how i did this, video + post w/ copy/paste instructions for your agents

sessions.bensbites.com/todo-board/

On today’s episode of @alreadyhereshow I switched things up and sat down IRL with longtime friend @bentossell while he was visiting SF.

We talked about how he’s playing with AI these days.

Ben is a self proclaimed “non-technical” person intimidated by the terminal but has ion at @FactoryAI.

We get into why he keeps his agent’s context in a folder, why he doesn’t use Claude Code, and why AI makes him want to spend even more time on his computer.

shout out to @thsottiaux for the reset because i had astra high fast edit the video. lmk what you think.

Other things you will learn in this episode:
02:20 Why building silly things helps you learn how to work with AI

04:23 How Ben built a mascot animator just to make his personal site more fun

06:04 How he built a visual dictionary to understand what coding agents are talking about

08:01 Which tasks he gives Pi, Factory, and Codex—and why he loves @get_bb_app

11:41 Why he agrees with @alexatallah that faster models can feel smarter—and prefers steering agents as they work

15:17 How a shared glossary can help you and your agent stop talking past each other

24:22 Why calling AI adoption a “skill issue” lets the AI labs off the hook

31:13 How Ben keeps his personal context portable across agents and models

31:29 Why he cut back on what his agent remembers about him

42:02 Why you should ask AI questions relentlessly

45:17 Why he wants more time on his computer with his agent

Ben @bentossell wants to spend even more time on his computer

on the next episode @hnshah shares the favorite thing he built at his time at @Dropbox and the importance of showing AI what good looks like

on the next @alreadyhereshow @hnshah tries not to get too philosophical about what it means if every app can do everything

on the next episode @hnshah explains how he asks himself this one question a couple of times a month

youtube: youtube.com/@alreadyheresh…
spotify: open.spotify.com/show/033YGqcUX…
apple: podcasts.apple.com/us/podcast/alr…
x: @alreadyhereshow

on the next episode @hnshah explains not everybody needs to be like @Teslaconomics and give their @grok @bot access to their bank

on the next episode @alreadyhereshow
@hnshah talks about what he doesn't outsource to AI!

on the next episode @hnshah explains how his librarian @bot organizes his @github repo

youtube: youtube.com/@alreadyheresh…
spotify: open.spotify.com/show/033YGqcUX…
apple: podcasts.apple.com/us/podcast/alr…

on the next episode @hnshah explains how most people don't know how to get the most from models and are lazily typing to it like they do with google until @grok @bot made it easier

youtube: youtube.com/@alreadyheresh…
spotify: open.spotify.com/show/033YGqcUX…
apple: podcasts.apple.com/us/podcast/alr…

On today’s episode of @alreadyhereshow, I chatted with @cathrynlavery, a self-taught agentic engineer who learns in public and constantly fixes broken open-source projects because she knows it will make things better for everyone.

What I loved about chatting with Cat is how day friction into something an agent can handle. At one point, I realized she had given me about 12 examples in two minutes.

She makes moving to SF with 2 kids for @ycombinator seem easy because of the house manager agent she built.

There is so much to learn from her in this episode. Hope you enjoy!

Things you will learn in this episode:
01:05: How Cat became a coder using Claude and Codex
02:32: How to start contributing to open-source
05:46: How agents can handle real business operations
08:10: Three layers of protection for AI-generated code
09:25: How to build a personal house-manager agent
10:47: Why effective agents need tools, not just prompts
15:10: Mac Mini can function like an $800 employee
17:50: How deterministic tools make agents more reliable
19:16: How to turn years of voice memos into searchable memory
20:48: Why AI is only as good as the context you give it
23:38: How to create a self-improving agent system
27:00: How to separate safe automation from actions requiring approval
29:03: How to synchronize skills across multiple computers
35:24: The best place to begin using AI

You can also watch/listen at your preferable dealer:

on today's episode of @alreadyhereshow we have @runsonai talking about how his @NousResearch hermes agent set up lets him do things that @claudeai can't!

I love how he clearly hammers home that most executives don’t need more AI tools they just need to review their repeated ose first.

He talks about the use case he came across where a team was manually getting the prices from Amazon of 186 products that took them a couple of hours to get until he showed them how to do it with AI.

His 10-80-10 approach and much more.

Hope you learn something new and are inspired to implement personal ai in your daily efforts.

---
Things we discuss:

05:16: How to meet people where they are instead of forcing every AI tool on them.

12:03 : Why you should begin with one repeated task and one useful result.

13:13: How to choose between a local agent and a cloud-hosted setup.

20:09: How to give different AI assistants clear responsibilities.

25:53 : How an automated meeting brief turns email, messages, and research into useful context.

27:03: How AI lead enrichment helps teams qualify opportunities faster.

34:33: How the 10-80-10 model keeps humans in control: define the work, let AI execute, then review.

43:10: Why building a personal project teaches you more than passively watching tutorials.

According to @mvanhorn if your prompting is good enough you should never read your agents plan

Last week I launched the @alreadyhereshow to showcase how people are already living in the future with personal ai.

While editing the episode for tomorrow's show with @mvanhorn I came across @mattshumer_'s gauntlet prompt for creating games and asked fable to create an open can come together and connect instead of a traditional landing page

I let fable decide on the infrastructure and vehicles and game play and most everything else.

Only requirement I made was to be performant and work well on mobile.

I gave it access to opus 5, codex sol, grok 4.5 and kimi k3 via my fable delegators skill github.com/contains-studi…

6.4 Billion tokens or so later I think its in a good state to share.

Here's the promo video it created with @ElevenLabs

And if you want to hop into a world with me I'll be in this one for next hour or so

alreadyhere.show/w/msg

In an effort to drain all our weekly tokens @msg along with claude, codex, grok and kimi created a game

@dwarkesh_sp AI is already superhuman at many things.

We are in the singularity.

Stop clicking on apps and have your agents do it for you

How @samgaddis using Hermes @NousResearch to post to his website for him

“The future is already here – it's just not evenly distributed.” @GreatDismal

I know lots of people who are already living in the future but how they are navigating the new world order in their every day life tends to constantly be in flux.

• How/What they delegate.
• How AI into their lives
• When they choose to stay in the AI loop.

so I messaged a bunch of friends to see if they would yap with me on camera and share how they are already living in the future and decided to call it the @alreadyhereshow

@samgaddis was the first person to agree to share his experience with the personal AI system he built for himself and his team that can turn meeting notes into email drafts, coordinate work through Slack, update his LinkedIn, maintain software, capture new ideas and even help him and his wife fight/communicate better.

One thing that stood out: Sam doesn’t use @NousResearch's Hermes agent for coding or let it touch client repositories. He uses it for wacky one-off ideas he doesn’t want to sit in front of Codex or Claude Code to build.

Watch/Listen to the show and let me know what stood out to you.

Timestamps:
00:00 — The AI that helps Sam and his wife fight better
00:47 — Meet Sam Gaddis and his AI system
02:25 — Turning meeting notes into client email drafts
04:05 — Pointman: Sam’s AI assistant inside Slack
05:18 — Use Hermes for wacky one-off ideas not coding
06:08 — Turning Twitter likes into website content
08:04 — Permissions, security and earning trust
09:42 — An AI agent that audits and repairs itself
11:57 — Controlling Hermes through Siri
13:58 — “I’m not clicking through apps anymore”
16:54 — Working with agents away from the desk
20:23 — Personal AI beyond client work
21:13 — Building an AI relationship coach
26:41 — Turning rough ideas into interactive prototypes
32:15 — Why your AI system should be model-agnostic
34:36 — The coming transformation of video
37:31 — Why so much AI content feels disposable
40:00 — Why we’re still at the very beginning
43:24 — Human taste is still the signal
44:56 — Where this is all going according to Sam