Jaydan Urwin

If you build it, you still have to tell someone

[Jaydan Urwin @jaydanurwin]

@roerohan @OpenAI @opencode Personally, I haven't had a great experience with sub agents every time I've tried them. I find it better to create multiple threads with specified tasks that I determine rather than the model, deciding what the tasks should be. Still exploring.

[Jaydan Urwin @jaydanurwin]

Grok being my default coding model wasn't on my bingo card for 2026 but here we are. This 4.5 model is phenomenal. Fast, persistent, smart, and cheap

[Jaydan Urwin @jaydanurwin]

Grok 4.5 came out of nowhere though. This model is such a workhorse.

[Jaydan Urwin @jaydanurwin]

GPT 5.6 is amazing, Fable is amazing, Grok 4.5 is amazing, Gemini 3.5 Flash is amazing. There is no shortage of great models to build with.

What a time to be alive right now. Go build something!

[Jaydan Urwin @jaydanurwin]

@thdxr Ya I'm coming to realize this is a place where people prefer complaining over praising. I think the timeline skews heavily towards negativity compared to YouTube or potentially Instagram even.

[Jaydan Urwin @jaydanurwin]

Still very early on the marketing side but I built this and now use it every day for both coding and general assistant use.

100% free for a single desktop device. Completely private, on device. Bring your own subscriptions and API keys. Mobile client apps incoming.

https://t.co/2pShPA1CV6

[Jaydan Urwin @jaydanurwin]

and now...we wait

[Jaydan Urwin @jaydanurwin]

Fable is on a different level. Based on how the last year has gone this makes me so excited for every other model to get close at a much lower cost.

[Jaydan Urwin @jaydanurwin]

Your ability to work successfully with agents is almost completely reliant on how well you can communicate both in your direct prompt and the context it already has.

[Jaydan Urwin @jaydanurwin]

@markproduct You are absolutely wrong

[Jaydan Urwin @jaydanurwin]

@nayli_ai Built my own agent / ADE that's model agnostic. Chat with local or cloud models

https://t.co/d7ocM7ANul

[Jaydan Urwin @jaydanurwin]

@jpschroeder I think the solution is to ship a skill that wraps the mcp server. Keeps the context from getting blown up because it's self contained.

[Jaydan Urwin @jaydanurwin]

@DanielSmidstrup I think branding and speed. If you can't catch up with them how can you copy them?

[Jaydan Urwin @jaydanurwin]

@mweinbach I've heard Grok is king for trading because of the X grounding search. I still need to try it for myself.

[Jaydan Urwin @jaydanurwin]

@MatthewBerman Do you use feature flags? It's not a silver bullet but it can allow multiple agents to work on the same feature together.

[Jaydan Urwin @jaydanurwin]

@HarshithLucky3 Gemini 3.5 Flash is my default coding model for anything frontend related at all.

[Jaydan Urwin @jaydanurwin]

Start sweating the details. It will pay off in the end.

[Jaydan Urwin @jaydanurwin]

@fofrAI This is why my default test prompt is "tell me a joke"

[Jaydan Urwin @jaydanurwin]

haha I'm pretty sure this is already broken...it was worth a shot

[Jaydan Urwin @jaydanurwin]

Thanks to this I now have Composer 2.5 Fast in my pocket

[Jaydan Urwin @jaydanurwin]

@gregisenberg What about local models AND cloud models...?