Yeah we totally agree. That's why we work on the end-to-end app, not on a single prompt. You pick what parameters become knobs in the frontend. So if you have a giant app with 10 parameters (say 5 prompts, 5 numbers), great, wrap those and they become knobs on our frontend. We override during the actual end-to-end testing execution kinda like a Statsig / Launchdarkly (only with typing).
Awesome work Cai and Andy. Do you have advice on structuring functions for best results? (eg, is it a good idea to give my agent access to things like a calculator "in case it needs it" or be careful about spoon-feeding it the tools I think it needs)
I don't think it's a good idea to give an agent tools "just in case."
We've opted for giving our agents access to a few distinct, but powerful tools, and then trusting them to combine these tools in a strategic way. It's going well so far, but we are pretty careful to clearly explain the tools.
However, I've spoken to other teams that have opted for giving their agents tons of tools, and it sounds like that can work pretty well. But everything seems to take work and experimentation.
Just tried it out in my dummy application, but it throws an error when you provide a suffix... which is officially supported by their documentation. Hoping they support this soon.
Too bad. Using suffixes means that maybe it can play Cards Against Humanity, if there is a way to constrain its output among a fixed set of choices. I can't wait to see the results.
I understand the cartoony feedback. We're working on improving the quality of the avatars to be more adult / work focused. We do believe in avatars - they are one of the key points in our product - because we believe that being on video all day is soul crushing, but audio-only feels disconnected. The avatar provides privacy and personality at the same time. FWIW, you can turn on video if you want it too, we just default to the avatar.
Nice, that’s a great angle. I think you should stick with it. Thanks for responding and providing more context.
I might be a bad prospect because I’m a very literal and surface-level sensing type of person. I’m terrible at reading between lines, seeing past facades, etc. Working with avatars would probably make me feel like I’m in a cartoon, haha. This is definitely not your fault.
I know my oldest kids would love this. They don’t like the formality of things like zoom and they find mostly-text interfaces like slack stifling. By the time they’re in the workforce, if something like this was around they would likely prefer it.
Good luck with this. I’ll keep an eye on it for sure.
Believe me being an avatar all day is more soul crushing. Actually I cannot see how being face to face can be referred to as soul crushing. That’s a pretty said commentary on the target audience.
In your opinion. For me, I’ll always favor video-off calls, especially when they are longer than 30 mins.
I find being on video calls all day incredibly draining. I can see how a tool like this would help. Though for me the addition of an avatar adds nothing, but it is a nice middle ground.
