Kevin HG
@kevinhg.com
There are few better than a good podcast and a new app (maybe outside and with food)
I can’t overemphasize how empowering it is to be able to take what’s in my head and create it in code 🔋🔌🌪️
I’m so hyped to share some of what I’ve built. I’ll post some slick gui here after a couple more late nights 🤌💎
I’m so hyped to share some of what I’ve built. I’ll post some slick gui here after a couple more late nights 🤌💎
December 15, 2025 at 11:27 PM
I can’t overemphasize how empowering it is to be able to take what’s in my head and create it in code 🔋🔌🌪️
I’m so hyped to share some of what I’ve built. I’ll post some slick gui here after a couple more late nights 🤌💎
I’m so hyped to share some of what I’ve built. I’ll post some slick gui here after a couple more late nights 🤌💎
Getting my projects jussttt right still takes time and effort (I easily clocked 60 multi-agent hours this week), but before 4.5, I couldn’t go at the same project for more than 6 hours before it would collapse under the weight of accumulated tech debt and detritus.
December 15, 2025 at 11:27 PM
Getting my projects jussttt right still takes time and effort (I easily clocked 60 multi-agent hours this week), but before 4.5, I couldn’t go at the same project for more than 6 hours before it would collapse under the weight of accumulated tech debt and detritus.
There were always going to be trade-offs as Anthropic tripled down on coding, and after a week of building at a level and pace I never have before, I’ll happily take this model that feels 40% worse at research, but 40% better at coding.
December 15, 2025 at 11:27 PM
There were always going to be trade-offs as Anthropic tripled down on coding, and after a week of building at a level and pace I never have before, I’ll happily take this model that feels 40% worse at research, but 40% better at coding.
That said: this model finally got me to upgrade to Max 💸👋
As much as I loved the thoughtful, well-researched recs from Opus 4.1, the context and usage caps made it tough to lean on for research anyway. It was a testament to the quality of Claude that it was still my go-to.
As much as I loved the thoughtful, well-researched recs from Opus 4.1, the context and usage caps made it tough to lean on for research anyway. It was a testament to the quality of Claude that it was still my go-to.
December 15, 2025 at 11:27 PM
That said: this model finally got me to upgrade to Max 💸👋
As much as I loved the thoughtful, well-researched recs from Opus 4.1, the context and usage caps made it tough to lean on for research anyway. It was a testament to the quality of Claude that it was still my go-to.
As much as I loved the thoughtful, well-researched recs from Opus 4.1, the context and usage caps made it tough to lean on for research anyway. It was a testament to the quality of Claude that it was still my go-to.
There are workarounds—deep research, planning mode, etc—that get solid results, but 4.5 feels optimized for token efficiency and brevity when not coding, and so it requires more back-and-forth to get to root of qs. Anddd I’m not used to this many hallucinations from Opus.
December 15, 2025 at 11:27 PM
There are workarounds—deep research, planning mode, etc—that get solid results, but 4.5 feels optimized for token efficiency and brevity when not coding, and so it requires more back-and-forth to get to root of qs. Anddd I’m not used to this many hallucinations from Opus.
Opus 4.5 doesn’t seem to read-between-the-lines, dig as deep, search as widely, or validate as thoroughly without extra prompting. It kinda reminds me of the transition from GPT o3 to 5.
December 15, 2025 at 11:27 PM
Opus 4.5 doesn’t seem to read-between-the-lines, dig as deep, search as widely, or validate as thoroughly without extra prompting. It kinda reminds me of the transition from GPT o3 to 5.
Less good @ research 🔦🫠
Off-the-bat I felt that the thoughtfulness and persistent effort that Opus brought to coding seemed absent when answering questions that required web search.
Off-the-bat I felt that the thoughtfulness and persistent effort that Opus brought to coding seemed absent when answering questions that required web search.
December 15, 2025 at 11:27 PM
Less good @ research 🔦🫠
Off-the-bat I felt that the thoughtfulness and persistent effort that Opus brought to coding seemed absent when answering questions that required web search.
Off-the-bat I felt that the thoughtfulness and persistent effort that Opus brought to coding seemed absent when answering questions that required web search.
And yes, Claude a functioning build to work off of, and tech debt still accumulates, but from that project to 6 others: I’ve never been able to create in code like this.
December 15, 2025 at 11:27 PM
And yes, Claude a functioning build to work off of, and tech debt still accumulates, but from that project to 6 others: I’ve never been able to create in code like this.
It works deliberately, looks around corners, asks smart questions, and just keeps at it till the job is done. I (basically) one-shotted a mobile friendly version of my research orchestration n8n that took me 20+ revs before.
December 15, 2025 at 11:27 PM
It works deliberately, looks around corners, asks smart questions, and just keeps at it till the job is done. I (basically) one-shotted a mobile friendly version of my research orchestration n8n that took me 20+ revs before.
Damn good @ Coding 👩💻🤤
I know it’s Claude, so this isn’t a wild take, but the coding—as all my poor friends have heard me say 10k times this month—is crazy. Gemini 3 and I had a good few days (the images are still 💅), but Opus 4.5 just works. It just works.
I know it’s Claude, so this isn’t a wild take, but the coding—as all my poor friends have heard me say 10k times this month—is crazy. Gemini 3 and I had a good few days (the images are still 💅), but Opus 4.5 just works. It just works.
December 15, 2025 at 11:27 PM
Damn good @ Coding 👩💻🤤
I know it’s Claude, so this isn’t a wild take, but the coding—as all my poor friends have heard me say 10k times this month—is crazy. Gemini 3 and I had a good few days (the images are still 💅), but Opus 4.5 just works. It just works.
I know it’s Claude, so this isn’t a wild take, but the coding—as all my poor friends have heard me say 10k times this month—is crazy. Gemini 3 and I had a good few days (the images are still 💅), but Opus 4.5 just works. It just works.