Alex Omeyer (ClickUp) Basically, yes! And obviously - The main goal is simple, faster conversations with AI for brainstorming purposes.

I don’t actually care what tools are connected, only that it takes forever.

Think using GPT-4o in ChatGPT vs. using Gemini 2.5 Pro in Gemini.

ChatGPT is way better for creative brainstorming because it’s so much faster.

Ben Valentin: Gooot it, thanks a lot!

Alex Omeyer (ClickUp) Conversely, the move to make it faster has resulted in MUCH less useful output from ClickUp Brain when asking about your Workspace.

Answers are nowhere near as good as they were in June/August 2024. They are often incorrect, never in-depth, and are missing information from relevant Tasks and Docs. They don't source that many Tasks or Docs anymore, either.

So this is somewhat unrelated to OP, but kind of related in a roundabout way, in that I don't want it to get even worse. - Can we please get ClickUp Brain to give us in-depth, accurate answers about information in our Workspaces, that does not miss out relevant Tasks and Docs? Like it was in August 2024, basically!

It's the entire reason I signed up for ClickUp AI, but I don't use ClickUp Brain anymore to ask about my Workspace because it has gotten so bad at pulling information. 🙏🙏

Alex Omeyer (ClickUp) I would take a "Search in-depth" button (default on) and slower results if it gives accurate, in-depth information that covers all relevant Tasks.

That is FAR preferable to something fast that is inaccurate or misses data.

Guy Mannerings: I'm with you! 2 big things coming: we just made an improvement to one of our AI search tools that almost doubled quality.

We're also preparing our "deep workspace search" POC for prod release. This search will be slower but a lot more thorough.

I'm hoping that between these two, search will be even better than last year!

Alex Omeyer (ClickUp) Great! Thanks for that.

Will it also add references back in? Preferably in-line where specific or at the end of its general. So it can mix depending on what is best.

Also, will it actually search ALL areas of the workspace and give back ALL relevant info? One thing with the old version was it seemed to be limited to 4 references.

Alex Omeyer (ClickUp) Will these be obvious releases, by the way? Or do I need to keep an eye out for changes?

Guy Mannerings: yup, we're designing a piece of UI which will clearly show Brain's thinking, the sources it considered, and list sources inline. And the idea with Deep Search is that it should really consider all relevant info.

Note: we aim to release the first iteration of Deep Search without this new UI but we'll ship it soon after :)

Re announcing release: it should be obvious, yes! That first iteration will add Deep Search as an option on the Ask AI homepage.

Alex Omeyer (ClickUp) Sounds great! Thank you for addressing that!

Alex Omeyer (ClickUp) Oh, one other thing. With the "copy" button, it would be great to get the option to copy WITH and WITHOUT the references. So we can see the references, but then copy the text output only and no reference links, or we can copy it all with reference links.

Also, when copying output, it should not copy the part at the beginning where it says "I have done this for you....." and at the end where it says "let me know if you need anything else" (or similar).

Guy Mannerings: Understood, thanks for explaining.

I've gone back and forth on including vs excluding the AI pre-amble in responses. It's been hard to find a way to make this work in every situation (i.e. how do you consistently figure out what's pre-amble vs what's not).

Alex Omeyer (ClickUp) I would almost rather it never even said an pre-amble and post-amble (as that a word??).

Is there ever actually a good use for it? Like... I already know it did XYZ because I can see it.

I already know I can ask follow-up questions, because that is a basic use-case.

So maybe it should just never output them in the first place, and should just do the thing you asked?

This would also give the benefit of taking less time to output and be quicker to read.

The only times it should say something outside of what you asked is if it failed/didn't understand/had a problem/needs clarification/can't find something you asked, etc.

Do you see any problems with this approach?

Guy Mannerings: Yeah it's an interesting question. We were just talking with the team about how Brain should generally be more concise (and defo limit pre-amble) when responding in the AI Command Bar UI since it's already quite small.

In general, we want to limit any useless fluff while also giving users the info they need to know what the AI just did / missed.

E.g. working with Claude Code, I like it to give me a summary of what it did so that I can say "you forgot X, overdid Y, and also now you mention it can you do Z".

Alex Omeyer (ClickUp) Yes, I think summaries in that instance make sense. Like they do for if it did a thing on a task.

E.g., Rather than simply saying "Done!", it is better when it says something like, "I marked this Task as XYZ Status, added ABC info to this Custom Field, changed this Dropdown to 123", etc. That is definitely needed.

One thing I note on this: It's so inconsistent with what it says or even does. I wish it was consistent. I can ask it the exact same thing - or at least very close - and get two different outputs. Like one where it lists everything, and another where it doesn't.

This is a problem with LLMs in general, and why any way to build in actual hard references (like @mentioning a custom field) is key - because LLMs always react differently, and you never know if they will be consistent. This is even more true when they push updates to them in the background and they change yet again.

Does that make sense?

Guy Mannerings: Yeah totally -- and the underlying LLMs change so often too!

We're working on improvements with how Brain handles context depending on the UX through which you invoke it. It's a complex issue to summarise here, but rest assured we're diving deep into it.

Thanks again for your feedback!