Get live statistics and analysis of —'s profile on X / Twitter

Unrelated(?) but I moved some apps to my Android to declutter my iPhone and the quality difference was crazy. Same Glovo that was blank on android on Saturday worked v fine on iPhone. Almost like Apple prioritises quality more.

This chat 3 months ago made me realise a glimpse of how important memory would be soon. I deffo don’t have the resources for memory-layer solutions, so this is a product I’m excited about. The premise of my chat with Claude was simply: > if I already use ChatGPT and have my preferences wired in, I shouldn’t need to start afresh when using Claude. I shouldn’t need to bring Claude up to speed manually. > which then means, there’s should be a way for all my LLMs to share the same preferences — I don’t want to restate my tone a million times, or teach Windsurf models my research flow with each new project. And yes, I know that creating a doc and pasting it into every project is an option, but it’s just a stop gap with poor UX. > while I don’t want to start afresh, I should also be able to choose which memory/preference to share at a given time, especially since too much context sometimes confuse LLMs. All of this call for having a plug-and-play “Context Manager.”

Working on this extension UI and all I’ve been asking Magic Patterns is “how can this be better?”. Works like a charm.

Surely can kick him out of their platform. But the weirdest thing about 2 is the fact that Beehiiv’s marketing is basically “hey look at us, we’re better than ConvertKit”. Literally non-stop and people compliment the “boldness” 💀

One of the changes I’m enjoying is this reduced conflict between what I want to do and communicating them to the Engineering team in a language they understand. If you think of your engineers as partly machines, then it becomes easier to not expect them to understand natural language. Or to understand natural language better when paired with sample code/technicalities. It then helps to break the initial barrier that keeps them from prioritising your request(s) since now, they don’t have to first think of what the implementation will look like because you’ve provided a base they can go off of. They’ll simply be building on what you’ve presented vs starting from scratch. Takes your request from being a pipe dream — from an engineer’s POV — to something concrete and achievable because they can process it. Tbf it's not really a new thing because it's much easier for you, as a writer, to quickly understand the kind of tone your company/client wants for their website/blog when there’s a reference. Same way it's easier for a designer to understand expectations with a moodboard. What's new now is people who can't code can bridge this gap in communication with Engineering using LLMs and it also helps take requests from "we want to change XYZ" to something much more thoughtful and specific. You're able to even have architectural discussions with your engineer because you've brainstormed your request with an LLM. Something the engineer was expected to figure out alone, or with a product manager. Obviously, this requires asking LLMs quality questions, so it comes back to how well you even know what you want in the first place — which is something engineers have complained about when handling Marketing’s requests. But the good thing is that LLMs can help here, too. It may just take more back and forth to land on where you’re going. If you're still having communication-related clashes with your Engineering team, then this is something you may want to add to your toolbox.

Unrelated(?) but I moved some apps to my Android to declutter my iPhone and the quality difference was crazy. Same Glovo that was blank on android on Saturday worked v fine on iPhone. Almost like Apple prioritises quality more.

Working on this extension UI and all I’ve been asking Magic Patterns is “how can this be better?”. Works like a charm.

Surely can kick him out of their platform. But the weirdest thing about 2 is the fact that Beehiiv’s marketing is basically “hey look at us, we’re better than ConvertKit”. Literally non-stop and people compliment the “boldness” 💀

This chat 3 months ago made me realise a glimpse of how important memory would be soon. I deffo don’t have the resources for memory-layer solutions, so this is a product I’m excited about. The premise of my chat with Claude was simply: > if I already use ChatGPT and have my preferences wired in, I shouldn’t need to start afresh when using Claude. I shouldn’t need to bring Claude up to speed manually. > which then means, there’s should be a way for all my LLMs to share the same preferences — I don’t want to restate my tone a million times, or teach Windsurf models my research flow with each new project. And yes, I know that creating a doc and pasting it into every project is an option, but it’s just a stop gap with poor UX. > while I don’t want to start afresh, I should also be able to choose which memory/preference to share at a given time, especially since too much context sometimes confuse LLMs. All of this call for having a plug-and-play “Context Manager.”

One of the changes I’m enjoying is this reduced conflict between what I want to do and communicating them to the Engineering team in a language they understand. If you think of your engineers as partly machines, then it becomes easier to not expect them to understand natural language. Or to understand natural language better when paired with sample code/technicalities. It then helps to break the initial barrier that keeps them from prioritising your request(s) since now, they don’t have to first think of what the implementation will look like because you’ve provided a base they can go off of. They’ll simply be building on what you’ve presented vs starting from scratch. Takes your request from being a pipe dream — from an engineer’s POV — to something concrete and achievable because they can process it. Tbf it's not really a new thing because it's much easier for you, as a writer, to quickly understand the kind of tone your company/client wants for their website/blog when there’s a reference. Same way it's easier for a designer to understand expectations with a moodboard. What's new now is people who can't code can bridge this gap in communication with Engineering using LLMs and it also helps take requests from "we want to change XYZ" to something much more thoughtful and specific. You're able to even have architectural discussions with your engineer because you've brainstormed your request with an LLM. Something the engineer was expected to figure out alone, or with a product manager. Obviously, this requires asking LLMs quality questions, so it comes back to how well you even know what you want in the first place — which is something engineers have complained about when handling Marketing’s requests. But the good thing is that LLMs can help here, too. It may just take more back and forth to land on where you’re going. If you're still having communication-related clashes with your Engineering team, then this is something you may want to add to your toolbox.