Brainfood Live On Air - Ep415 - How to Use Claude for TA (Talent Acquisition)

0:28Are you? brain food and say, Oh, isn't there a briefing call? I said, no, man. No. This initially, the reason why is because brain food live was started as an open mic show. So the entire idea was that I would just get on screen. If you want to come on, just come on. It's like, literally, that was it. And that was pre COVID. So pre remote, no one had the gumption or the interest to come on camera then, but it might come back in some format. So I'm kind of tempted in 2027 to run more of an open mic session saying, you know what, I just kind of want to get a kind of a chat roulette type of thing going on.

1:09And then just hope everyone's going to, you know, play it safe for work and so on. But yeah, it'll be a good way to get more people on screen and sharing dialogue and so forth. Anyway, folks, welcome to Brain Food Live on air. I'm bringing it to you every Friday, no fail. We've got a brilliant show for you today. Of course it is. We are talking about artificial intelligence. We're talking about TA, how those two things interact, and particularly how we're going to use Claude for talent acquisition. It's a massively oversubscribed show. I think today we're probably going to get more people watching this than we have all year.

1:43I can already see we're something like 880 odd people are subscribed to this, probably two, three hundred live already. And we're multi-streaming in this, hopefully in different places. So very important that I establish firstly that you can hear and see me okay. So audio folks, can you hear and see me on that little monologue? You should see Ala Pavlova and Adriana Herdman on screen as well. If you're watching this on LinkedIn, let me know if the audio and video is okay as well. We should be live streaming a number of Linkedins. Definitely my LinkedIn should be going live. So let me know if the audio and visual is fine there and we should be good to go.

2:23Okay, Tracy. Thank you, Emin. Appreciate it. Let me just check my phone because that's how I check my LinkedIn, to be honest. Am I on? I really hope I am. God, why am I on live there? All right, maybe I screwed up. It's possible. These things can happen, you know. Anyway, I'm live on Twitter at least. And I think we're live in different places. All good. All right. Brilliant. Let's get on with the show. Folks, let's introduce our wonderful panelists firstly. Ala, let's go with you first quickly. Who are you? What it is you do? Hey, everyone. My name is Ala Pavlova. I'm based in Amsterdam. I am a hands-on recruiter, but at the same time, I do train TA teams on how to use technologies.

3:14AI was one of them, of course, and also other tricks and tips that can make your life more productive and easier and more fun. So I'm excited about this show because I'm a huge fan of Adriano as well. Yeah, fantastic. Great to have you on the show here, Ala, as always. Adriano, great to see you back on the show as well. Why don't you quickly introduce yourself, who you, what it is you do? Yeah, thanks, Ala. Thanks for having me on. I'm Adriano. I'm co-founder of Move. We're an embedded talent consultancy that supports technology businesses in hiring. But you may know me because of my content on LinkedIn.

3:51So I'm a big advocate of AI in TA and obviously Claude as well. We use it a lot across our business. So looking forward to sharing what we're learning and also understanding a bit about how people are using it themselves. Fantastic stuff. Let's get into it straight away, guys. I've just shared the LinkedIn's to both Ala and Adriano. I do recommend you follow them immediately. Just connect with them. I think they're probably going to be okay to connect with you. But if not, then just follow their content because it's really interesting stuff, always producing high value contributions to the community there.

4:28So good stuff. Folks, as you're watching this, I want you to direct your attention to a poll I have set up on the right-hand side of your screen. there should be a bunch of icons there click on the poll please and can you sort of just let us know where you want or where you feel you want to see a bit more action on Claude like what do you want Claude to do for you in TA for the folks who are watching this via audio I'll read out the options we have sourcing as one option assessing as another option scheduling, coordinating report stroke intelligence admin and then other please comment. So obviously, if there's another thing, do let us know.

5:12But let us know in the votes, where is the weight of work you want to have Claude support you? Interesting for us to just have a view on that. Okay. So whilst these votes are coming in, that's good stuff. Let's kick off firstly with just a quick basics, guys. Claude, why is it Claude these days? How come we're not using open AI in the same way? Why are we so enthusiastic about Claude? Can you give us a bit of your own take as to why you've gone this route rather than any other model? Adriano, you first. So, yeah, it's funny. Full disclosure, I have recently started using ChatGPT quite a bit again, but I haven't let go of Claude completely.

5:59Claude's amazing for lots of things. It's just we can talk about how maybe some of the guardrails you need to or stuff you need to be careful of when using Claude. to produce stuff because it's not perfect. But yes, I started using it a few months ago and was blown away by its capability and things have evolved even since then in a crazy manner. So it used to be like, okay, I'm a bit nervous about using anything other than chat. And you have like co-work and then you've got clawed code. Now it's so sophisticated that it doesn't matter what route in you do. The idea is not the limitation anymore.

6:36It's the theory that goes into it but anyway Claude's phenomenal for I guess like the infrastructure behind the scenes that connects all your tools now Chachapiti does that but it doesn't do in a way that I think Claude does it's also phenomenal at producing visual assets that Chachapiti doesn't quite get right as well so those are the key differentiators I think it's like in terms of copywriting it's not the best it's not it's like really verbose sounds robotic I think it's because maybe loads of engineers are using it perhaps and it's just starting to talk like an engineer um they're correcting that a bit now but um the capability behind the scenes in terms of like connecting top of funnel to what happens in pipeline is just it's just crazy like we can unpack that later but it's it's mad what it can do and you don't need to like honestly you don't need to be technically savvy at all which wasn't the case six months ago yeah yeah fantastic stuff um how about you ala So what's your take?

7:29Why Claude specifically compared to other tools? Or is it simply the case that you've made an, I wouldn't say, arbitrary choice? But we do kind of pick one or the other, right? I mean, not one is better than the other. Microsoft, Android, it's like, you know, back in the day, it was MySpace versus Facebook. It's like, okay, how do you choose? I mean, maybe they're pretty samey. What's your take? Why Claude? Yeah, so I think I do like A-B test. and Claude just showed working with text much more effective. I've tried to build my tone of voice with ChatGPT and my tone of voice with Claude. And the one that Claude had the latest model built was really sounding like me, sounding like I wanted to sound.

8:16So it's a long-known reputation of Claude models to be really good with text. And, yeah, I think I see that, like, companies are having enterprise versions of ChatGPT, but people at home really using the law of Claude. Because one of the reasons, I think, as well is you have projects there. So you build skills. There's, like, a lot of connectors. And when I talk to people that I train, they say that it's a little bit easier to get acquainted with Claude rather than probably when they start with Chargipity. And in addition, Claude has an amazing feature that I would like everybody to try Claude Design.

9:04it's doing crazy great things like websites like one pagers motion um cvs that are really great so i think yeah but i'm not saying the charge gpt is better or worse i do think their model to create pictures was one of the ones that still stays one of the uh you know good ones um but i do see personal usage a lot of people will just go for plot yeah really good and both you mentioned the design aspect so maybe it's a good place to start um and and perhaps uh adriano this is a good time to to to present an aspect of design um because you i wouldn't say patented the idea because it's obviously just you know it's it's not something you can close down but your one pages that you publish on linkedin always viralize i don't know how sort of many impressions you you've had using this technique probably into the millions about definitely into the millions and deservedly so because the one pages are so impactful that everyone just wants a copy of it and it in fact stimulated this conversation so i wonder whether you could just flash that up on screen for us it'd be good for everyone to have a look at this and see where see uh sort of firstly an example of the design but secondly let's use it as a guideline yeah yeah how we're going to talk about this yeah so for sure it was made in claw design as well so that would make sense if it was i'm glad it was used in in the tool that it's talking about um let me share my screen here we go so my linkedin so you see the impressions pop up actually uh no that's just the engagement not the total impressions can everyone see that okay it looks okay let me just see if i can boost it shall i zoom a bit bigger yeah yeah that big let me make it full screen is that better yeah and also people people who are watching this can also go full screen themselves i think so um they should be able to see it so yeah okay what are we looking at here um cool so before i dive in i just want everyone to know that this is a a plugin on that you can download uh you can add to your cloud right now if you go to my profile There's a link there and you can set it up in a couple of minutes.

11:25It's a marketplace. It's got all these skills already in it and it will be updated live every week as myself and the team evolve what these do. We're testing them live on a bunch of different environments across different clients. So if you use it, you'll get the most up-to-date version every week because we're going to be keep updating it. So that's just a quick like how to. And it's on my featured section on LinkedIn. in. So these are a selection of skills. And on the left, on the spoke are the categories within which those skills operate. So we kind of cover it from, let's say, like intake through to, you know, closing a role.

12:03So top left is before the role kicks off, and then all the stages until you close off a role, and then you're going to do a wash up and report back. So that's the structure. and then on the right outside of each spoke are the individual skills that make up the the categories i don't know if there's anything you uh that the community wants to focus on i can't see the chat right now but um these were built up over a few months and they enable you to go maybe a little bit faster or produce something that's a little bit more valuable to the stakeholders you're working with um quick caveat one thing that's i think needs more evolving more broadly just in TA, we've got amazing, we've become phenomenal at like improving our workflows.

12:47But ultimately in this, I was chatting to Luke Eaton the other day and he made me think of this, what the candidate sees hasn't changed much in the last 24 months. They still get an email and they still get a call from a recruiter and they still get an email to close off stuff, right? So I think there's a bit of an unlock there. There's some good ideas maybe, or potential openings for how to improve that side of things. But this is much more like what the recruiter and the hiring manager sees just uh just as a heads up is there anything that we'd like to kind of narrow down on perhaps yeah i want to talk firstly i want to make sure everyone's on on the right level uh does everyone understand uh what a skill is um if you don't just let us know um because you know there's no bad question here out first of i'm i'm a little bit confused i understand the skill to be essentially a prompt a prompt that you basically store and then invoke uh as you go forward is that right adriano that's that's that's that's correct it's a it's a it's a skill is a you have a repetitive task and it captures the process of doing that repetitive task one so yeah it's just it's if let's say for example post intake uh with a hiring manager you have a kickoff call you upload a transcript typically you do that on chat chp t or claude i imagine many people do that right they upload the transcript and go please capture the ideal the essential requirements from this role.

14:05Now you do that all the time. A skill could be as soon as you upload transcript backslash intake skill, and it would give you that codified requirements straight away without you having to prompt it. So it's basically automating the repetitive. Yeah. And I can feel that this is pretty similar to how, I mean, the work of a recruiter here is actually moving a lot closer to how software development traditionally is. I mean, back in the day i mean they still do it's a case of having uh libraries you can pull in sort of uh essentially uh pre-written pieces of code and it plugs in rather than hand typing stuff uh essentially skills is is that um it's basically something that you have stored um and it can then invoke in future uh or have uh claw to invoke and presumably you can then string skills along into a sequence.

15:00Is that right, Edwin? Yeah, that's right. So yeah, I simplified it a bit too much. In terms of skill, a skill could be multiple prompts in one. So for example, we've got a skill that, again, using the intake because it's the core pillars where a lot of these skills are really useful. It encapsulates three or four prompts that normally you'd have to do sequentially and it does them all for you in one shot. So it just saves a lot of time. So that's a skill. And then obviously the other levels to that are plugging skills together. they become um yeah you can get plugins as well which is when they do multiple tasks for you back to back yeah excellent um okay some people are saying in there or in the chat the presentation is stalled um it may just be lagging folks because there's basically zillions of people on on on the live stream uh but it is just one page and so they we're not it hasn't moved yeah the the value of uh of Adriano's output is principally because it's not a paginated experience.

16:00It's like you can glance at this and potentially just print it off and have that as a thing to review. Okay, so what happened to all of this to create this, Adriano? You've basically, I guess, created these skills. And how does that help your customers? like uh what do they want so they so i think it's it's primarily so we're thinking about what the what a recruiter does day to day in their role um and how can we improve that in some way so that's the that's the obviously the guiding principle that will improve the customer's outcomes um so they started over the last few months on what do they do that's currently very manual that takes a lot of time, but the output is largely like, I don't know, the difference between superb and good isn't going to change the game massively, right?

16:56So for example, let's say after you do a kickoff call and before you start a search, you need to align on what good looks like with the hiring manager from a profile standpoint. We call that calibration. Most people call it calibration. I think that's probably the technical, the industry term. so what it involves is all right i know what you look i know what you want i think i know what perfect looks like i'm not just going to go away and find you like two slam dunks because that's obvious like i know you want someone from if it's a fintech client i want someone from a fintech that's experienced a lot of growth with the right stack and also been promoted a couple of times in the last three years and you know whatever they say that's usually quite common i can show you that person in two seconds we're not going to learn anything but let's say for example i find you someone in an adjacent industry, et cetera, et cetera.

17:42So we want to both expand the talent pool, challenge the brief a little bit and make sure we actually know what the ideal candidate profile is. That used to be a recruiter going away, finding six profiles, thinking about it quite a bit, thinking, okay, what am I testing here with this hypothesis? Is this definitely a tier one candidate? That would take a couple of hours, right? Hiring manager by that point has done something else. You send the profiles back to them. You can sometimes do it live, but there's a bit of a risk to that. It really takes a senior operator to bring up profiles live on call quite well if they're going off the cuff.

18:12So that would take a couple of hours. Hiring manager will maybe go away, forget about it, come back, look at the profiles, forget what they're doing. Maybe 24 hours later, say, yeah, we're aligned. Let's go and do a search. That normally takes like 24 hours. With Claude and the skill we've built, you can upload the transcript, upload the job spec. Now this is, not everyone will have this, but with Clay, which is a data provider, you can get a free account, plug it into your Claude. You get like 200 credits a month and it will be able to, based on the spec and the intake meeting transcript, pull together six profiles across three tiers in two minutes.

18:48So, and you're not breaking any rules. This is all for internal use. You're not doing outreach for them, but it will give you why it's picked them based on the criteria that the hiring manager is interested. And you can share that back within 10 minutes. So what we call time to alignment is aligning what good looks like. and can we go, go, go has become like, depending on the hiring manager's feedback can be 10 minutes. It used to be 24 hours. So that's like a really like key example of where we've had a massive unlock, but the quality actually has gone up as well because talent partners often don't have time to say, I've picked this profile because of X, Y, and Z.

19:20What do you think hiring manager? Whereas if you prompt it properly and give it guardrails, it can say I've chosen this archetype and I want to test your risk factor for domain adjacency, for example. and then you can really get alignment on what they actually want to see in their search. So yeah, that's just a short example of how we've gone about building it. So we did that and then we plugged that into the marketplace. Yeah, I mean, as a person who's been in recruitment sort of 25, 26 years now, like literally an old man in the industry. I mean, I can see some of this early work, which would have consumed a huge amount of time, massively compressed.

20:01and there's two outcomes to this folks. Firstly, you might be able to manage more jobs, which by the way, I don't think is great for internal TA, but it is great for an external agency, right? So I'm hearing a lot of positive feedback from third-party recruiters, agencies that can now run many, many more jobs and then increase the amount of revenue generating opportunities because they can absorb that work and convert it into revenue. TA, for us, we have a different dynamic. Do we want to increase the numbers of jobs that we're running per recruiter? Arguably, we know. We want to try and increase the depth of work, the quality of work, maybe per vacancy per candidate.

20:46But that's an internal debate we need to be having with the business. But this is an example of how technology is going to just basically change people's sort of, whether it's good or bad, it really depends on what position you are in the ecosystem. Okay, cool. Okay, Adriano, we can stop the screen share now. Appreciate it. This one-pager, and in fact, all of your one-pagers, where do you sort of store them? Are they freely available, or do you send them to people as a DM? Do they need to sign up to something? yeah they need to they need to put the one page the one page is visible for everyone it's on my feed um but the the asset behind it is um on my featured posts and pop your email in and you'll get an email with the marketplace and the instructions to download it onto your cord um so it's got the 24 skills bank and then you can use it in your environment and also share it with your team straight away it doesn't come with the connectors obviously but um most of them are ready to go just as they are um and you'll see which ones you can play around with i'd love your feedback as well um uh if you get a chance to to play around with them link specifically in the chat there adriano yeah sure around for it um and then and then people can take a look um uh click through and see where you're at when you click on the link by the way folks um make sure you open a new window new tab don't leave the show uh to see adriano's uh stuff um but yeah if you've got a chance to take a look um and give feedback live to adriano go ahead and do that He's a big boy.

22:24He can take the slings and arrows of recruiters scrutinizing the work. Okay, Ala Pavlova, great to see you. Very busy in the chassis. No, you keep doing. I love how you do that. Let's firstly look at the poll, by the way, folks. Ala and Adriana, take a look at this. I hope everyone can see the poll. It should be visible in terms of results, right? But again, for the people who are not watching this on Crowdcast or otherwise just listening, I'll read it out. What do you want Claude to do for you in TA? 38.7%. Oh, we have late runners making votes. I have to really lock the poll. All right, we've got about 40 % of people who voted saying sourcing is number one.

23:11And number two, very interesting, reports and intelligence. So 30 % of people want Claude to do a lot of data analysis. I wonder whether that is on market intelligence or even sort of performance stuff. Because I think I'm interested in both. Admin, we have about 8%. And assessing about 13%. Scheduling, interesting enough, it doesn't seem that we want Claude to do that. Maybe we think that we have an alternative solution, a vending solution to it. Same with coordination. And then other, of course, we have a bunch of comments, I think, coming through. But really interesting, sourcing is still number one.

23:50Recruiters inherently just want more candidates. It's always the case. Okay, good stuff. Okay, Adriano shared his thing. He's also going to leave, is he? No, sorry. His tongue's coming in. It's burning me. That's all right. I thought your mic dropped your way out there. He's done. All right, Alapablova, you've been chatting with people here. I think one of the questions was, look, do you want to see how a skill is sort of built from scratch? We have 15 people thumbing it up. So let's see a skill built from scratch. I've never seen this. Cool. Yeah. Yeah. So it should be easy. So let me share my screen.

24:28And also, like, sourcing is number one. I'm happy to comment on this one and show how you can do it. But there are some caveats to that as well. So let me do the entire screen and go to my clothes. So, you see my screen, right? I can see it, yep. Okay, sorry. You can go to the skills in different ways. So, you can start from the plus and you click skills. And from here, you either select something that you already have or you can browse for skills, meaning that there are different skills that are already developed for you. and you can type whatever you want to do sourcing maybe but sourcing you have to be careful it could be um yeah i have my own skill but there are tools that suggest that they can do sourcing like what are these are these third-party like software companies have said they've produced this yeah just random people have submitted stuff and it's like it's not really random it's uh approved to be, let's say, part of the skill library.

25:33And there are not that many of them. And so you see it's like official, the companies that are somehow connected. But yeah, so you can switch between the skills that are yours and the skills that you found in the library. So for example, I really recommend to check out the skills for marketing because they're really, really applicable for us in recruitment because they have like email campaigns skills. You see, skills are here. You can read them, what's inside the skill. And you can, let's say, copy and ask your quote, hey, I like the skill, but I want to modify it because I want to do it differently.

26:15We don't use Notion. Let's say use something else. And you can really adopt it and then save as a skill. And it will be yours. but it's already kind of big library of skills that you can try. But if you are like, okay, I really want to build my stuff like myself, different ways how you can do it. So let me go to yours. So you can, if you click on the add button, let's say somebody like me or Adriana is giving you the skill, you can upload it and immediately use it. There are marketplaces for skills, but I want to comment something. Please be careful because there are really strange things that can be put in there.

27:00It's like you have to trust the source. It could be things like, you know, mining something in your laptop or sending some information. So just be careful. You can create the skill from scratch, like just explaining the problem. And one important thing, one skill, one job. So your skills should be very, simple in terms of this skill only does read the intake and brush it up. This skill is only for scheduling. It finds the time in my calendar and schedules. So that's very important. You can write the prompt that will have multiple skills on it. That's how I did for my, for example, tone of voice.

27:43It has Alice tone of voice, my imperfect English, of course, because if you see my LinkedIn posts perfect English so no it's not me even not my AI so but I use multiple skills inside like for example delete AI words are inside my voice skill as well yeah so because it's stop press a second there so multiple skills within a single skill so this is like a nested prompt isn't it so basically I can set up use your example eliminate AI speech in any copy I produce and that's the skill in my library and then I can basically invoke that skill within a new skill I create and then basically pull it in okay good is that it for everyone by the way?

28:36I think so it does make sense I'm sorry I will go back so you can enter from here as well it really makes sense because you sometimes don't remember like you know there's complex things, for example, first you read the intake, but then evoke, wake up the skill that is going to create the sources strategy. So it makes sense. It's like a small little automations within each other. Another way you of course can create this cloud. But one of my favorite ways I don't know, Adrian, have you tried this or not? And have you tried it? I haven't tried it, but I really want to do it. I wanted to like track some I did today on LinkedIn.

29:15Is it good? It's good. So basically, recording your screen, be careful, of course, because it gives access to all your computer. And yeah, so what you do, like let's say you complete the regular task that you do. Let's say it's going to be sourcing. Like you open one website, you take this, you put this, or you work with your invoices or you work with Excel files, something repetitive, especially with reports. so you record your screen and then cloud building the skill out of it and test it what it was good at it completely repeating what you're doing this clicking so if there's something that you have to click many times take things from one place to another one this is like a perfect way to do it okay so let's say i've got a spreadsheet i manually check for data entry on that.

30:09I need to update it and just correct. Let's say I need to remove all the caps lock on all the text forms in the spreadsheet. The way you can do it with this is record your screen, do it once, do that action once. Claude will read that and then think, all right, this dude basically wants locates on all of the text. And then in future, you invoke the skill. And I guess you want, you've got to point it to the spreadsheet. How does that work? Yeah, so this is a next step that I really would recommend people. Before you start with anything like a chart GPT or Cloud, create a folder on your computer that will have the necessary documents.

30:50Let's say this Excel file will be in a special folder. And so, I don't know, maybe you download some CSV or anything. And you give access to Cloud to that folder so that the skill can be executed. Does it make sense? Yeah. So you basically almost create, it's not a true sandbox, but you basically create a kind of a pseudo sandbox and say, look, on this folder, this file, this is the place where you've got to just look at stuff. Please don't scrutinize, those are the spreadsheets I have that might contain information I don't need you to see. Or you can create a project. And I will, yeah, you see some of my private project, like anti-aging, yes, I do that too.

31:37I've had a collapse in the last 12 months. I look terrible. Like my bookshelf and stuff. So, yeah, sorry for like showcasing things as well. But project, it's like your friend that remembers everything you told your friend. But it's powered with, it has access to any skill. So, basically, you can get the skill in any chat in any project. and yeah, it's just like a folder with all the context and you talk to this folder and yeah, this is how you can also eliminate repetitive tasks. And I think I don't want to like talk a lot, a lot, but sourcing is one of the most popular requests. So let me probably just to showcase it quickly.

32:25Should I? Yep. Okay. So sourcing. um always be careful because yeah anything linkedin related um you have to be careful and i would not like recommend to use llms for sourcing in general for um for the purposes of you know in countries like europe with gdpr and etc so you you still get some some data that probably you're not supposed to get right so always be mindful about that and when you reach out to candidates always mention how to find their profiles it's like a hygiene um that's we have to follow but if you're curious to do maybe like some like um quick search uh and quick sourcing um for very niche and like difficult role or something like this um there is an easy way a nice way to do it um using connectors and people were mentioning exa in the chat um and it It has a free tier, but if you use it extensively, you can just run out of the credits also, like if you do a lot of sourcing.

33:34So it's not completely free. There are free web searching tools. I think Parallel Search is free. It's part of the Cloud library. But yeah, if you need something like a little bit like more deep, EXA is the one. EXA AI is the connector that you need to give access to at your cloud. So you go to connectors, you find EXA, and then you connect to cloud. It will ask you to create an account with EXA. And you choose the free plan. and just to give an idea a free plan is enough to source maybe let's say up to 50 hundred profiles a day something like this so I think with a free tier it's okay but if you do more so yeah be mindful and it's also depends what information you're asking for so you get the connector and then you go to uh the uh the search and here you can just say hey use exit to source um uh if you can do this so i can do the live demo that's that's everyone wants to because i don't see the screen so han tell me if i should do the live demo i don't know folks are you interested in seeing a live demo from ala let me know in the comments let her know in the comments good damn it um by the way folks um i really appreciate everyone getting active engaged in this um if you're watching this on linkedin um i'll also contribute we've got people monitoring the comments on there everyone's yes of course that's a rhetorical question folks yeah yeah i'm ready to show because i already ran the search so i just want to prove there's like hundreds of people watching this because you can always i don't trust the numbers but i do trust people's comments i mean that's definitely happening right so um okay what are we looking Can I hear, Ale?

35:29Yeah, so I already did the research on the role that's just random. Somebody asked me to make it as an example. So this is just a copy from the job description online. And what I'm asking is, like, use EXA AI. Find candidates only in the Netherlands. And add their titles, add their location. and also explain why you think it's a fit and also give me some gaps on the job description versus the profile information, the source you're getting it from. And it's also, if you ask, you can also ask to draft the reach out as well. And that's basically the file you get after you're sourced. And you can do, yeah, it's already did the outreach.

36:19You can also do the shortlist of candidates that you would like to start with yeah and they're the long list so basically that was all done from just asking hey this is the job description use X to find let's say 20 profiles add the reach out, add the where they fit and you can do some other things from there so I will stop sharing but that's the sourcing the easiest sourcing how everybody is doing it right now Okay, fantastic stuff. And tell me what the difference here between APIs and MCPs. So basically, APIs, I think we understand, right? Basically, it's a way in which different software systems, it's a protocol that basically communicates with different software systems.

37:09So what is MCP and why do we need an MCP instead of an API? Because you're basically connecting the tools between each other and you allowing the tools to talk to each other. So let's say you have your applicant tracking system. You can have API basically to try to write maybe some customized things, something that is only internal, available like some kind of only for internal usage. But with MCP, let's say you get your ATS connected to, I don't know, like greenhouse allowing MetaView to use the MCP. So both tools can use the benefits of each other. So, Adriana, I don't know if you have a different...

37:58Asking chat to you too, what the difference between API and MCP is, because it's nuanced. Yeah, they both allow them to talk to each other. That was my understanding. I thought MCP was... It discovers it can connect more and have them talking to each other, where API would be like unidirectional, but I'm not 100 % sure on that. Yeah, folks, API specific, Jonathan Duarte, he's coming in with some great chat there by the way follow Jonathan he's always hilarious online as well I mean they do it for the comedy as well as for the value guys okay you mentioned Greenhouse by the way and I should actually do a quick stop check because we have to thank our sponsors because I'm your co-host here without our sponsors basically Brain Food Live cannot come and do this every week and Greenhouse have been one of the amazing companies that have always sponsored and supported this newsletter and basically these webinars.

38:54They are, of course, one of the leading ATSs out there on the market, have been for the last 10 years or more, and they're constantly pumping out some really interesting innovations. I think they've got something really exciting under the pipe. In fact, I'm not sure whether I'm even able to say it, so I can't really say too much right now. But I do recommend you keep an eye on their announcements because they're going to do some, I think, very significant moves in terms of their product development and their wider product vision. Follow Daniel Chait, their CEO. He's a person that talks very publicly.

39:30He's pretty much a founder-led marketing style person that if you want to know where recruitment tech is going, someone you should follow. In the meantime, I'm sharing a link with you. They've published a really interesting guidebook on what questions you need to ask your vendors when you're looking at making big decisions in terms of software purchases. So do check that out. It's going to give you a bunch of things you need to keep in mind before you get bedazzled by brilliant demos and so on. Okay, cool. Right. What else have we talked about? We talked about projects we've talked about connectors we've talked about skills um how do we sort of actually go to you on this adriano because you're speaking to companies and functions um and um i want us to basically i want you to give us a little bit of a view as to where you feel what the the median company is right now and i'm assuming people can reach out to you uh are probably a little bit sensitive to the idea, hey, we need to move forward.

40:39So they're kind of maybe a little bit, I wouldn't say ahead, but they're above the median line, I guess. But where are they usually with the AI enablement in TA? Yeah, it's funny. I had it prepared just in case, but you've asked me, so it's awesome. We did a survey on LinkedIn a few weeks ago. We've had 162 self-reported filled in. Most of them were heads of, somewhat ICs, all in-house across various stages of business. And we, this is a self, we created this leveling. It's not a codified leveling that anyone agrees with, apart from what we've done with learning and speaking to people. So we did like four levels.

41:20So the first level was ad hoc. So it's individual use. You've got one super user, other people don't use it. There is a tool available to the team. Level two would be experimenting so there's been some tools tested uh methods vary but they seem there's some level of like sharing of knowledge and then the three and four are when you get to like okay things are starting to kind of be connected a bit so level three we defined as operationalized which is shared workflows and manual handovers and then level four would be orchestrated which is connected workflows um talking to each other with human oversight so that's like the you know the north star nirvana um yeah 162 89 were at level one or two so ad hoc or experimenting um and 69 were this was this was when we ran we did them initially on call and then we just did a self-serve survey and what was um interesting was some were had an amazing stack everyone had clawed everybody had like lots of tokens given to them, but there was no unified way of using AI and or shared knowledge.

42:27Nevermind the legal blockers for some of the larger businesses. The main thing was there was no like codified best practice of AI. Tools are there. People don't know how to use it. They don't know how to share that knowledge. So that was the main thing. And also when you ask them, what do you want to achieve with AI in the next 12 months? Not many have thought about how to measure its impact um so it's yeah it's they're quite far let's say we're very early stages based on so the message i'm hearing from you by the way again adriano if that report i can share that yeah i posted the findings early in the week but i'll share i'll share it again it's on it's on my feed there's no uh it's not hidden there's no guard there's no um what's the word i don't know the word it's not gated gate yeah share it and share it in the chat mate because people love to see that and the reason why that's important folks is because i don't feel as if you're behind ahead or whatever you know the worst thing you can do is feel this way uh understand that this is all a journey we're all on the journey and we're going to be you know up and down forward and back there'll be a breakthrough and then you know the the method of progress is really this jagged line that we're all going to get towards so no panic although let's understand the urgency because we have the opportunity to get significantly better and faster, we should grab it.

43:46Yeah, good advice. Everyone's feeling the same, you're right. They're all like, oh my God, I'm losing out to my peers. But you're right, that's not the case. Okay, I love the progression. Al, I want to get your take on this as well. And here's where I put this on a slide deck recently. And this again was from the dialogues I've been having with different types of companies. and I think they map quite accurately to Adriano's four-level tier. So number one is basically individual experimentation. So we've got individuals that are using Claude, you're doing stuff like we've just seen. So maybe you individually might be quite advanced, but you're not coordinating with anybody else necessarily and we might all independently be doing different things.

44:31So that's a natural state of phase one of AI adoption. So phase two is when you are starting to share stuff. So maybe there's an internal marketplace, perhaps you kind of all share a bunch of things that you've done, and you have a weekly, hey, what's working this week, folks, and you start selecting certain things. We've seen a library of skills, it seems that those can be shared internally within team, right? So you can imagine a scenario where, oh, actually, Adriano's got a better way of doing this than I did. I'm going to delete mine, I'm going to adopt Adriano. So stage two is that coordination.

45:08Stage three is when maybe you start to say, right, there is like a single winner, so to say, on this thing, but we're going to roll it out across the entire function. and this is codified as to how you use it. You don't actually do more vibe coding. This is the way it is. Maybe it even gets rebuilt by technical people that can look at all of the issues that you're doing, make sure the guardrails are there, and then advance it and scale it out. And then sort of tier four, I guess, is along those lines. I'm not sure whether in my framework it fits into tier four, but maybe it goes external at that point.

45:47Who knows? you contribute something that is a little bit more open source and say this is how it works but yeah anyway Ala Pavlova your take on things and when you're training companies and training people what's the sort of level we're at right now and where do you think we're going to be in 12 months time if we progress the way that we've been doing so far yeah thanks so much I think shareable resources internally is really really powerful especially the onboarding part or I know that some companies they use Gemini notebook alum for like a library so anyone anyone external can just join and read a lot of files but you know why it's so powerful because you can talk to the documentation so you remember the times when you join the company and they have like like five books for you to read but like you can't talk to the books until you read them.

46:48So I think that technologies are super helpful. Let's say you create the project in Cloud. You upload all the context. This is how we interview. This is how we decide what this good looks like. This is our outreaches. And then I, as a newcomer, can just ask Chagipiti instead of being like, you know, flushing, like with red cheeks coming to somebody like asking, can I ask you a stupid question? And this kind of thing. So I think that reduces. it was notebook now it's called the Gemini notebook but similar works with projects although one thing if you're a personal user and you create something that you share please be careful because there are a lot of things that you share that goes to the internet so I don't know if you know but if you like the chat let's say if you give thumbs up thumbs down in clouds the whole chat is sent to the customer service team to check it out.

47:48So whenever you share something, yeah, Adriana, I'm happy. So you're learning from me. So please, whenever you - I'm going to start deleting chats. Do they, if I delete them, does it delete on their side as well? I don't want them knowing about - Yeah, I don't think it does. I don't think it's even possible to do. So this is a really important point that's worth underlining. I think Shrini mentioned it in the chat as well. We in recruitment, I think we're quite sort of new to all of this, But we need to start learning about what it is to have local AI, on-premise AI, maybe even air-gapped AI at some point.

48:22Because we are using intelligence that's not owned by us. It's actually owned by a company that's a profit-making business that, in fact, may turn into a duopoly at some point. At which point do we want to have that? And secondly, we've already seen evidence, and Adriana, you intimated about this in guardrails. We've already had evidence of things like sort of refusal to do work because of guardrails that may be right or wrong, but who set those guardrails and who appointed you as the paternal sort of overseer of what's good or bad. Number two, there's also model rerouting, often undisclosed. So you might be a paying customer to a model that you think is the best.

49:08But you know what? There's evidence that they've reroute you to something that's cheaper. And they've given you like a poor service. Okay, what's our take on this? Thirdly, do you want to put all of your IP into a service that presumably will be providing that intelligence to your competitors as well? You know, if you are saying, hey, I'm a recruitment company, I want to open a market in this territory, here's all of my strategies to do that. You put it all into a cloud-based vended solution online called whatever it is, OpenAI. cloud or open ai whoever it is is going to learn how to all that information it's going to provide you with the best answer however competitor number two might ask the same question what it's going to do it's not going to withhold the information that i've put up there um it's learned it's developed the intelligence has improved i have trained it and it's going to give value to your competitor uh so at some point folks we need to move to open source on-premise ai i hope that's the future we are going to go toward but we need to support it whenever we see that argument this is one of the reasons why we need to endorse it okay run over folks I'm sorry I think I got you very excited about the chat privacy things when you share but to finish the thought when I see us so one important thing that I would encourage everybody to click all the buttons like know what's the artifact you know what is you know why what's the difference between connector and plugin and that's going to something that I think we need to all you know attest the tools no matter what but also I would like everybody to encourage to use Cloud Code.

50:58Please do not think it's only for developers. Just try it. I make such a funny nice dashboard. Like I created attention dashboard for myself because I have I have really a lot of emails, subscriptions, newsletters. I just can't get my attention focused. So I think if I didn't have ADHD before, so I may have it now with all the things that come in my way with information. So attention board helps me to focus on the things. And it's very easy to build with cold code. Go to cold code, explain what you want to build, and it's going to vibe code for you. And you can ask questions. So please don't be afraid to click on that button.

51:40Yeah. And folks, would Cloud Code be a very good sort of episode two for this? Let us know in the comments whether that would be a good exercise because I immediately feel a bit terrified. But then I kind of am aware of a few people that I don't think are any more competent than I am. And they dive into Cloud Code. They're producing apps. They're producing bits of software. And this is the vibe coding sort of journey we're in. And I think vibe coding does have a role to play, particularly when it comes to personal productivity, definitely when it comes to personal sort of marketing and content generation.

52:16But also, I think, again, we tie it back to eventually to on-premise local localized AI. And then suddenly maybe you can vibe code with a lot more confidence because basically it's stuff that doesn't leave your computer. OK, lots of people saying yes to this, which is great. Yeah, they have to do it. Before we go to the, I mean, coming immediately to the end of the show, which is super fast, but we've got to talk about sort of AI agents here. So all of the things we've talked about so far, it's still very much a human being making operations. And I'm like clicking the buttons and saying, I want you to do this, that and the other.

52:56where do AI agents come in and how do I let's say there's a process there's that sequence on Adriana's one pager how do I set a digital hung on to do that shall I go on that I think agents I think we can get wrapped up in the meta language of skill uh agent workflow like it an agent is just a collection of tasks uh where it can make some element of judgment across those tasks to finish that task there are multiple stages right so skill would be like multiple prompts and it's it's fixed an agent will have some kind of decision tree but you've you've coded that in already but ultimately it's just a more complex task you're trying to get it to do and you can literally like Allah was saying it's it's go to go to flawed code or even even co-work it still does vibe coding for you um and tell it what you want to achieve no matter how ambitious it is and it will build the agent for you and then you built an agent so like it's it's it's really like the meta language is almost irrelevant now it's like okay what's your task what's your objective give it the theory behind which to achieve that objective properly so like a lot of the tools that i've seen are excellent at finding let's for example, sorting, because a lot of people are talking about it.

54:24They're excellent at finding those 20 profiles that are really high signal, that keywords are really updated, right? I'm like, I'm in fintech, I use React, I'm a senior principal, or if it's sales, like I'm in president's club, whatever it may be, they're easy to find against the spec and sourcing tools are good to find that. But they're getting pounded by everyone, because they match a brief. And so yes, you want to find them quickly but they're going to be hard to convert a good recruiter knows all right they work with another 10 people in that business they haven't updated their profile ever i'm going to also ping their colleagues who don't mention anything on their profile so i think like teaching the agent this knowledge is where you're going to get the biggest unlock because the build is no longer the barrier so like you like so that that's my that's my take around agents it's like go there i want to achieve this multiple steps walk me through it great technically it's possible, but what's the theory that underpins that quality output?

55:20Yeah, really interesting. From my side, though, I'm an ignoramus, right? So I've actually deliberately kept myself ignorant purely so I can ask dumb questions with authenticity. This is the sacrifice I make to the community.

55:38So where do I find this AI? It's good framing. I love the framing. where do I find this AI agent like I'm on Claude I want to create an agent that does all this like how do I do that yeah let me unmute myself so when people talk about agents so I also would like to see what they mean by that because automations that Claude allows you to do immediately it's a scheduled task and I'm curious if anyone tries to do a scheduled task because this is something that's like I think to build a real agent, I will need some time. So I think I can't be put on the spot like this because I have to think of what problem it needs to solve, what access I need to give it, because if you want things fully automated, so, for example, you want a scheduling agent.

56:31So this agent needs to have access to your calendar and your probably Gmail, and maybe you also want to give access to something else where it's taking the information. So I would love to hear in the comments what type of agent you want to have because, again, agents like, let's say, like ChargPT, they have DOTS. If anyone tried DOTS, I don't know, Adriana, have you tried DOTS because you just came back to ChargPT? No, I don't have the enterprise plan, apparently. It's just for those. It's annoying. so yeah scheduling uh agents i see people wants to so um uh they uh i think clothes is soon going to have their own agent so i don't know for which plan so we will see that as well but if you're looking for um agents immediately to try and to test it i would suggest maybe people can play with crockbot although i i know that's like always controversial everything related to Elon and everything.

57:34But I would definitely, I saw a lot of agents' automations built as Crockmots. But if you go to Cloud and you go to Schedule Tasks, I can show a very quick example of one of the two tasks that I have. It's related to creating content. Just a quick share of the screen. I see people asking for agent built live so we can do it I mean if you allow I wonder whether you can live stream it properly though but it's a bit time so you have to so one of the things like I use my tone of voice as a skill and I have this scheduled task I will show you, you go to schedule you click on a new task and you create it and then it runs automatically like I'm looking for let's just see who is hiring like daily market ideas but for example one for content let's say I want to create content like regularly posts on LinkedIn but I want it to be already like written in my style and this is doing for me every day it generates three posts like it's using Excel to scrape information from the internet to find what's good, what's not and so then it's suggesting me and then I have every day some ideas to post.

58:58It's automated. And my next step could be that I just go and connect it with my social media and get it out without probably me checking. But I will not do it. I will check still because I'm a control freak. No, I think there's a couple of things. I mean, we don't have the time to do it all here today. But definitely, let's do how to build an agent or how to use an agent. As you mentioned, GrokBot's out there, Muse is out there, Dot is out there. Some agents have vended to us, so we may be able to crack through and use some of those. And yeah, how do we use Claw Code? Maybe there's another fork in the tree.

59:40There's another episode. Folks, before we leave you, there's a couple of things that we shouldn't forget. Firstly, Bring Food Live, of course, has to come off air. It's very frustrating sometimes when we have such a great conversation. But we are over time already. So one thing that I think is really important, though, is to make sure that BrainFood not only starts conversations, but makes them continue or enables them to continue. And there's no better way to learn than to connect with other people who are on this show and make sure that you can extend this conversation when we do have to go off air.

1:00:14So the way in which to do that is very straightforward. Just grab your LinkedIn URL, share it in the chat stream if you're watching this on Crowdcast. If you're watching this on any of the LinkedIn's, just do the same from there. Put it in your LinkedIn URL as a comment and then just connect with everyone who's just done the same. Basically, you can take as read that there's permission for you to connect with those people once they've publicly shared this thing. And of course, one thing you will know is that everyone who's done that basically is into this idea. So go ahead and break out and have a separate conversation.

1:00:48Create your own group. Talk about it. Hopefully it'll accelerate your learning as you go through. There's no point in trying to, well, there's point in trying to do it yourself. Of course there is. But there's also value in just going faster with other people. Okay, that's it. Adriano Herdman, Ala Pavlova, thank you so much for your time. Brilliant. I've learned tons on this. So fantastic for you to share your knowledge for us. Really do appreciate it. It's also given me a few ideas as to what to do for early next year. We're already looking into how that looks. The Brain Food Live, of course, we're in the final sprint.

1:01:26We've got another, I think, six more episodes to go. If you're interested in any of those, please take a look at the channel. They're all up there now. Most of it is going to be look back review of the year. So we're at the point now where we want to just say, okay, what the heck has happened in recruitment in 2026? And the last five shows is going to be focused on that. So we're going to talk about sourcing, sourcing agents. We're going to talk about talent acquisition, talent attraction. We're going to be talking about talent ops and workflow automation. So a lot of this stuff will come back in.

1:01:57And then we're going to have a big overview at the very end where we're going to get some thought leaders, quote unquote, to say, okay, what is the verdict on recruitment in 2026? And that's the last show of the year. So make sure you join that. You can sign up to any of these shows now. Next week is going to be GEO, what do we know? So how do you show up on artificial generated summaries? I don't know whether you know, folks, but basically there's been a massive spike of the amount of times in which your inquiry on Google is triggering an AI summary. I think previously, six months ago, it was something like 30 % of your inquiries would produce it.

1:02:38Now it's up to 80%. And you can imagine in the future it's going to be 100%. 100%. So people are going to search, put a search in there, who is the best company to work for in this sector, bang, you're going to be an AI summary up there. And how do you get your brand in there? How do you get your job in there? How do you make sure the words that are being commented on you are what you want the AI to say? That is something we're going to talk about next week on Friday. So make sure you tune in for that. Okay, that's it, folks. Enough for me. Thanks to Ala, thanks to Adriano. Thanks to everybody.

1:03:09We'll see you next week. Thanks so much. Thanks. Bye. Bye. Cool. All right, guys, that's it. Well done. So many questions. No, and so many people watching it. It's legit. There's tons of people. This is why you can see on the comments. So hundreds of people watching this, which is fantastic. Shout out to people like Joe Atkinson, by the way, and Andrea Lungulescu, I think, also pumped this to help get people signing up, which is fantastic. And yeah, thanks to you two for coming in and sharing the knowledge. It's been brilliant. Of course. Thanks so much. All right. We're going to end the session.

1:03:51Appreciate it so much. All right. You have a good weekend. I'll see you soon. Thank you. Good weekend, guys. See you later. Bye. Thanks so much. Bye-bye. Bye. I think you're still recording.