After 7 days letting AI agents control my email inbox... 📮
I built an AI agent to take over my inbox for my past 7 days, sharing my learnings & how I built it
Get free Hubspot research of 100+ ways business are hacking chatGPT today: offers.hubspot.com/using-chat...
🔗 Links
- Follow me on twitter: / jasonzhou1993
- Join my AI email list: www.ai-jason.com/
- My discord: / discord
- Inbox agent GitHub: github.com/JayZeeDesign/inbox...
- AI agent for schedule meetings: www.crafters.ai/ai-meeting-sc...
- Extract email/reply pair from raw email thread (RelevanceAI): app.relevanceai.com/notebook/...
- Extract FAQ from past emails (RelevanceAI): app.relevanceai.com/notebook/...
⏱️ Timestamps
0:00 Intro
1:20 AI draft everything
3:22 How people are hacking chatGPT today
4:20 Train AI with past emails
12:09 Agent branch behaviour
18:38 AI agent for scheduling
👋🏻 About Me
My name is Jason Zhou, a product designer who shares interesting AI experiments & products. Email me if you need help building AI apps! ask@ai-jason.com
#langchain #autogen #gpt4 #autogpt #ai #artificialintelligence #tutorial #stepbystep #openai #llm #chatgpt #largelanguagemodels #largelanguagemodel #bestaiagent #chatgpt #agentgpt #agent #babyagi
What else should I build for this AI inbox agent? Leave some comments!
If you are interested in the meeting schedule agent - leave your email here and I will send a blog to you: www.crafters.ai/ai-meeting-schedule-agent
Could you do same agent for Outlook? Thanks. Great videos!
@@mantasvarna yes i think you can just change the trigger from gmail to outlook, the actual agent part is the same!
Could we hire you to build this for our account?
Hi Jason, amazing video. The whitepaper is really dumb, i am assuming this was a sponsorship, but your video is really high quality. Would love to see how I can create a decision tree based on slack export of over 20 channels in slack, to help my customer care team to do QA
There's so much garbage on KZhead with people taking already published docs and regurgitating it. Nice to see some original content with real use cases and challenges.
The heart has been added by AI.
This is another Zapier ad, so I’m not sure how well that applies.
I love how your videos are immediately about what's written in the title. No fluff, no redundant stuff, it's great
AI Jason is my _primary_ source of AI coding that is useful and not just copying someone's docs on youtube. We love and admire you man. Thank you.
Your videos are absolutely top-notch! The quality in each one really sets you apart as one of the best content creators in the AI field. I follow your videos religiously :) Keep up the amazing work!
jason making json files... nice
Just popping to say I genuinely appreciate the videos man! You've taught me so much, and it's changing the way I work for the better. Thanks and keep up the amazing work!
Man .. I really thought I could go to bed just a little bit earlier today. But, this blew my mind so hard, as has happened before with this channel, and it's soooo much new, very compressed content, that I'll be up until at least 4am again, just trying to reconstruct and comprehend the full video. You rock!
Goat I love how deep you go into the details and thoughtful process such as knowledge fact faq, which is something embeddings don’t manage so well. Doing the extra step is very important 💪🚀
This is SUPERB content! Love the in-depth projects you are doing. Maybe taking the business use case one step further with an AI that can negotiate for item prices or project payments, that would be cool.
Duude! This is Insane , I love the content , simplicity and function , Tahnk you so much Jason
This man never misses legit puts out the most fire content in the space.
Thanks bro!
This content is an absolute fire. I am definitely going to use this!!
Really Impressive Work! Thank You for sharing…
This is amazing @AIJasonZ. Definetely first step to full assistance automation. Any chance that you will combine some of your video ideas, fx. this one, vision agents, scaling ai automation etc. to create a semi automatic pro assistant?
Love the videos, please do one for scheduling as well 🙏🏽
Wow this is next level!
Amazing video. Thanks for the valuable materials. I have learnt a lot.
excellent example of doing this purely with lookups and prompt context
Looking forward to this content 🥰🥰
man you are legend. legit explanation. target audience is clear. You lift up a LOT of work and make it real for a lot of us. Keep them coming
need to also say that problem with your videos are that you forget what time it is. so interesting you realize it's already bright outside :)
Love your content!
Awesome video Jason, very good insights and top notch knowledge your providing, thank you!! Planning on building such an AI agent for my self to answer my e-mails iam getting, but its hard for me bc i dont have an background in tech. especially the programming part looks hard to duplicate but I will try my best.
Jason is my favorite AI KZheadr now
Thank you bro you are genius ❤
You are truly excelling in producing remarkable content on your KZhead channel. I stumbled upon your channel just recently and I have to say, I thoroughly appreciate and take pleasure in exploring your content. I eagerly anticipate witnessing your future creations. May you have a fantastic week ahead and continue to excel in your endeavors!
you are so damn cool. Thank you so much for your content. Keep it up! 👍👍
Thanks for sharing ❤
Congrats on the sponsorship!
A version of this AI Agent with MemGPT would be very interesting!
You're probably going to want to revoke your Google Search API Key (it was displayed around 17:30). Cheers, thanks for the video!
Great video! I am interested in the AI agent to schedule meetings
That's is amazing , thank you , can you do a video about using some Open source AI model from Hagging face and extracting relevant information based on my preference and get the info as a CSV file.
Thanks!
This is fantastic! Are you planning on improving on this system further or making it into a service?
@AIJasonZ awesome video! What was the total cost of training your email manager (and running cost as well)? Which Zapier plan do you use? How much are you paying for Relevance AI to run this?
@AIJasonZ - this is really cool. one thought occurred to me; have you thought about prompt injection attacks? how would one avoid that in this sort of situation, since emails and web sites for research would be considered untrusted.
Great content as usual! I wonder if you could make a centralized inbox with Instagram dm, Facebook messenger, WhatsApp and email, and then have something like a custom gpt connected into the flow.
Hey Jason, Please make a video for the beginners on how to get started in AI & related programming. How would you do it, if you were starting today from scratch? Appreciate your work. Thanks
"I can't help but anticipate Jason's advice to be: 'Master Python, am I right AI Jason? 😄 Although I completely lack coding skills, I'm captivated by his videos, fascinated at the possibilities. I'm deeply entrenched in branding, business, design and e-commerce. Hoping to integrate these insights into our ecosystem soon. Excited for what the future holds!"
That's impressive! May I know how long did it take for you to build the entire process?
Awesome
That's one of things I really wanted to do for months, except I didn't ever get to it. Now I've got no excuse not to just follow this tutorial ;P
Dude you have 25, 157 unread emails in your dock!! Better this AI working on the weekends to churn through all of them! HAHA
I was waiting for him to say "Jason JSON" lol
I'm sure you're working on it, but I would love a video from you about the new Assistants and GPTs that OpenAI released today!
Fantastic
This is amazing work, I’m going to try to do this for my wife for her business to help her. Wish me luck, I’ll need it 😂 side note: all of this is accelerating so fast videos I’ve saved a month ago are ancient. We are on the hockey stick. This is happening.
You are step by step reverse engineering our brain 😀
Question: Did you stumble on any adversarial content in emails? I’ve seen a couple of proof of concepts with either Gmail or Google docs integration with ChatGPT, whereby the adversarial content in docs/emails/input hijack the prompt and change the instruction (one was demonstrating data exfiltration, the other targeting [deliberately in this case] flawed function calls) We’re there any “interesting” drafts, like trying to send private info, or any unexpected parameter values in function calls?
i'm sure it loves living in your inbox.
Hi amazing vedio but kindly tell me how to automate email AI Agent with zapier
Interesting. Your essentially 1) creating translation tables to a degree. Email Inbound x matching Email Outbound and 2) having it sweep through the emails to extract information already available.
Great video Jason. One question why did you not directly upload your knowledge base to the assistant?
You mean openai assistant api? I found their RAG not good enough so want extra config
What about an implementation using autogen and memGPT?
Signed up. But it says I have to wait to receive?
is there a comparable app out there with the same level of personalisation that can be connected to ones gmail-account and used out of the box?
Is this available for outlook/microsoft 365 as well? or can only be done with gmail?
Would you please give us some info about costs of agents per run?
How can this solution be productized and scaled? Please can you provide your insights.
Hi Jason, Thanks a lot for this important video. I'm a React developer using your Python code with my email inbox automation. When I command "python mbox_to_csv.py" then generate it and I show "past_email_mbox.csv". but when I command "python email_cleaning.py" it does not generate and I can't show any CSV file, it shows an error [ FileNotFoundError: [Errno 2] No such file or directory: 'past_email_final_mboxt.csv' ], Now how can I fix it. Thanks
13:02 After watching the video multiple times, I found a cameo 😁appearance.
Can you please do a video for Visual Studio Co, you've skipped a lot of steps and we're lost
Pls bring the AI agent scheduler
by any chance, have some of you managed to force answers to be picked from a pre defined list? forcing the answer to be exactly one specific and not chatting around it
Hey Jason, great work I love your vids. I tried to use relevanceai and when I'm using the API I always get a timeout on their end. I had to split my "tool" in multiple api to the point that It wasnt a flow it was just one call to openai so I'm not seeing the value in their tools using the api. Have you encountered such problems? Keep up the great work !
Hmm that’s weird; would you mind share the tool you created with me?
sure thing@@AIJasonZ i'll write an email. thanks ! :)
@@hellippshey what was the solution? Thanks
Damn Jason how can I purchase this to use 😭
Unfortunately this is just not possible for my type of job. Answering emails requires often for me to log into Adobe Workfront and cross-reference different projects, or do complicated excel tasks. Wish something like this was applicable to my job. 😢
Really good insights 😮out of curiosity - what is your job?
How much does this AI agent cost to run per day? Per email?
This is quite interesting, however several of my email addresses (with the most interesting content) fall under NDA or trade secrets. I am not sure if or how to extract this knowledge. My personal email has--unsurprisingly--personal emails, which I also don't want to be ingested. I like this solution, but the implications on security and PII are huge and not easily solved for.
Is this a use case that auto gen could make better?
I actually tried auto gen for this, and it does actually improve the agent reasoning; however i found is a bit unnecessary as the cost goes up with multi agent systems
Hi, please help, I can't find the code from 6:14 minute.
Hey man, I have no coding knowledge at all, do you think I would be able to replicate this with the scripts you have provided and this video? Either way thanks for this video, really opening my eyes everyday into what is possible out there.
thanks! yes i think you can, just need to update the file names to yours; use chatgpt to help you fix the code!
Hey Jason, I've modified parts of your code that use GPT to use local LLMs (running mine through mixtral 8x7b) Let me know if youd like me to send over the code or if it'd be useful to anyone else!
Hey Sludgy!! Would love access to the code using Mixtral, it's something I wanted to implement myself!
I am interested in AI agent for scheduling meetings.
Thanks a lot for these cool insights and examples! One question: when you described creating a knowledge base out of your previous emails, you vectorise the original message (around kzhead.info/sun/fdqYlZqKq2asin0/bejne.html). Why not vectorise the original message + the answer? Is it possible that new incoming messages could match (by vector) better with the response [part], not the question ("original message")? I build one RAG setup to automate a help desk, and vectorised both the question + answer. I haven't tried vectorising only the question, curious which approach may work better. Thanks again for your work!
I wondered this aswell. Thought I must have been missing something.
He answers that in pinned post here kzhead.info/sun/lsOnc86kinRnjaM/bejne.htmlsi=vvdYNALTGTfyuVP7 Summary: vector is for searching so only vectorize the questions. Answers are still used to craft responses. Keeps cost down and more effective.
Why not integrating memgpt into the process?
cuz i dont need unlimited memory for this, email threads normally are fairly short
Can this be done on Microsoft outlook?
Yes, absolutely you can! Just use the outlook API, or zapier
😮🎉
After cleaning up the CSV file, i have 3000 sent emails to go through. Is that too big of a database? I think the email cleaning .py is going to be costly
Okay Jin Yang....
Yeah I just made a keylogger so I can get my own data to see what I can extract facts about myself ^_^
I wonder how I could extract my FB messages because its not turn based like an email is. Friend: Hey! Friend: Whats up! This is msg #2 without reply. Me: Hey! Me: This is not like a gameboy pokemon battle where we... Me: Get turns, its just rapid fire Me: You know.
Do you operate in Romanian language?
Learning and usefulness aside I see a big data protection issue.
99% of emails I get are notifications of some service (shipping, logins, privacy policies), hardly any humans actually send emails… to me at least
I think u leaked your api key at the function search(query)
thanks dude 😂
The clear advantage of doing this as someone with hard skills, is that we nerds simply don’t get many emails 😅 8.57 mails a day is half an hour for others 🥶
😂
Wow, this could be used in a really bad way. Advanced phishing
most my email is spam anyways.
Is AI Jason even real? Are we sure he is not using a code to deepfake his face and voice to automatically create an AI video?
Oh my Allah 😮
If people know that most of the emails are generated by AI, why do they still want to read them? Perhaps you need another AI to assist you in reading them efficiently!
Short answer, nothing happens
Lol, how’s that legal? Sendung private emails to OpenAI without consent.
-> I believe we are meant to be like Jesus in our hearts and not in our flesh. But be careful of AI, for it is just our flesh and that is it. It knows only things of the flesh (our fleshly desires) and cannot comprehend things of the spirit such as peace of heart (which comes from obeying God's Word). Whereas we are a spirit and we have a soul but live in the body (in the flesh). When you go to bed it is your flesh that sleeps but your spirit never sleeps (otherwise you have died physically) that is why you have dreams. More so, true love that endures and last is a thing of the heart (when I say 'heart', I mean 'spirit'). But fake love, pretentious love, love with expectations, love for classic reasons, love for material reasons and love for selfish reasons that is a thing of our flesh. In the beginning God said let us make man in our own image, according to our likeness. Take note, God is Spirit and God is Love. As Love He is the source of it. We also know that God is Omnipotent, for He creates out of nothing and He has no beginning and has no end. That means, our love is but a shadow of God's Love. True love looks around to see who is in need of your help, your smile, your possessions, your money, your strength, your quality time. Love forgives and forgets. Love wants for others what it wants for itself. Take note, love works in conjunction with other spiritual forces such as faith and patience. We should let the Word of God be the standard of our lives not AI. If not, God will let us face AI on our own and it will cast the truth down to the ground, it will be the cause of so much destruction like never seen before, it will deceive many and take many captive in order to enslave them into worshipping it and abiding in lawlessness. We can only destroy ourselves but with God all things are possible. God knows us better because He is our Creater and He knows our beginning and our end. Our prove text is taken from the book of John 5:31-44, 2 Thessalonians 2:1-12, Daniel 7-9, Revelation 13-15, Matthew 24-25 and Luke 21. Let us watch and pray... God bless you as you share this message to others.
Nicely done... BUT this is still the most stupid thing ever. First of all, it will contaminate itself after a while by training on mails it sent itself. You should exclude those from training but then it'll be stuck in the past unless you still write mails yourself and train it exclusively on those. Naturally this will bias training more and more to exceptional circumstances (the ones in which you decide for whatever reason to write yourself) and after a while all its mails will sound like dramatised apologies or non-consequential chit-chat or some mix between them but it will no longer sound like the normal "you". We already see that effect all over GenAI - especially visible in pictures which train more and more on things GenAI created rather than actual fotos of real things. As long as AI outputs aren't clearly marked as AI generated (and, proudly, they all try to avoid just that for obvious reasons), this will render GenAI exponentially more useless over time. Secondly - sending an email personally addressed to you off to a cloud based AI system is not just reckless but probably illegal. Thirdly - if you get more mail (communication between humans... still) is more than you can handle because it is oh so tedious and time consuming, you should probably work on your life in more constructive ways than automating your part of it. Lastly - if you really think this is progress and, in the long run, makes you more productive, think again. What will happen is that the "people" you are "communicating" with are using the same "solution"? You'll bombard each other with a rapidly growing mountain of meaningless fluff. Man - go out and touch some grass and do something really useful with your skill.