Let’s talk about this…
Where do I even start? It seems like a lot of people are feeling this too. The constant barrage of new products, models, capabilities, threats, whatever is one thing. But seeing people with 20 personal agents for managing their social content or just life.
What are we doing here?!
When I first started the newsletter it was to keep up with everything that was going on. It was 5 days a week. It was so noisy I decided to cut it down (along with our revenue) to just two days. I try and cut through the noise with the links here but the noise is still there, it may be louder.
I read two excellent posts yesterday. Here’s a snippet from each
You’re doing your best to keep up. You’re 9-9-6’ing, sometimes 0-0-7’ing. You didn’t actually rest during rest week, did you? What are your agents up to? Hopefully getting a lot done while you’re reading this. Have you tried the latest personal assistant? Optimized your workflows to accommodate the strengths of the latest model? Have you gone viral on X lately? Have you been angel investing? Have you locked in your Bay Area mortgage? If not, you know you’re fucked, right? You’re enslaved to your keyboard, timeline, and Slack. You’re tired, but your ambition is also Icarus-ing. Your FOMO is spiking. Your capacity is redlining. Is your moral compass spinning?
The Sane Asylum by Colossus
To be blunt, instead of having to sit down and “do” a thing, you can, while hungover in sweats on your couch, grab your laptop, fire up Claude Code, Codex, GrokBot, whatever, and just tell it to start building whatever epiphany is top of mind at the moment. Oh, and you don’t even have to type it anymore. Hands too sore? You can double tap the function key and your space bar, have Wispr Flow start transcribing, and say, “I want you to build a tool that each day goes through all newsletters I’m subscribed to, find the key highlights from each that I should care about, and send me a summary. Make this a daily routine, thanks.” Then go back to vegging out on your couch while your little token muncher cranks away.
The Difference Between “Can” and “Should” by Jack Raines
I start far more than I finish. I have a course that’s ~70% done, I have a tool for non-techies using personal agents that’s 80% done. I have 2 outstanding funds with one I should finish fundraising for. I want to try a new sponsor model that I should ping companies about.
I should do things because they make business sense. But I only like doing things for the fun of actually doing it. I can only be motivated to do something if it feels like play.
And I think play is missing for a lot of people. We’re pushed by our companies or peers to do more. But more for what reason? It’s busywork. It’s cleaning my desk before starting what would actually be work. AI was meant to handle our work, but when it does it turns out there’s endless work that you can do.
Just one more pull of the slot machine.
You don’t need to try every tool under the sun. You don’t need swarms of agents running and optimising your life. I hate optimising anything.
I recently un-subscribed 60k+ people from this newsletter because they were not opening it. But the numbers are good for sponsors. It’s just not the point of this whole thing. I want to share what I find interesting, what I do for fun. I tinker, I build random shit, I share it. And ultimately I want it to feel like you can do more stuff with AI by seeing what I can do, but not to the extremity of productivity-tokenmaxxers.
The videos I did I found myself trying to pick them apart to optimise them. Am I trying to become a fucking YouTuber? Absolutely not. Don’t like the videos? Guess what? You don’t have to watch them. I’ll make them when it feels right, and I’ll try and make them good.
I’m thinking a lot about all of this stuff (as you can probably tell). I was the guy who shouted about not needing to code when the world told you you did. Now I’m getting lost in multiple agents trying to optimise and automate everything because it feels like I’d be more productive.
In a company you should usually only hire when there is a pain that hiring would solve. So why am I hiring (and managing) multiple agents for things that aren’t that painful?
Most people can only hold 4 things in their heads. That’s maybe a stretch for me 😂 but the point stands. Top engineers can churn out tons of work and multi-task, and those are the companies I want to invest in because that may be what it takes to build a generational $Trillion company . But for people who want to move fast but also considered, there’s got to be another way.
I want to focus on fun. AI absolutely gives me power to do things I never thought possible, but they don’t need to be valuable, they don’t need to run constantly. If I want to do something, of course I’ll use an agent to help me. But I’m not working for my agent, it works for me. It’s a computer. It’s not a person I have to keep busy and keep managing (and checking it’s managing it’s own agents). That sounds horrible.
I don’t have the answers. But for the next 30 days I’m going to try and do a few things:
‘morning pages’ - first thing i dump out my random thoughts on anything, clear my head
only use agents to build things I find interesting and fun
no new tools unless it’s clearly solving a pain for me
no summarising long posts, read it or ignore it
don’t do anything just because ‘it makes money’
I already can’t stop thinking of ways to package up some form of ‘move fast and slow with AI’ into a business. But I’m going to consciously spend much more time thinking about it before throwing it to an agent. I’ll play around with video, written, visually interactive, anything that gives me energy at the time.
I didn’t mean this to be that long. oops.
If this resonated or you’ve got feelings or thoughts on this, please hit reply and tell me - I want to hear it!
p.s. I have to give a huge shoutout to Ashe who DM’ed me after my tweet and we chatted on a phone call (omg whaaat). I’ve admired watching her go head-first into her own vibe and she gave me some great reflections back to consider what I am doing.
My feed
Google has 2 new models:
Gemini 4 Carbon. Much better than Argon in benchmarks, but only available to cyber defenders for now.
Genie 4. A world model that can create 3D scenes and edit them (like changing the weather or adding characters).
Suhail’s new company Particle is training coding models to get to an answer with fewer tokens.
4 personal agents go through a nine-task test. Which of them actually gets things done?
A free, open-source guide to assembling IKEA furniture, with 3D instructions and voice.
How agents influence experts - 86% of the time, the experts go along with an agent’s random choice.
Simon Willison thinks we’ve reached an inflection point for vibe-coded games. They are actually fun to play now.
Readwise Books - New reading and tracking app that can connect with your agents.
Clicky now generates custom visuals on your screen when you ask it about stuff.
Afters
The line about hiring agents for things that aren't painful is the one I needed. I built a few automations just because I could, and the only ones I still use fixed something that annoyed me every single week. The rest I stopped looking at without noticing. My test now is simple: would I notice if it broke?
Comment regarding: "Most people can only hold 4 things in their heads. [...] Top engineers can churn out tons of work and multi-task."
No one belonging to the species homo sapiens can multi-task! Multi-tasking is a dangerous delusion that our minds play on us!
Interesting article on that: https://www.theguardian.com/science/2015/jan/18/modern-world-bad-for-brain-daniel-j-levitin-organized-mind-information-overload
Another fact: homo sapiens can hold between 4 and 7-8 chunks in working memory at one time. No more. This is the max for all types of working memory chunks (auditory, visual, conceptual, + others). The kind of work discussed here (coding and other complex involved knowledge work) consumes all of your working memory at one time. Sure you can listen to music and code simultaneously, but you cannot read an email and code simultaneously (of course, I mean "code yourself" not telling a bot to do so on your behalf).
Multi-tasking does not exist in humans ... it's always sequential switching. Every switch costs energy (calories). That's why you can easily feel completely brain-fried without actually having done anything real besides busily switching back and forth between different tasks/screens/stimuli all day (which we do on average every 46-47 seconds these days).
If there is something we may want to consider optimizing, it should probably either be keeping the number of "discrete (valuable) tasks completed in a day" somewhere between 1 and 4 maybe, maybe even 8 - 10, depending on task complexity and involvement, but certainly not 25 or 30 or more) or minimizing the number of task/context switches in a day.
In this insanely fast moving world, we should pay very careful attention to how and what we pay attention to. If we lose sight of that we are going to lose one of the few truly unique human capabilities.