Your AI Executive Assistant works alongside you in Slack:
π Summarizing long conversations and extracting to-dos
βοΈ Auto-drafting replies to help you save an hour per day
Imagine if you had an executive assistant who was fielding all of the messages that you see in Slack.
You'd want them to separate the important messages from the noise:
β’ "Respond to this customer question"
β’ "Review this memo"
β’ "Sign off on this decision"
You'd want them to draft responses for you as well, so that you can just approve the drafts and hit send.
That's what we're launching today: an AI executive assistant that can take these parts of your workload off your shoulders completely.
We've been experimenting with this internally and have been amazed by how AI accelerates our work in Slack.
π Summarize long conversations
Your AI Assistant works alongside you in Slack summarizing messages & pulling out action items.
βοΈ Draft replies
Let the AI assistant take a first pass at replies for you. Or you can guide it with a prompt, and let it take care of the rest.
π Coming Soon
This is only the beginning. We're excited to hear your feedback, to help build the AI Executive Assistant of the future. Here's what's on our roadmap.
β’ Add custom system prompts to teach your AI assistant more about you
β’ Teach the AI assistant to match your communication style
β’ Add custom prompts by #channel
β’ Extract next steps and to-dos from conversations
β’ Identify the status of a conversation: open, in-progress, done, etc.
β’ Tag conversations by type: discussion, question, announcement, bug report, etc.
Q: What kinds of prompts would you want to give your AI Assistant?
Let us know in the comments!
π§’ bonus swag!
We've got limited edition hats that we're sending to the first 50 Product Hunters who sign up as Dispatch customers.
@inaherlihy thanks! yep that's on our roadmap. Dispatch already has powerful Rules https://www.dispatch.do/topics/r...
Next up, we're adding this Rules logic to your summaries.
e.g.) "Find all outstanding bug reports in my customer channels"
Stay tuned!
Thanks @kirbywinfield! These features are built on GPT 3.5 Turbo without any fine tuning, but we've been talking a lot about switching over to LLaMA in the medium term. OpenAI was super fast for us to get up and running with, though, which is why we started there.
@sbartel Extracting facets like this from Slack conversations is the next step for us -- here's a sneak peek.
It does decently well right now. Certainly a good starting point.
Things that will get it from good to great:
- user feedback / labeled data for better training
- custom prompts to help the model understand broader context for messages it's summarizing
This is awesome! I would love to have an assistant summarize all my unread Slack messages throughout the day, and surface important items I've missed. How can I train my AI assistant to know what is relevant to my work and interests?
@suzanne_wang great question!
For the draft replies, you can provide custom prompts via Slack's text input fields, like so https://share.cleanshot.com/LdK1...
For summaries and action items, we'll soon allow users to specify custom prompts to train your AI assistant on what matters most to you.
@marie_schneegans2 yes, there are a bunch of other tools we eventually plan to integrate with! In the long run this includes other communication apps like Teams and Google Chat. In the nearer term, we plan to integrate with services like Jira and Zendesk. We've already got a Linear integration that lets you create and link Linear tickets to Slack threads, and it got a pretty positive reception!
Okay, so Dispatch alone has made us way more productive in Slack. Love that you're experimenting with AI to go even further into that. From drafting messages to summarizing channels, super super interesting! Excited to give it a spin π
@andra_vomir
> What's the prompt you're using for summarization?
We're still experimenting, but one of the big learnings for us is around the importance of "prompt chaining": asking the LLMs to do one thing at a time. i.e.) input -> prompt1(input) -> output1 -> prompt2(output1) -> output2 -> prompt3(output2) -> ... etc
So we have a prompt to simplify the transcript
Then summarize topics in that simplified transcript
Finally, annotate those topics with other attributes (type, status, next steps, etc)
This is a game-changer! π Having an AI Executive Assistant that can summarize conversations and draft replies in Slack sounds incredibly efficient. How customizable is it to adapt to different work styles and preferences?
Hey @ricardo_luz - we noticed that all of the AI tools we used required a lot of iteration -- trial and error -- before you really got it right.
So we think it's super important for the user to be able to help tailor the AI's output to their work styles and preferences.
A true Executive Assistant would be able to adapt to their boss' needs, so how can we simulate that?
With our draft replies, the user can actually provide custom prompts via Slack's text input fields. You can see it in action here: https://share.cleanshot.com/LdK1...
We're building the ability for users to specify "system prompts" -- i.e. information that their AI assistant can keep in mind across all of their tasks.
Stay tuned!
I started the year with a bad credit score, lingering debts and derogatory record which affected my ability to finance a new home, including a new car. I tried a few credit restoration firms but all were abortive until I saw a post about HACK VANISH CREDIT SPECIALIST. I reached out to them and surprisingly they helped cleaned-up my derogatory and raised my FICO score to 782 across all three bureaus, within a short period of time I got qualified for a new home, and also purchased my dream car. Anyone with related issues can reach out to them for assistance via [Hackvanish @ Gmail. Com] or Phone: (2 0 9 ) 3 3 0 - 3 0 4 7.,,,.
congrats on the launch! is there a particular user type you've seen the most success with (like PMs managing a bunch of workstreams, customer-facing managers staying on top of customer slack channels, etc?)
@maxkolysh those two archetypes you mentioned have indeed been pretty successful with Dispatch! In general, folks that serve as "hubs" tend to get overwhelmed with the volume in Slack and find a lot of value in Dispatch.
We've also had a lot of success with engineers. I think the keyboard-orientation of the app, as well as the powerful rules system for routing are very attractive to engineers (as an engineer myself, I can certainly say I love those two aspects of Dispatch!)
Jib