpose Electron app for ~$3 of tokens. Demos his Leave a Comment annotation app for Google Docs.
09 · Summary + CTA
30-second rapid recap of all seven tools, then like/subscribe close with the I-make-multiple-videos-a-week frequency promise.
Visual structure at a glance.
Named ideas worth stealing.
The four bottlenecks an agent stack has to solve
- Input speed (voice over typing)
- Context quality (visuals + history)
- Cross-app friction (clipboard + launcher)
- Tool-fit (custom one-prompt apps)
Riley's implicit organizing principle. Every tool in the list maps to one of these four jobs.
The 7-tool AI-agent peripheral stack
- Wispr Flow (voice to text)
- Raycast (clipboard + launcher)
- CleanShot X (visual context)
- Paper (AI-native design canvas)
- Readwise Reader (second brain)
- Excalidraw (AI-generated diagrams/decks)
- Build Your Own (one-prompt Electron apps)
The spine of the video - seven Mac apps Riley keeps docked around Codex.
Build Your Own - one-prompt Electron app recipe
- Identify a tool you wish existed
- Prompt: please create an Electron app that does X
- Ask Codex to download it to your computer
- Iterate on it until it solves YOUR specific problem, not everyone's
Reframes the listicle from apps to a worldview: agent-first computing means treating tools as disposable, single-purpose, and personal.
Lines you could clip.
“These AI agents are powerful, but they are only as good as what you feed them. The input matters. The context matters.”
“Whenever there is a tool that you wish existed, you should try and build a desktop app that does the exact thing that you want.”
“This app cost me like maybe, I do not know, $3 of tokens and it is an app that I can use that has full storage that I actively use in my workflow.”
“The more you get in the habit of just building tools that solve your own problems, you are more likely to stumble on something that other people would want as well.”
“There will be a lot more tools coming around that are made specifically for your AI agents.”
How they spent the runtime.
Things they pointed at.
How they asked for the click.
“If you enjoy videos like these, please hit that like button. Please hit subscribe. I make multiple videos like this every single week, so make sure to follow-up for that.”
Soft, low-pressure, paired with a frequency promise (multiple per week). Comes after a 30-second tool recap that quietly re-establishes the value of subscribing.
Word for word.
Steal the format.
Pre-declare the axes that your tools fight on, then list one tool per axis, then close with a mindset twist that reframes the whole list.
- Open with a single-sentence thesis that names the bottleneck (here: input/context/speed). Every item in the list must visibly solve one of those bottlenecks.
- Pin a talking-head insert in the corner of every screen-capture demo - parasocial retention is free and most stack-listicles still skip it.
- Show the actual prompt you would type/voice into the agent, not just the tool UI. Real prompts are what makes the demo feel usable.
- End each segment with the price in plain dollars. It is the easiest way to make a stack feel real and trustworthy.
- Close with a mindset twist - Tool 7 should not be a tool but a worldview. That reframes the recap from shopping-list to way-of-working.
- Recap all items in 30 seconds before the like/subscribe ask - gives the algorithm completion data plus rewards skim-watchers.
- For a JoeFlow version: rebrand this as 7 Tools That Make Claude Code 10x More Powerful with JoeFlow as Tool 1 in slot one. Same skeleton, your worldview at the close.
What this could mean for you.
The single highest-leverage upgrade is not a smarter model - it is removing the friction between you and the agent.
- Install a global voice-to-text tool (Wispr Flow on Mac, or JoeFlow on Windows) and use a single keystroke to prompt the agent instead of typing.
- Turn on Raycast (Mac) or Windows clipboard history - a multi-day clipboard, especially with images, will change how you give the agent context.
- Use a screenshot tool with arrow/text markup (CleanShot X on Mac, ShareX on Windows). Annotated screenshots are the single best way to tell the agent exactly what you mean.
- Save anything interesting you read to one place - Readwise Reader, or even a Notion DB. Then point your agent at it. A second brain only pays off once an agent can read it.
- Try the Tool-7 mindset: the next time you say I wish a tool existed that did X, stop and prompt Codex/Claude Code to build a tiny Electron app for it. $3 of tokens, one weekend.
- Skip the paid tiers until the free tier becomes your bottleneck - Riley pays for Wispr/Paper/Readwise but explicitly says Raycast/Excalidraw free are enough.












































































