AI, tech, business, and solo game dev. PhD in EE. 20 launched games in 12 years. Christian, father of 4.
North Carolina, USA
Joined October 2026
- Tweets308
- Following94
- Followers76
- Likes219
Pinned Tweet
We are in an era of automation: automate this, automate that, automate the next thing.
The bottleneck is not our ability to automate. It's the mental fatigue that comes from continually having to think in higher levels of abstraction.
2
3
121
Context still matters so much.
I was having AI optimize an app, trying to get the startup time down to reasonable levels. I tried three or four times, telling it to investigate, to explore, and find things to speed up, and it found only marginal improvements every time.
Then I asked it, "Tell me what the data structures are that we're using and what kind of programming patterns are being used during initialization." It came back with a bullet point list of six things: six patterns and data structures that were being used, why they were a bad fit and were slowing us down, and what good alternatives would be. From there, able to get it optimized, no problem.
The act of telling it to look at specific data structures forced things into the context which completely changed the outcome.
1
20
I run text through a cheap model to clean it up before text-to-speech in my apps. No more reading tool calls and data tables aloud!
Every text-to-voice button should do this. @bcherny
1
19
Computer Use is all you need.
Effective computer use is the biggest technological advancement of the last several months. My agents now:
-Navigate complex and buggy websites (cough facebook ads manager cough)
-Read threads and conversations on discord
-Setup billing information for subscriptions
-Generally operate my phone, laptop, or desktop remotely, including configuring settings and installations
Computers are increasingly a thing to be used only be the Computers.
1
37
One great feature of Mantis is being able to set up scheduled agent tasks, and I can say which model I want to run the task, what machine is going to be running it, what project it's in, and what it's doing.
1
19
Today, I built a full replacement for the ChatGPT and Claude Code mobile apps. I call it Mantis, and it's the command center for everything I need to run my business.
I'll probably be showcasing some of the features I built over the next couple of days.
With my phone asleep, I can say "Mantis," and it wakes up the phone, opens the app, starts a new thread, and starts speech-to-text dictation. I give my message, and then I say the keyword "Osprey," which ends the dictation and starts the turn.
First thing the Mantis app does from here is spawn a lightweight model like Tera to infer which project the prompt belongs to and what model should be used to run it. Then it kicks off the agent in the right context!
1
22
The lack of steering in the Codex mobile app and the lack of queuing in Claude finally pushed me to create my own model-agnostic agent harness. I'm calling it Mantis.
1
42
Tonight's task: hacking my Fiet Electric light bulbs to turn off the stupid power outage reset behavior. Go, Astra, go! 😁
1
59
A huge thing is just being aware of what is possible. I told Codex what I wanted to do, and it comes back and tells me, "Oh, I can't really do that." I go back and I'm like, "Are you sure we can't just flash the firmware?" It's like, "Oh yeah, you're right, we could do that. I found an open-source package to do it, but we need Linux."
I'm like, "Are you sure we need Linux? Can't we just port it to the current operating system? Can't we just port it to Windows or Mac?" It's like, "Oh yes, we could do that, except..." and so on and so on.
If you know the right places to push back and can propose reasonable routes to follow, you can get so much more leverage out of the AI.
30
"Just ask AI" is the next "just Google it."
Despite Google being in the AI business, I don't think this is a net positive for Google.
1
1
44
Matthew Miskiewicz retweeted
I’m sure you have had animated conversations with people you “clicked” with, where everything flowed naturally and you didn’t even realize the passage of time. That kind of communication creates truly new ideas, makes deals happen, creates friends and lovers, and makes you feel alive. AI doesn’t do that kind of thing. It’s uniquely human.
Also, if you read biographies of famous founders or other top performers in any field, there’s always a chapter about something they experienced in their childhood or in business or wherever that lead them to a major conclusion or decision that changed the world. Often they also meet one or several other unique individuals in their life, and the combination of them leads to something huge (Steve Jobs & Woz, the Wright Brothers, all the members of Led Zeppelin, The Rolling Stones, and the Beatles, etc). It’s impossible to create any of that in a vacuum.
Human beings are the most complex things in our known universe, by a long shot. All this AI technology is super impressive and I love it, but it’s nothing compared to a human brain, even though AI outperforms it in many dimensions.
AI is incredible at doing things bound by rules. Mathematics, programming, making a plan, organizing things, etc.
Running a business also involves rules, but most of it is an intense dance with chaos. It means dealing with people, and nothing is more unpredictable and complex than that. It also means doing some bartering, making a deal with someone over dinner, and relying on goodwill and connections. It also means not making a deal because someone rubbed you the wrong way because of how they treated the waitstaff.
AI can’t do any of that. Running a successful is too much of a human endeavor.
That’s the way I see it, based on my personal experience. Every successful entrepreneur I know is either a people person, or someone with an idea working closely together with a people person. I’m sure others see it differently and I’m sure there are exceptions.