DevHub Live 31 - Building AI Agents & Bitte Bounties

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:29

   what you want want oh

2. 1:30

   hey hey he

3. 2:59

   oh you n

4. 4:00

   n hey welcome to another edition of Dev Hub live I am your host Joe uh thanks again for joining us today uh we have a

5. 5:02

   great episode uh I am by myself today because we are action packed here with multiple developers talking all about AI agents uh this is born out of a few different things so you've been following along we have had bit protocol on a few times to do actual uh like workshops and some tutorials especially around the redacted hackathon on things like that we got to a point now where you know bit day has really started to

6. 5:33

   grow this out they have a full registry now you can see analytics and they've actually been doing a bounty program for a little while outside of hackathons to allow Builders to uh contribute different agents that are beneficial to the entire ecosystem and just in general things that are like necessary or that you maybe would want as part of an AI agent system and today uh we have a

7. 5:58

   really awesome episode where two of those Bounty winners are going to come on and talk about what they built for bit day we also have microchip from bit day to talk about building agents and on top of all of that we're announcing that uh at devhub we matched those bounties to these Developers for the work that they did um so we've done a bounty matching program in the past and we felt

8. 6:22

   that it was important to recognize developers so these are all really exciting things that I think are a big part of what we're trying to get across here on near and just again what bit is doing with all their AI agents so with no more delay uh let me introduce everybody and let's get started with uh talking about AI agents all right thanks everyone thanks for joining me today we have here uh slime we have

9. 6:51

   this Winder and we have Micro uh and guys maybe we could just start with talking a little bit about what you guys actually built uh and what that look like for uh the Bounty program Jaz you want to go first sure yeah I can go first quickly um so I built a dow agent on top of uh bet protocol um the Dow agent lets you interact with your sping V2 contract at the moment so you want to query your

10. 7:23

    Dows you want to create proposals you want to vote on proposals you want to transfer funds near or fungible tokens you can do all of that stuff uh there's a bunch of new features coming soon as well so yeah all the things that you want to do with this spting V2 contracts uh already part of the Dow agent and some new features are coming along awesome awesome and uh slime what did you build hi I built a staking agent it can

11. 7:52

    do staking I staking and that's all nice that's all you need from a staking agent yeah no awesome great great stuff and micro you want to talk a little bit about kind of uh what you guys are doing to enable them to build agents like this yeah sure um so I mean I think we're when um so we're building this framework right this agent framework uh that you know it's pretty simple uh you pretty much any API can become an agent and that's kind of like what the what these folks uh did were like they built

12. 8:31

    an API attached a file like a single manifest file that kind of explains what the API does and then um our runtime when ingests this this data um makes these API smart right so that's kind of like what we're building and um you know it's very nice to see like enabling uh developers to do this because they don't need to like create open API keys or or not sorry open AI keys they don't need

13. 8:58

    like any of this stuff uh because just plugging in into the bits of runtime uh you know we are the ones paying upfront the cost of the AI uh inference so um I think it's just a a very good enabler for developers in general that don't want to like spend a lot of money to experiment with these uh these tools yeah one of the things that you guys have is so there's two parts of it one is obviously your like main registry

14. 9:23

    but then you also have a playground where anyone can just you know from their local setup actually start building and trying this out right yes exactly like uh you can run these things locally uh connect directly with the runtime and uh works the same way as it would work um live right yep perfect so Jazz let's start with you and and what you did for Dows now this is something that uh has obviously been an issue across all ecosystems is just Dow Tooling in general why did an AI agent

15. 9:55

    make sense here right um I I feel like um for me personally I essentially use butnik V to contract a lot and after a point of time I was kind of um done with the uis for multiple reasons so I was actually doing CLI to do it and it was bit complex obviously um I feel like the AI agent the whole natural language processing interface to interact with the smart contracts or even just general data is just an easy way to sort of interact and

16. 10:34

    I feel like even in the UI uh doing the going to the exact uh point of doing something is is difficult as compared to just saying something like what are the proposals I can vote on uh show me the details of a certain proposal I feel like it just makes it very very easy like this stuff is easy in the UI for sure um and sometimes uh uis can be clunky and whatnot but with with natural language it became like really really

17. 11:06

    simple right it's it's like interacting with the V2 contracts in a more organic way than interacting with it in fixed seven things that a UI can do right so stuff like and I'm thinking to add much more functionality to it in which you can do probably more intelligent stuff say something like what are the most important proposals I should look at right now and not just the top five

18. 11:32

    event what are the proposals I need to give more uh importance to right now stuff like that so just interacting with all the V2 data in a very organic SL userfriendly way is is the ultimate goal for sure uh but yeah to me uh because I keep on sending my friends and other people of the different Dows that I'm part of to either create proposals or sign proposals or just push them to vote on

19. 12:03

    proposals that are pending um and then they have to go to I don't know multiple clicks to find that out but with the agent it was as easy as saying uh get me all the Dows I'm part of get me all the proposals I can vote on show me the details of this proposal vote Yes vote no Etc so it's super easy and uh it was it was a personal pain point for me so I actually just wanted to build it myself and my and my other Dow members initially and also kept it

20. 12:35

    in like beta for two months and not told about it to anyone before it went to production yeah no I think it makes sense and slime this is a good time to kind of bring up also what you built with the staking part is that one of the things that agent seems to do really well is make crypto transactions more accessible to people I no longer have to to go through my wallet even though this is actually inside a wallet it's inside the B wallet but it doesn't feel like a wallet it just feels like I'm right

21. 13:07

    having a chat Sim did you see that as an opportunity for the staking piece there yeah definitely but actually what I wanted to do is I wanted to stake in my bet wallet I actually wanted to do this but I found out that uh there's no such thing inside the wallet but uh then I found then I thought uh it might be uh like an opportunity to create a something completely new that is U like a new era of UI so you don't have to choose which

22. 13:40

    validator you want to unstake with you don't have to claim after that you just say I want to unstake like 5 15 year and U it sees what validators you are staking with it will claim if you have already some balance unake waiting for you to claim it will unclaim from one one validator or unstake I mean if not enough it will unstake from the second validator and uh it literally everything happens with one click of button with uh one prompt and you don't have to have

23. 14:15

    all this knowledge of how things work you just and that's all really that's really interesting and so it's not only just trying to simplify the process there but you're Al actually like changing how all the mechanisms are are working without having to present that to the end user which I think is really interesting um in and this is for anybody but maybe micro you might be the best one to

24. 14:44

    answer this was that part of as you guys were starting to see the potential here building your own runtime and putting it inside the wallet was the idea of like look we can start to enable developers to do something like this we're not offering staking we we're wallet but we don't have a staking button but this is a great opportunity where like hey devs can do this and they can actually make it better than it would be normally if you guys had a button that said stake

25. 15:09

    here whatever it is 100% that was uh so we you know we were very conscious that we were building a wallet and uh we had to build all these different things like staking transfer nfts do this do that I mean the list goes on and on right but as we started integrating AI more and more into the wallet our goal was to let anyone build whatever features they want to see in the wallet and just make it very accessible for everyone right so that's uh exactly these are two success

26. 15:41

    cases for us because you know uh we don't want to build a staking we want someone that's you know that knows better how to build it stuff and uh wants to kind of maintain that to do it or Dows or you know any use case and that's uh 100% uh the reason why uh we we let everyone do this yeah uh jazz or slime either one how was the actual development process for something like this what was you know both um what challenges did you maybe

27. 16:10

    face and kind of just overall what was it like to build something like this versus something more traditional like a web you know front end and back end for me it was like a just a format macro like you get um some data in input like uh the the validator address don't here and the amount and you basically just uh multiply the amount by some uh by 10 to 24 and uh just return the transaction and uh it sends it so for me

28. 16:43

    it was not uh like uh anything difficult it was like just an API point that uh forat the data the way the taking agent will the AI will understand it and uh yeah interesting so really data driven right the can do the can do workflows on its own you just have to give it a specific command that the user might not um give the command in the spe see the correct format for the first time but um if you have like even a hardcore prompt

29. 17:22

    you can uh just instruct it to do certain things and it will do it nice right yeah there's actually something that I copied from Slim's code shamelessly which was just returning a prompt and a data as well so instead of returning an unsigned transaction which was what I was doing earlier and was seeing some hallucinations I started returning a prompt saying this is the data please use the generate transaction tool to generate the transaction instead

30. 17:57

    of returning an unsigned transaction and it was so much more accurate uh with with the transaction data otherwise earlier I was seeing some hallucinations so that kind of made me realize that you can actually return anything from your apis and not just transactions because my earlier understanding was that you return unsign transactions and then the and then the wallet just lets you sign it um and the and the only thing that the uh runtime does is pass your natural

31. 18:28

    query into to API Arguments for for the specific API that are available uh but it actually also does a bunch of stuff on the on the returned output so it's not like the returned output goes into a dumb system so to speak it is also managed by an llm and you can actually you can actually do prompts you can say here's the here's the Dow data but show it as because for instance the Dow data is returned as a Json from the from let's say the pike speaking API or the

32. 18:59

    near blocks API you can return a prompt saying this is the proposal data but show it as staes on the UI instead of returning the data so you can do I've been thinking you can do some sort of prompt engineering with the return data as well you can say stuff like display this adjon um display the token ID separately stuff and have used some of that stuff so yeah uh but yeah to iterate Slim's Point developing the API was easy for me with the Dow agent the

33. 19:32

    difficult part was that I did not know the Sputnik R2 contracts that well uh so there was a there was a learning for me to dig into different functions and different um arguments and what not parameters uh so that was fun uh but it was it was relatively easier uh to spin it up and then it iteratively keep on adding more features to it awesome so I mean you both have been Developers for a while you've built a lot of different things both in near and other places is this process now working with

34. 20:10

    AI does it feel like it is its own learning curve of just trying to understand the system too or is it something that like you know it's just like any other project that you're now working on and uh you're focused on the data in data out it definitely has uh some learning curve because AI is like a whole new way of programming like you have normal programming language languages and you

35. 20:43

    have natural text pring and you have to adjust it to the language model that is used you have to kind of understand how it works like it works by tokens by appending data and it cannot go back and uh add data before the answer so you have to have some understanding of how AI works but right now I think most people even not Engineers have a pretty good understanding of um how to use chpg

36. 21:15

    because uh it's like almost the same thing so yeah yeah for sure yeah and to touch on ja Point um the promp engineering you know like returning instead of returning data I mean it's still data but like you can return a prompt um that's actually true like our runtime will ingest your return your result whatever if it's a prompt it will try to to make sense of it and uh I think in the future this might enable things like you know you see a lot of this in like Lang chain for example like

37. 21:49

    chaining prompt and stuff like that so we're we're thinking a lot of like how can you know like the return of this uh execution then be changed to another agent execu in this way we we enter the multi-agent multi-chain or chaining prompts of multi-agents basically uh it's still early days for us for this but I I guess this is definitely where we're going and um yeah it's going to be super fun to like connect all of these

38. 22:13

    different different agents yeah that brings up a question that I was I had as well is so for people who are coming into trying to you know want to build something on P what are like what is there out of the box so like do you know I know obviously like you said the run time is linking to uh open ai's API and you're handling that piece of it but are you handling anything else that's like kind of already in there that people can

39. 22:42

    leverage um yes so uh I mean we are definitely doing the inference uh we have a set of tools like primitive tools um that you can add to your agent things like generate transaction generate Media stuff like that that's uh out of the box um and I think that's pretty much it now we're um removing or not removing but like taking the runtime out of the wallet because it's still part of the same system it's a um you know that that was how it was built but like now we're

40. 23:15

    removing the runtime and we'll be open sourcing it to like I think this is out of your question but like the thing is like we're trying we're trying to make this uh more open so people can contribute to to the runtime to how everything works um but yeah let me let me try to share my screen maybe and I can show you uh the boiler plate uh one of our examples to start with yeah that would be great and as you show this I definitely would be curious I think for people who are are watching looking

41. 23:48

    to uh to build at some point it might be great that you know slime and Jazz as you see as you go through stuff maybe parts that were uh I think important to note because I I I've played around with a little bit too building an agent myself and it's a pretty you know straightforward process for the most part but there are a little hiccups here and there so I think it' just be good to

42. 24:11

    just again recognize like where some of the things might come up for people so take it away micro okay yeah so bits of protocol SL agent next boiler plate this is a you know simple next project um and uh yeah so I just forked it I cloned it uh here you can see two main things so API there's two folders AI plugin this is the Manifest this is how you describe uh all of the different available routes um each one is treated as a tool in our runtime and

43. 24:48

    some other things like the title or or the name of your of your agent descriptions instructions how you want it to to act the Primitive tools you want to import this for example generate transaction will will be able to generate new transactions um and yeah that's pretty much it this is a open API standard this follows the open API standard you can see version 3.0 and yeah then the tools each one each folder here is an API route on the next project and has the actual logic to

44. 25:19

    to run um you know the actual tool logic so we have built uh the make agent make agent is our CLI tool for uh running this locally deploying um it helps with all the different steps of the life cycle building an agent to deploying the agent uh and I'm just going to run it so like buav will run uh my next project and make agent so I can just do Bev it should all right I'm already

45. 25:51

    authenticated this means that I've already signed a message uh we need this message to pretty much every single time you change your code um we need to send a request uh with the changes you've made back to to us and we use this key uh that you sign with your wallet maybe I can can even delete this and show the whole process wait a sec well that's a quick question there about that so as a developer and as you're building on this oh as you change your code is it auto refreshing the

46. 26:25

    background like what do you have to do to make sure that the agent is actually working with the latest uh changes that you made uh so in fact we just need one thing we just need to update the Manifest in our back end um so we we store this information it's it's you know it's already it's public when you when you um when you deploy the agent so we don't really keep any private information or anything um but for now I think make agent reacts to every single change in your file system under you know the the path you you started make

47. 26:58

    agent but this is something we're going to change uh probably next next week uh because we just realized we don't need that we just need to like know whenever your manifest changes so we uh send this um this update request what happens sorry what what happens when that goes on is the because you know the AI is obviously retaining some information right so because it's like it's context

48. 27:22

    aware for that moment is it kind of like resetting itself a little bit or are you still kind of of in that same singular session when you update yes I guess not sure if I follow your your question so so yeah I guess like if I if I make a change to to my manifest right and so I I've changed how where I'm going to get data from does that current session that I'm in right now will allow me to know access that or

49. 27:56

    do I need to refresh anything okay yeah yeah yeah exactly so like if I change let's see this name for example um near thehub system I don't know so now you can see change detected and it's attempting to update it's updated so I guess now our UI will will suck at this but uh it should change eventually here so let me send a message maybe hey so you can see see that this didn't update but uh the name here already

50. 28:31

    updated so every single thing you change in this file like our our runtime will be aware of got it got it yeah so you can see near da Hub assistant here and the yeah so yeah where was I going I mean yeah this is running you can see that I signed the message this is my my signature to validate every single update I make locally you can see the bit of config this includes things like the URL this is local we need to do this because our runtime initiates HTTP requests to your local

51. 29:03

    tunnel your local server and in order to do this you need to make it live somehow and we're using local tunnel serve you uh those services to make your local server live and now I mean this is you know we can change anything here maybe what do you guys say like can we try to maybe like get Reddit posts this was the thing I did last last session here sure uh you can see there's um a tool it

52. 29:28

    explains you know like this can fetch and return list of posts of the Reddit front page it tells me the schema of my response so you can see there's an array of items each item includes author title subreddit score bunch of things and um well didn't work let me try again interesting always with lives no problem sessions you know like um funny yeah I realized refreshing the page always

53. 30:12

    helped should I should I just refresh the page and now there's a weird thing going on okay um got RIT there you interesting yeah interesting chz you had a question do you want to ask you question JZ yeah um so in the in the environment file I saw the B the key with the the developer account and the and the key so every update is being signed by the by the developer account or something yeah so for for example if I change um I don't know let me add just

54. 31:03

    the character here maybe uh Q whenever I detect this change um interesting it is supposed to send a request yeah see plugin updated successfully for so for every single change we sign this with your with your key and send that to our servers and then on the servers we validate whether that change comes from the developer which we check against the account ID here and we validate if this

55. 31:39

    account ID signed the actual message and I don't want to show my whole signature here so but yeah that's interesting so essentially saying another developer cannot make changes to your agent because it gets verified your signature uh yes for now but we want to include multiple accounts here so this is this is part of like our this is an our backlog now to support multiple accounts um from the

56. 32:12

    plugin so like you add the accounts that are uh you know that should be available that should be able to um make updates to your manifest or your agent is uh is can you maybe just take a few minutes and talk a little bit about um since this is kind of more developer focused about the run time and like what's actually happening kind under the hood obviously there stuff that's proprietary you you can hold on to that but um anything that you can kind of share like what's really happening

57. 32:45

    there yeah so right now and let me open this maybe here so we can okay let me so this is um um me go to the AI so basically uh the runtime is very simple we um we well first we get messages right and then we we based on whatever messages you send um we um we we we use we we send this against U nlm right now I think the the one we're using by default is 40 uh GPD 40 but we support a few others um this is undocumented for now because we didn't have like enough good results with the

58. 33:33

    function calling but um they are there so like we can load multiple llms um in the runtime and then based on the agent you select uh we um load the the routes or you know like the different routes that are available per agent as tools so they they get picked by uh with function calling so for instance if we go here um let's say Gecko and we try to get near rice you'll see that this calls you know

59. 34:08

    the uh simple Price tool to get the the new price and then you can combine multiple agents in one chat so like if you go to let's say uh ref Finance where is that here so all of this is kept on on the same chat and uh so like multiple agents can like share the same context and uh do different things I'm not sure if I'm if I'm like answering your your WR time question but it's it's very simple it's just literally we load

60. 34:41

    different function calls uh and based on whatever prompt is sent we uh we just you know just uh generate text basically yeah I did not I did not realize about the the multiple agents using the same context that's really interesting so is that is that a product of the way that you are handling sessions yeah yeah interesting so let's say if you want to actually I was actually this one is

61. 35:14

    interesting we just uh launched today with um base support and you can check what's what are my balances on base yeah so I I use it before uh to let's say swap 0.8 usdc to dgen on base this is using CH abstraction and we'll generate the transaction the evm transaction and then will print really nice UI for me to sign this transaction so you can see the evm transaction data and then I'm signing it with my new year

62. 35:59

    account but yeah so I got my balances and what I wanted to do actually was I want to switch to coin Gecko and what's my what's my port how okay how much how much is my portfolio worth on base I did should get like the the previous messages about my oh you see but so uh L with the sorry I'm interrupting you with the shade context could we do could we do something like uh we could let the Dow agent have the context of a of a swap from The Ref agent like would something like that

63. 36:57

    work because slime and I were also just talking about it recently U the Dow agent doesn't do swaps right now U but the ref agent does could we somehow get to some level of even if it's inaccurate but could we could we mix match somehow yeah uh so right now there's there's no way to do that b you know the user we need to select individual agents uh and send the prompt but we're we're thinking of how we can um how your agent

64. 37:31

    for example like the Dow agent can initiate a request to the Bre Bre Finance agent um and yeah that's that's where we're we want to go next and that was the initial goal actually of uh this you know we call this smart actions I'm not sure if it's still somewhere um I think we've been renaming things but uh yes it's your smart actions so like the goal of smart actions was to like uh write sort of like scripts like prompts and then um

65. 37:57

    send this to the run time the run time decides whatever agents he needs to talk to um and execute things for for the user so we're definitely going that direction not sure exactly how it's going to look but I know we're uh we're again like as I was saying like we're removing a lot of these things from the wallet and uh making it for example like the we're we're building an API so so

66. 38:20

    like you can you can talk to different agents um with only HTTP requests and and I want to try to show some Alpha but I don't know where that is actually so maybe I need to stop my screen uh to search for this stuff but I think it's also interesting now that we're here I can show for example Dow agent uh if you click here you you're going to go to like a really nice um sort of like dashboard where you

67. 38:51

    can see the things are just messages sent individual messages sent to this agent so you can see that um and we started recording this maybe like three four weeks ago uh and yeah you're having a daily average of eight uh you can see the development history so like when Jazz works on this thing you can find the the repo Whoever has like an open source repo will verify immediately and show it nicely on the UI so like we incentivize people to open source these things and you can see like multiple

68. 39:22

    agents so like meme cooking I think this was a hit we having we're having a couple really interesting activity here um yeah and a lot of other stuff like staking it's pretty nice too let me try to stop my screen for a while so I can find the AL yeah so the the idea here is that um one like just like any other app like you you you get a sense of you know the development still going into it um is this an active project all of those pieces of it can people still submit though a

69. 40:05

    closed source project or do you require it to be open source they can definitely do it yeah well you know using the the registry you just need to pretty much make it live you know like accessible on the internet and send and register that that link then we check we make uh we check if it has valid manifest all those things but it doesn't require it's not required to be open source um got it yeah yeah but

70. 40:31

    it's nice if it is open source right yeah yeah yeah sure and um we're we're now thinking more and more about monetization and this is something I'm trying to push um for early January to be kind of like we have a you know like a live vzero sort of sort of deal uh where individual tools or individual API routes are monetized so let's say if you build for example the D agent if you want you know to monetize your um Dow Dow prop proposal submission Tool uh you

71. 41:05

    can put a price on it um not sure exactly if we're going to do like near uh only or support multiple currencies or you know have sort of like Fiat and then we uh convert that into the multiple currencies that you know like you can pay to the to the agent but the goal is for every single time you use a tool uh you pay for it and the developer is kind of like making money of it of it

72. 41:29

    yeah really and so because of something like that is that where even though I know you want to pull some of this stuff out of it is the wallet piece always going to be there for you like do you see how that goes like long term or what does that look like I guess yeah I mean the wallet was very important for us um because again we our goal with the whole wallet AI thing was always to simplify how people access crypto uh and

73. 42:01

    we know like for for a while we've been in the in in the space and we know that in order for crypto to kind of win you know like we need we need to make it accessible for everyone and uh you know we Tred to do that with nfts first then the wallet was to simplify you know use Rel layers and all that all those all those things to you know make it possible for people to use apps without

74. 42:22

    crypto and AI is just you know like um the next level where you can just use whatever language you use uh even if it's English Portuguese Spanish whatever you can still interact with these applications somehow like it's still not it's not perfect but again like it's uh one step closer to make it widely accessible so to answer your question yes I still think the wallet will be there uh and it it's our playground for

75. 42:49

    a lot of these things but more and more we want to abstract like how people access uh our infrastructure so we want we don't want to force people to go through the wallet uh to do these things we want you know you to import the runtime wherever you are maybe it's a CLI tool maybe it is a website maybe it is a native app I don't know but the goal is for um to let anyone um you know access through whatever client they want

76. 43:12

    and that's where that's why I'm trying to find uh this thing to show you how we're we're what's our first step into into that uh um I my internet do very slow somehow but yeah why you looking for that I W to Simon Jazz I want to just talk as a developer here from like an effort standpoint and time standpoint is it's something where you see yourselves spending more time working on agent things not just for

77. 43:50

    within bit but just in general or is the traditional route still just as important because that maybe that helps drive an agent later like does it kind of go hand inand for you guys or do you see this kind of yourself moving more towards let me work more on the agent type of area I think both are important like um for more advanced users having um like a smart construct interface or a normal user interface um

78. 44:23

    is uh would be um more use you can like do more of it and you don't have to type a long sentence of what you want to do you can just uh do three button clicks but uh for new people who are new to crypto who don't have technical knowledge they just want to say something um it's easier for them to type a sentence rather than click uh do three clicks so I think the base should be the normal UI and uh the agent should

79. 44:56

    be for onboarding more people to make it more easily accessible to to my grandma MH yeah so the the you're saying the building process should kind of include both right we don't want to just go full agent only right it's just like we need this progression here of tools yeah J what are your thoughts here yeah no I was also going to say that building agents at the moment are very very tempting and um I haven't been doing much been doing too much but uh

80. 45:27

    yeah I I also agree that I I kind of split my time in doing the traditional stuff as well uh there's there's learning component and there's just I feel like some of this stuff can be readily consumed by an agent whenever I want so I kind of just split my time spending in both both sides but yeah building agents is tempting and it's super helpful uh but yeah some of the traditional stuff is important and uh I feel like it's also just creating more

81. 45:57

    and more resources for the agents indirectly uh anyway so yeah I I personally try to do try to do both yeah I think that's a great Point especially around you know if like the concept of on crypto smart contracts is really important still right because it has to get the data from somewhere right it has to you still have to have that chain interaction and so it's not just about throwing more agents at stuff it's a

82. 46:26

    it's a nice mix next year right now and it seems like that the opportunity again with that bit is providing is to get up and running with an agent as a additional feature as well for sure and if you don't have an interface you if you only have a smart contract then I believe someone from te made a tool that creates Agents from Smart contracts ABI and uh you can use it without instead of uh UI

83. 46:57

    if you want to test your contract before creating AI so it can be a big developer experience booster yeah absolutely absolutely I mean that that's a great great point so for developers that's huge because how much how often do you get to that point where especially more complicated contract that you're trying to test out different things and yeah you can write test cases and you can do so from the command line but it's nice to be able to against speed up development by able just to talk to it

84. 47:29

    or just talk to my contract see what's happening make sure I'm not missing stuff it's not perfect but uh absolutely it is a productivity boost yeah I also feel like with uh B being fairly popular it also gives you sort of a sort of an attention space for you to just launch the interface as soon as possible because otherwise you would have to Market Your solution uh yourself but with adding an agent the the registry does it for you so anybody who's who's going to the BET interface

85. 48:09

    to use the mem coin agent or the staking agent will also see your agent listed there and maybe give it a try so it kind of solves that z21 problem for your initial users as well for me at least it did so great plus Point too absolutely absolutely Mike are you ready um actually I I found something but I don't think it's ready to uh share worries I put you on the spot no cool yeah but the goal is again like we we have the the the repo is open source too

86. 48:48

    so like you can find it I just wanted to show it but it's not quite ready yet but uh the thing is like we're uh you know for f developers that want to integrate our ch this will possible very very soon so like it's a react component you can integrate on whatever app uh just import beta AI chat and all of a sudden you have you know like your your app is enabled with AI and you can talk to all of the the agents in the registry and and things like that nice so very very

87. 49:14

    soon stay tuned yeah yeah yeah yeah that's that's great uh look I you know this is a direction that clearly a lot of developers are starting starting to put more time and effort into um we've talked a lot about on this show about the different components that go into AI both infrastructure products on top with it agents things like that so it is a nice opportunity for developers to kind of stretch their wings a little bit and and see what's happening I'd love to

88. 49:45

    just get a sense of again this is not holding anyone do this just kind of like where do you see the next six to 12 months of development in the space does it keep going with agents is Legion right now just kind of this first wave what are your guys initial predictions here whoever wants to go first can I mean I think yeah can I can can share my my opinion I I mean this is we're definitely in the in the first wave of Agents I don't think you know

89. 50:20

    we're we were really exploring the full potential of what this could could get um especially right now it's very it's it's just a very good thing at find at predicting tools uh or at least that's how I see it you know like you just s The Prompt it's very uh catered for answering or assisting people and uh equipped with a bunch of tools and it would just pick the right tools and and help achieve whatever but it still needs a lot of a lot of handholding you know like it still needs to be interactive

90. 50:51

    somehow I I don't think we're right we're there yet with autonomous agents but yes I guess in in the next couple of months we'll see more of that in fact I just hired Devon uh to our team I've been out thinking about it for a couple days and uh you know F mode into it and uh it's actually building a PR for me right now and this is where I think things will go over the next months maybe in terms of development and uh how it will help developer

91. 51:20

    teams uh and I mean we'll we'll try to adapt again like this product product uh wi I think we need to be um you know we need to be exploring and and and trying new things all the time and that's where we're going to go with the be run side and the framework and everything will will be enabling more and more of the autonomous agents and whatever comes next yeah cool before we go to the other

92. 51:46

    two guys can you just people who do don't know what exactly Devon is can you talk a little bit about that yeah it's an AI agent I guess semi-autonomous because it it's it reacts to requests it can uh you can it can react to uh or be triggered by uh GitHub issues or slack messages things like that so for example I just gave Devon which is just a product you know like uh on I gave access to one of our repos then on slack I just said like hey Deon can you help me with this and uh if you want me to show how it looks like I

93. 52:20

    can actually do it because I think it's very exciting yeah yeah yeah um so you know I just sent a message this is for slack I was very uh I just said like Devon can you help me refactoring how tools work this is actually the boilerplate example I was showing before and maybe improve the AI plug-in manifest I didn't even give like any Specific Instructions I was like go crazy with this high level request and

94. 52:51

    uh yeah I kept sending stuff here a lot of messages you can see you can replay all the things that Deon did so like he cloned the repo then did r a couple of commands did all this stuff you know and you can even see the browser lans's view not sure if this works uh shell all the all the things that it did uh and yeah I think it created a pool request so like maybe we can go to refresh this page and we have uh new pool request that was opened by

95. 53:29

    Devon with my request it changed a lot of things a lot of crazy things here yeah did some enhancements I guess definitely hallucinated here we don't have this uh the stuff in our Docks but I mean I think the more context you give it the more the better we'll get so yeah it's just this was the first request I asked you know but yeah yeah no really interesting I look I I think all these things well not perfect yet are really interesting

96. 53:59

    to play with and again for yourself right as someone who is leading a team it's got to be interesting to look at like how does that affect uh you know how many more team members you need to bring on what does that mean for your own time and Staffing um you know right like I find myself like you know the AC of coding is a lot longer than me doing code reviews right so if I can have the

97. 54:24

    AI agent you know write all this code and then I spend even if I have spent an hour doing a code review that's less time than it would have been for me to actually maybe code everything by myself so it's like finding those balances uh I think are really interesting so I'm definitely curious to see how this works out uh for you over time yeah I I think for me the decision making to start because I know this wouldn't be perfect from day one but at least there are two

98. 54:49

    things I wanted to to improve from day one and I think the two two things are removing the need to spend so much time on fixing bugs and you know our backlog is still filled with a lot of bug bugs to to be fixed and I I I really feel like like our team is more productive if they are focusing on new things or like uh and and we're we're Innovative team right we we're always trying to innovate

99. 55:12

    and you cannot really innovate if you're like always trying to fix problems right sure um so that's one I really want to see if this helps us with um fixing a lot of the bugs so like we have a QA guy QA guy can talk talk to Deon and ask for those things and then the the team can can review uh if Deon did good work and then the second thing was I I still feel there's a lot of there there's um the a

100. 55:37

     cold star problem with when working on new tasks uh and a lot of that I think can be helped with better uh explanation or or like spending more time writing down what the task should be about instead of like you know right now we have you know things like work on this but we don't explain what this is and there's a lot of human communication there so I feel like solving this cold start problem and spending more time

101. 56:04

     defining tasks and then let Devon go crazy with that and only then like I think the dev team start working on that stuff uh would help a lot uh with our team so that's where what I'm what I'm exploring here yeah no that's that's that's really interesting it's and I I like that Philosophy from a developer standpoint too so if you're a small team or even solo Dev right of like not

102. 56:29

     getting caught up in trying to bug fix these stuff before trying to keep pushing forward because that is something that does stall people all the time right like how much time you get get going and I face it myself all the time of like and now I'm going back and now you're stuck and uh especially if you didn't write yourself great documentation um sometimes going back is a bigger pain in the ass so no I think that's a a really interesting aspect uh Jas how do you guys see this in your

103. 57:00

     workflows I know you both are fairly independent developers but uh how do you see this working out for you yeah um honestly um my day-to-day workflow right now is very very AI dependent honestly uh but most of the stuff that I'm doing is as uh Louis was saying is mostly fixing things with AI because I don't want to waste time going to stack overflows to to fix things although I kind of enjoyed that in the

104. 57:33

     past uh but and and also just the cold problem where if you have an idea you brainstorm a bunch and then you you get a get a Proto prototype and then you can start hacking adding new new ideas if there's a syntax or other kind of things uh right you can you could get that very easily as compared to going through the documentation of doing a Google Search and going through a bunch of stack Overflow stuff so I I I honestly use it as a mixture of uh reading the initial

105. 58:05

     docs and then uh finding some answers really fast that I wouldn't do on stack Overflow or other sites that that fast um but then also just solving bugs because it's it's annoying sometimes and AI agents can sometime lead you the wrong way so you have to keep uh showing them some some Direction but it's still easier uh and for me the biggest pain Point all my career has been I just suck at UI uh so that has been made like that came as a blessing to me with AI agents like now I'm a UI

106. 58:43

     engineer as well if you want me to be so so it's it's fully integrated in my in my workflow right now and yeah very cool uh slime I use even more I think if I really hate doing uh front end uh this HTML CSS stuff so right now I think in 2024 about % of all the sites all the U iate was uh mostly AI generated it was like a clo it was web Sim all these tools um I basically do almost nothing unless it's

107. 59:24

     something with near because the I don't have much knowledge of near right now yet we have to make a lot of documentation and have the cers uh CW our documentation and libraries to improve this or train our Med models but for now I think um we can do pretty much uh something like like on a almost a junior level developer can be almost replaced by AI yeah and this is going link to just a grow almost um in like in one year you will not um you will not

108. 1:00:07

     predict what would happen yeah no I great great point there um especially around you I I also hate front ends so if I don't do HDML anymore I'm I'm there for that uh yeah this is something that I think is definitely interesting and exciting for Developers uh it's not just about like building AI it's also just how you interact with it so um this was great uh I really appreciate you three taking time out of your day to come and talk about this uh and everyone should go check out these

109. 1:00:40

     agents that they built uh again uh if you are interested in building an agent uh on bitte and it maybe fits into their Bounty program uh if you go to b. they have you can see the links to their Bounty program uh and if it fits in within kind of Dev hubs uh perview as well we will match that Bounty so a lot of opportunities to be rewarded uh and really build some pretty interesting stuff and again this is something that

110. 1:01:10

     is new so you know if you are creative you have different ideas I'm sure uh micro and his team are open to hearing uh fresh ideas as well too um so before we go though uh maybe just kind of go around anything you want to promote anything want to uh shout out anything else like that um we take a second but we'll wrap up from there uh we'll start with you slime what do you want where should how do people uh stay in touch with you see what you're building try out your

111. 1:01:40

     stuff check out my staking agent um it is sponsored by um five pools that are not actually monary sponsoring me but uh I just picked some recommended once from my Twitter in comments so if you don't have uh if you don't know who to St with you can just uh ask it for the St U for recommended pools nice so sure to do it nice and you have a you have a couple telegram Bots too as well right yeah I have a beter bot it's a dragging bot it

112. 1:02:13

     has notifications it has a new CLI buletin it has a it'll be it'll take like 10 minutes to explain everything that it has yeah no that's awesome yeah uh I know you're building a ton here on near I definitely appreciate everything that you that you do looking forward to seeing all the other stuff and we'll put all the stuff in the show Notes too uh Jaz anything you want to shout out or show off here yeah um I would just

113. 1:02:43

     recommend everybody using the Dow agent for your day-to-day Dow needs and let me know the feedback uh if you see any bugs please let me know on telegram or Twitter you can find me on Twitter easily um and yeah if you have more bit agent uh recommendations suggestions if you need want to see an agent uh that's also something I would want to do so yeah hit me up perfect perfect and uh well on you micro what do you

114. 1:03:13

     got I mean the the usual bit of protocol on Twitter uh GitHub there's a lot of stuff bounties um documentation you can actually like there's a few things we will we will share and I think next next week uh on on Twitter like the different agents were very interested to see you know to name a few like intent and year intent um agent for example there's a lot of stuff going on on Twitter we definitely want to see some of that uh

115. 1:03:40

     yield farming that's uh you know that uses stuff Like Ur a or different different evm protocols and I I would be a you know especially in Europe now uh with the m law we cannot be earning uh with usdc anymore on Cur BAS so like we definitely need need some agents that help um everyone navigate the different protocols and the the you know the risks um and know you know we'll be sharing a a list uh and and yeah I mean follow follow us on Twitter I think bits posted

116. 1:04:13

     a Twitter a tweet oreet I don't know how to call them these days but uh with all different DS in the team and uh follow us and yeah that's it perfect perfect well uh again thanks so much everybody for you guys for coming on uh this was a great episode and looking forward to see everything you guys built thanks guys thank you Joe thanks uh awesome stuff everybody uh that was really really great and it's

117. 1:04:42

     something that we want to continue to kind of keep talking about if you have ideas for agents things like that definitely go check out bit day um that's an awesome opportunity to kind of get started today uh just a couple quick housekeeping things uh so I'm scrolling at the bottom of the screen again if you haven't signed up for the devhub newsletter we have all the things happening in the developer ecosystem each and every week that comes out on Mondays uh definitely sign up check that

118. 1:05:07

     out see what's happening uh and stay in touch with all that stuff uh dev. n.org is the place to go to get started building on near uh everything and all your resources are right there uh to get going and if you need developers support our telegram group is linked there as well next week uh we have a special episode kind of re bringing the race of sloth team back on uh we're going to talk about some of the different changes

119. 1:05:31

     that they have made some of the different things that they're they're building and growing on uh how they're kind of opening it up to all projects as well and just kind of a a general recap there a really great thing for developers and different incentives and things from there so we're going to talk to that team uh next week Al we go into the end of the year here as well uh we'll have some more workshops we'll

120. 1:05:54

     have some more end of the year kind of wrap up stuff and all the way into the new year uh we'll continue to have uh more great content so if anything you're looking to see you want to learn more about please reach out to us um my personal Twitter handle and the near Dev hubs is there on the screen just reach out to us and say who you want to see and we'll get them we'll try to get them here uh on devhub live uh but until next

121. 1:06:18

     week uh I am Joe I've been your host and we will see you same time next week at devhub live take care everybody what you want you want you want you want