Severe performance issue with AUTODESK ASSISTANT

[ca_steve_foobar]

I have been using the new Autodesk Assistant extensively over the past week and for very long, sometimes, multi-hour sessions. I have consistently noticed that after about a one hour session, the response time with the assistant gets extremely slow. At some point, it gets so slow that my typed characters to prompt it don't even display for sometimes as much as 5 seconds! The only way to reset things to normal seems to be to start a new chat session. Unfortunately, if a new chat session is started, no context or history from the previous chat session is available for use. The Assistant has no persistent memory or even the ability to review the historical chat sessions, apparently.

[lauri_barnhart]

Hello, @ca_steve_foobar.

I'm just following up to see if the Autodesk Assistance performance has improved or if you are still experiencing the lag after about an hour of use. Please let us know with an update here how things are going.

Thanks,

[ca_steve_foobar]

Unfortunately it has not. I have determined that it is not elapsed time related but instead related to the number of total characters/words in the long chat session. For example, today I purposely kept one long chat session while I was using Assistant and started noticing extreme input prompt key lag as I typed to the point where the session was unusable. I had to start a new session and as you know, that means it loses all history from the previous session. I copied all the text into Microsoft Word and got a Word count. It is attached. I hope this helps. 

[Phil.E]

The reason this happens is the context becomes filled. Would you use a feature that compresses the conversation by keeping just the most important parts? Alternately, what if new chats could access the memory of older chats, would that help. Thanks for your time and consideration of these questions.

[ca_steve_foobar]

Firstly, I don't know what AI back-end and API is being used or if you have any control over it. When I use third-party AI bots, they NEVER slow down my input typing regardless of how long or how many hours the session was open. This would be the ideal behavior. If that's not possible, shortening the saved memory to just the critical information, would likely help with the speed issue, again depending on what AI you are using. Regarding, current chat session access to previous chats would be amazing, however, not a single AI bot I use from the major vendors has that capability, so I doubt you would be able to do that. I also assume that it would dramatically worsen the typing input (and response) lag issue.

What I have noticed is that the Assistant is wrong or goes down wrong solution paths at least 70% of the time! The only exception is if I ask it something very simple and direct like "How does the ..... command work?". At times it gets things so wrong, that I have to close it and figure things out myself. I have also gotten into the habit of closing it and starting a new session for every simple prompt because I don't want my input typing to start to lag.

So having said that, I have connect Claude to the MCP API and have started experimenting using Clause instead of Assistant. I don't have an opinion yet on that.