DadBodGeoff

When I lost my job at @HelixeSportsFX I thought I’d never find a job that I loved as much Over the last 8 months I became obsessed with building with AI. Now I’m working full time on an engineering team developing an AI platform. Work harder, nobody cares.

I wonder if this plays into any of the model degradation we see over time. Does the j space develop certain tendencies and new alignments over time?

One prompt that has improved my workflow: “Spawn an independent senior engineer subagent that continuously audits your work while you execute the plan. Have it review completed tasks for: bugs flawed assumptions deviations from the plan missed edge cases

Resetting everyone’s usage the day 95% of the userbase already resets to celebrate 5 million is a massive L. Could wait til Monday or Tuesday. @thsottiaux

Recently lost my best friend of 12 years. When I moved out at 18 I got him, he taught me how to be a dad, he kept me alive and was the best dog ever. I know you’re not suffering anymore but I miss you so much old man.

opus 4.7 in the Claude desktop app using Claude Code is unusable. It’s truly sad. Opus changed my life and career with the release of 4.5 and from there it just went…backwards.

Asked Codex to generate an image of what it thinks I look like based off my memory + chronicles here's the results Imagine having all those monitors, we cooking on my macbook 13"

My thoughts around GPT 5.5 I believe this is “Baby spud” likely several RSI checkpoints behind what internal OpenAI employees are using and will be the first model we feel gets better over time

What’s the first request you’re giving Opus 4.7 or Spud?

Claude Code and Codex usage over the last 40 days. Claude code is now exclusively planning and Front end work for me. Anyone got me beat?

Hard work, determination and self belief is all you need to chase even your wildest dreams. In 2025 I hated the path I was on. Didn’t wanna go back to restaurants and spent 9 months straight for 12+ hours teaching myself to get where I wanted to be Work harder, nobody cares 🫡

How bad do you want it?

If your building infrastructure around autonomous agents I want to connect with you. Reach out!

AGI is here but we can’t be trusted so it’s not for everyone. That’s cool.

Claude’s had a massive influx of acquired daily users but haven’t bought the compute to support it so instead of supporting the ones who’s supported them they blame users for the limits seeming different all of a sudden and tell us to user sonnet. A+ customer service

1 hr 45 min commute turned into 3 hours this morning. How do we not know how to drive in the rain?!

Watching NASA’s coverage of Artemis 2 has me feeling like it’s 1972 again. After getting use to the standard SpaceX has set this is a snooze fest!

starting to think this is the layer i want to build something that sits between the agent and real world actions and handles: when to ask when to snapshot when to let it run and how to make rollback / review stupid simple so autonomy doesn’t feel nerfed

think a lot of agent safety is being built wrong too much asking too much interruption kills autonomy better model: if it’s truly dangerous, ask if it’s recoverable: snapshot it let the agent keep going show me a timeline after let me revert if needed