The 2026 Meta Playbook: Creative Formats, Personas & Scaleable Structures

0:00We're doing a lot of mashups now that are specific to certain personas and then we are doing lots of variations through like more vo or sora based hook variations

0:15yeah where do we want to start i want to start on creative i guess yeah let's just start on what's working right now some some just top level bits that are very actionable uh that people can take away um so one thing and this just like seems very obvious but um i haven't really spoken about much on on the podcast at least in q4 and i guess it kind of gets forgotten about sometimes now but like mashups um so i know that sounds very very obvious but really starting to bucket creative assets by persona because we're doing a lot of like mashups now that are specific to certain personas and then we are doing lots of variations through like more vo or sora based hook variations so again just a more one that i wanted to raise because we we're seeing like a lot of top spenders coming through in multiple accounts that that are feeding into that and i think a lot of the time we're trying to like reinvent the wheel and new kind of audacious big swings or using ai to like recreate concepts when really like you're sitting on a gold mine with the assets that you have available to you especially now like for all those brands that have been deep in this creative volume game for six to eight months like their asset libraries are going to be huge like if we look at some of our longest standing clients there's thousands of videos in there um i think a lot of that is an exercise in like good file management it's the boring stuff because it's hard if you've not done that to search through obviously you can get support from like frame ai and things like that but yeah i think this makes a lot of sense like you say but booking by persona how are you approaching the is it is it is it is it building a script first and then finding assets to piece to that or are you approaching it from a different lens yes it's kind of modular in a way so um we the the key is it's like we're not a lot of the time this isn't just like sitting within your cloud storage or frame or whatever or whatever you use it's the in its simplest form of say you don't have that like back-end asset management setup literally going into the ad account and especially if you're using naming conventions by persona, like one that we did is we looked for one of our accounts, we literally just sorted by persona top spenders over the last 90 days, downloaded all of those video ads, and then made obviously all of them follow like some form of kind of, you know, an iteration of an ADA framework, let's say.

2:45then from say those 10-15 ads there are hundreds of variations that you can do once you file those into a modular approach so obviously hooks, problem agitation, solutions etc etc. So that's the way that we'd approached it through that lens. I will say that we are working on a lot right now in terms of that asset management side um obviously we've we've recently hired a director of creative ops who's working a lot on this in q1 so there'll be a lot more value that we'll be able to share there in terms of how we're structuring that going into uh into the the the second half of q1 nice yeah i think that's great as well because it's looking inwards and like doing more with less um and cost of creative just like try to drive that cost of creative down but it's obviously easy just to continuously add resource it's like if you can make that stretch further through those through the through the asset library makes a lot of sense yeah absolutely and i think the key there is that we're we're we're utilizing new hooks to obviously get around the creative diversity piece or similarities so that's just something to consider as well i would say like the most consistently top performing founder framework that we're seeing working in across all of q4 and i've seen this from multiple of our accounts now is the um is the founder podcast objection handling ad i knew you were going to say that i'm not as close to the data but i've seen multiple of ours i've also seen a number of other other brands doing the same where it's just direct call out of biggest objection in the hook right yeah exactly or like it opens with a controversial statement linking to the problem so i won't say the brand but the hook was um would you let your kids start um the day guzzling a can of coke yeah um but then it's like the skeptic comes in and starts pushing back with core objections and then the founder is is uh is there as the the person to to combat those objections.

4:57Conflict, I guess, which we've also talked about previously working well. I think the founders' concepts, again, there's probably a number of brands watching that have tried it and shelved it, but it just always works when you get it right. I would really recommend as well, if you're the founder listening, you're working with the founder, someone on your team, do a scripted approach to that, but also do an ad hoc version where you are just throwing objections at the founder and just seeing what you get back. Because we did that with Dan from Heights, who was on the podcast a couple of weeks ago.

5:34And we got some really nice moments from just actually going a little bit more spontaneous with it. And layering in satire as part of that worked quite well. He brought a bit of a sarcastic edge to some of the objections, which I think just helped drive that controversy as well. Makes sense. So another iteration format that we're starting to see really work is essentially like an X. What do you call an X post now? I still say a tweet. Yeah, it's a tweet. Yeah, X just sounds weird. It does, doesn't it? An X. But yeah, it's essentially where you have a headline, which is an X post or a tweet, but then you have the video underneath it like someone had posted a video onto x so again it's like a if you have a best performer that's working within the ad account take that best performer put it into an x video with a new headline that sits above it so it's still a video ad but you're incorporating it into a slightly different variation of that format it's interesting i imagine there's other ways to do that as well with other other sites other visual iterations of that exactly i feel like there's just a there's just a forever evolution with when it comes to those native formats i saw a lot of those re working well again over q4 and into jan actually we've done some of the um the classic black friday email statics what note from founders if you're a wellness brand you can reuse that format through q1 it's like it doesn't need to just be used over the Black Friday.

7:15It's just, again, the copywriting and the intention and the message needs to be right. Definitely seeing that work well at the moment. Yeah, absolutely. And then the final one is green screen reaction ads. So it's essentially a creator that is overlaid on top of another UGC ad that is reacting to the UGC behind them. And I think this was, I haven't seen these for a while, but it's something we started doing again. I know we, you know this is something that has been done a lot in the past but not for for i haven't seen it for a period of time and um that's working really well so it's almost the individual that is watching the creative behind them but then reacting to it in the form of a testimonial so almost like this was the ad that i saw that took me on the journey to to solve x problem makes sense oh yeah they've a lot of brands were doing them there's a lot of volume of that in early 2020.

8:09It's just not seen as many of them recently. Again, it's quite easy to do. It makes assets go further, it makes assets work harder. If you can just get a couple of creators to churn those out, it makes a lot of sense. Where do you think Hyprod's going to go next? We've obviously got these proven, we've got the founder, we've got the podcast, we've got the ring doorbell if you class that as Hyprod. Are there any innovative bigger swings that we've seen there come through? yeah so we again i'm just so bullish on like taking youtube formats and applying them to high prod so there's a few that we're doing there right now one is have you seen well actually it's more it's youtube and tiktok but have you seen the subway takes guy um i guess you're not on yeah it's the ghost not even a ghost he's not there um so it's essentially a guy that like sits on the New York subway and someone comes with like a controversial take and he says like 100 % agree or 100 % disagree and they kind of just like have like a funny interaction about it it's terrible my mind went to sandwich when you said subway I haven't seen the subway man that makes sense so that's one that we're testing again it's just like an evolution of like the agree to disagree concept which was an evolution of like objection handling podcasts um so we're we're leaning into that as as one option i think another one is like thoughts from the bridge again of really famous or organic tiktok account which essentially is like a person holding an old payphone and the camera is like from really far away and it's slowly zooming out and it's almost like fly on the wall we'll fly fly on the wall style yeah where they are kind of quite authentically and vulnerably uh opening up about something we're seeing that working for like more um emotionally driven brands uh where there's like a clear problem solution and the individual is like talking through their problem and then into the solution um i would say like yeah generally they're two of the stuff two of the things that we're testing right now but i think the like the dialogue heavy I know this your thoughts kind of going down that how do we how do we break through AI almost in that high prod style yeah yeah like what's the what's the what's the what's the next evolution of it like where does it go I think I saw Taylor Holiday put a tweet out about this I think but about the concept of like live streaming and where how big he thinks that's going to be as like a medium of creative because that's impossible to well hard very very hard to ai and you see like people like that speed who is like the pioneer of live stream content and how that becomes like millions of clips and like he believes he was saying that he believes there's going to be a brand built through that lens like that's going to be their primary means of content distribution and then that's going to become their strategy down so so is that the end state of like the furthest possible you can get away from ai like live in the moment content that forces you to be creative as a business not just um in a process which i thought was quite an interesting thing to think about um but i'm sure there'll be other big big like there's what's the next podcast there'll be something this year that's the next podcast I think it's just when we find that yeah when yeah I think like it's it's it's really difficult to say but I do just I still believe like how if we can through the lens of like emotional customer testimonials I still just see that being really hard to like replicate yeah I also think the one an example of that is that that Brez had that did the rounds of them interviewing customers on the phone like we've reproduced we've we've iterated on that a few times for clients i think that's a format that again if every brand got it right it would probably work for most businesses yeah exactly especially like problem solution first um brands so yeah i do think it's like you want it you need to be careful not to take too many big swings and try and do just be be throwing out crazy concepts and focus on those fundamentals like we always say but interesting to see where it goes yeah i still think we're like a i still think we're a i I think UGC is getting really close now.

12:37I think high prod is, although from a visually, it's very close. I still think there's a disconnect there from it being like dominating and driving performance over high production as a mechanism. There's like elements of those high prod ads that will be driven through AI, such as B-roll and things like that. But like Arcad's 2.0 is a great example of like they're really getting close now and I'm quite impressed with it. Yeah, I think it's getting... I've had a very... I think this is probably a whole episode. We should probably do a whole episode on AI tools that we've tested and... Well, it did.

13:17We did that once and it did really well. Yeah, we should do it again. Yeah, I think it's close for anything that's still quite formulaic. I think it's not close for TikTok shop, long form narrative storytelling, but I think that is hard to replicate, but it's going to be an interesting path this year, see where we get to. Yeah, you make a really good point there. Like those native TikTok shop style ads, I think again, like that stuff is, the way it's produced is very difficult to replicate from an AI perspective. It's not that like single creator one shot where they're in front of camera and usually it's like, you know, we're camouflaging it through more overlays and things like that.

14:04It's almost the rarity of it makes it more authentic. Yeah, definitely seeing all the demographics. I see that whole AI creative being, I've already seen it in a couple of accounts I've audited where if they're selling to over 45s, they're just, they're using a lot of AI generated UGC because they just can't tell the difference. Yeah, it's a good point. But as you get younger, it becomes more obvious. Any more creative bits? No, let's jump into some AI stuff. It's a good segue. I wanted to just touch on creative testing first, because I think that's a segue from this. I don't want to go into, I feel like we could go into a complete black hole around what's the right creative testing set up in accounts.

14:43And to be honest, we do it differently in accounts, depending on where's our constraint. Is our constraint we don't have that many assets and we're trying to learn and we're trying to drive better feedback to create better output? Or is the constraint that we have so many ads that we need to be careful not to spend too much percentage of our budget in testing and therefore inflate the overall CPA of the account. So you can do that ABO, you can do that CBO, you can do it ASC. There's so many ways to cut that. Probably the number one question that we get. It is. I'm a big fan of ABO, but I think CBO is very useful when you want to be very intentional with, I'm only going to spend this percentage of budget I want everything to take a minimum spend to give me a signal.

15:32But we prefer ABO by concept. But one of the things I'm definitely noticing is promotion of assets to core campaigns being a challenge at times. Like something can perform amazingly well in an ABO testing and then you'll spin it into an ASC or whatever that core driver of scale and performance is. And it will either not take spend or just won't pick up performance like it was previously. So I think it's really important right now to still be trying that. And obviously, hopefully that doesn't happen in the account, but still scale up anything that works in testing. We used to do this a lot on TikTok.

16:13We'd scale individual concepts. You want to be maximizing that horizontal scale as well as the vertical scale in your call campaign because it's proving to be quite limiting. and it may not be a problem like right now because you might have an asset that is like just taking all the spend in that ASC but I think you can get into a position where you've got spend centralised so much that incremental reach starts to become a problem. Just before we move on because I know there'll be some people the next question off the back of that that feeds into what you just said is before we're like horizontally or vertically scaling how are you deeming a scaler?

16:52I know that again differs brand to brand, but if you were to generalize that to some degree, what should brands be thinking about there when they're promoting winners?

17:06Significant spend. You want to get significance. So it needs to be on or within a certain percentage of your CPA target at a significant spend, but that significant spend is relative to your AOV. so we'll usually set a daily budget and let things run for a specific 3 day period but the budget will be set relative to the AOV and then we can make a decision on whether it's it should be promoted or it should be scaled up or whether it should be just left like maybe that performance needs to maybe the performance is tending towards the right CPA on insights and therefore we may leave it for a few more days see if it reaches that that point um but yeah that's the that's the approach we take um i think also as you there's a lot of nuance in account decision making like you i think once you've spent a lot of time in an ad account and you understand a brand you can start to look at leading indicators of performance and make quite informed decisions around whether you need to spend money on the asset anymore it's like if you put something in after like a day or two it's spent a couple hundred quid and it's meaningfully lower CTR and like there's a couple of correlating metrics that like are so off what you know to be good in that account you can make faster sharper decisions but.

18:32So an element of intuition almost. Yeah I think there's a bit of intuition in ad accounts certainly as you get more volume I think what that shows though is that like ad set wise and campaign wise there's an element of like delivery by persona and that whole Andromeda piece either meta still isn't able to effectively deliver in an ASC across multiple assets consistently which is sometimes the issue hypothesis there may be like it's centralizing all of the spend to the absolute best performer but it would actually be but there's actually like multiple in there that may reach multiple audiences I don't know if that makes sense so your ceiling would be higher if it was sharing spend across because you were reaching two completely different audiences in that campaign so what this has given more and more rise to for us is just ASC by persona scaling campaigns by persona that back down into your testing strategy if those personas are super clear and you've got consistent winners in both Right so persona on a campaign level I think that's when she's starting to push we're obviously working with a lot of brands in the UK that are pushing some significant spend for this market um and when you look at the ad account it's like if if a if another agency audited it they'd be like what the hell is going on in here like it's an absolute mess there's no no best practice like but we need to do it because we need to there's there's metrics we need to ensure happen today to drive performance two three weeks down the line and a lot of that is around new audience reach frequency cpm incremental reach um and i I think Facebook meta isn't great right now at scale if you have everything centralized to one campaign because it can often just force spend to one persona.

20:26So that's just something to test out. If you're hitting that incremental reach limitation, you're pushing multi-five figures in spend, you're struggling to reach that next layer, I think really think about your campaign setup through the lens of persona because I think it can help you get to that. If you've got a good creative strategy that backs into that can help you get to that next level of volume. Yeah, that makes a lot of sense. I think the key there, as you said, is like taking a step back to the route though, which is actually ensuring that there's real clarity around those personas. Yeah.

21:02We're speaking about those personas and how that feeds into the strategy. Yeah, there has to be, I think there has to be enough differentiation. And that drives the campaign setup. I do think there's probably an opportunity in some accounts to even test it, test it for like the same persona, but just different, like very different vehicles. And a hypothesis would be that that may work as well. But I do think there's still leverage in that account setup, especially as you start to spend more and your volumes increasing quite a lot. So worth experimenting with another quick hack. Shout out to Ben, who actually put us onto this, who we work with on a few accounts.

21:46Bidcap. He gave it the name the Bidcap Graveyard where you put all of your past top performers Bidcap relative to your CPA and just leave it. It's not very consistent but what it'll do is in periods of peak demand a lot of the time it will pick up incremental volume and it'll spend and it may spend a lot. So weekends paydays uh it can just capture at quite a fixed cost a fixed rate low risk um extra spending volume that you just wouldn't have captured otherwise so that's worked really quite well in in january for a couple of our brands since we rolled that out and implemented it so it's definitely worth testing so you see that working better in what peak periods for the yeah because you set a cpa a cpa bid and meta will only spend if it's if it can achieve that target for you um or it will only bid in the auction if it believes it can reach that target.

22:45And therefore, as demand in market increases, and that has an element of cyclicality, evenings, weekends, paydays, you'll see that campaign spend, but it'll often not spend anything outside of those periods. It'll spend very marginal amounts. So it just seems to be working. It seems to be a bit of a quick hack for a bit of extra volume, so it's worth trying to roll that out on some accounts. Yeah, no. I've been watching. Is that like new customer acquisition? Yeah, it would all be, it's very, we're proving to be quite incremental as long as you set up exclusions, right? I'd say exclusions is a good topic actually.

23:25That's one of the things that I see when we're auditing accounts, just people aren't intentional enough with exclusions. Meta has become so much better over the last 12 years, 12 years, 12 months. at delivering and finding new customer but only if you set it up to do so so like make sure your event match quality is really high make sure you've got all your cappy set up make sure you've got engaged customers defined as broad as possible so instagram facebook website make sure you've got the same with customer list and purchase all the data flowing in because as soon as you start to let existing customers leak into campaigns like it's an easy way for facebook to hit a cpa target sort of just spend a lot of budget towards that.

24:10Yeah. So we've seen that be a big thing that people need to tighten up. So yeah, but if you set that up correctly on bid cap, we're also seeing cost caps in accounts alongside a core ASC setup be quite incremental and non-competitive as well. Like we're having both campaign setups run alongside each other, haven't been too cannibalizing. So I think that's also worth testing if you've got the volume. I was going to say one more thing that just came to my mind. Just entered and left. It'll come back. It will come back. I guess just why you're thinking on that is good opportunity maybe to just talk off the back of what you were saying about personas.

24:55Just if we're taking it back to the actual creative strategy part of like how I'm just maybe thinking slightly differently about persona mapping and taxonomy. Yeah. Because I think especially when you're driving like real scale, one thing I, the mistake I see with brands that are probably spending like north of 500k a month, say, still like isn't hugely significant at that level. But I would say that that's where we need to start thinking about layering personas in different ways. So we're using like this, I'm starting to think through the lens of persona strategy in like this three layered taxonomy, which is the macro persona, which is the who.

25:36So the type of person or the state that we're speaking to. So that's where your more standard persona table will come into. That's holistically sits across multiple micro personas, which is the middle layer, which is the why. So this only exists conceptually, essentially. it's the dominant motivator or the tension inside each of those personas so an example of that is like you have your macro persona let's say it's like a a protein um supplement so one of them is the you know one of the personas could be the the protein gym goer that's uh the and it's the male protein um gym goer that's the macro persona but then the micros would be going into slightly different layers.

26:23So for example, one of them might be the taste realist. So you're breaking that into primary desire. So enjoyment without compromise, the core fear, protein usually tastes grim. And then the belief, which is protein powder usually tastes like crap as an example. But then another micro persona might be like the identity keeper. So the primary desire is the core fear and the belief are different so even though they all feed into the same macro persona it's about how how do we split off the different belief systems core fears and desires for those those almost subsets of that macro persona so we we can when we are labeling our creative by that we can speak to it slightly differently and as a result it's going to help with incrementality because we are speaking more specifically to that persona in different ways versus having like a blanket way of speaking through that messaging.

27:25So that's just something that I'm starting to think more through the lens of when we're driving more significant scale. That makes a lot of sense. I think intentional persona mapping is definitely going to be really big over this year. I think we were obviously rolling it into our process through that work. How are you thinking about prioritization across persona? So I think that's a really important part of, well, prioritization is a really important part of any strategy, but especially creative strategy. When we're building it for a new brand that's maybe coming on board and we map out and we've got four personas, how are you prioritizing across those?

28:04Is it just like data points, volume of data across the research? Yeah, it's a mixture really. So we obviously have, we break our research into market, audience, competitor and organic. And off the back of that, we then that funnels into strategic opportunities found within each of those sections. So it almost filters out all the crap and then builds that into prioritization. So that is a, off the back of that, we will build our persona, personas. and then all but all of those strategic opportunities then we can use that as a data source for ai and and that by the way feeds into customer reviews it feeds into you know market analysis etc etc all of the we we use ai to like scrape reddit as well that will be specific to different pain points personas um motivators um and and ai does a really good job now of being able to consolidate which which are the priorities and which aren't i mean in its absolute simplest form if you weren't doing all of that you could literally if you've got significant data in terms of customer reviews like we we we run this for heights and they had thousands of reviews so we were able to like literally take those reviews and it would give it and then cross compare based on the personas that we gave it and it was actually able to just say it essentially labeled um the volume of reviews that related to each of the personas so even through that in its simplest form you can get to a good level of prioritization i think it's that but then there's also as you said from a media buying perspective like a level of intuition involved as well yeah yeah from the context that we've experienced and from the brand side exactly makes sense um there's two interesting new sources of customer research that i was chatting to a couple of people about that maybe you quite quick wins well one's a relatively quick win for brands that aren't already doing it one's a little bit more involved we've touched on it briefly in previous episodes but i think both um could provide a new way to uncover a lot of insights i was speaking to someone the other day who had implemented a post-purchase survey on the check on the checkout completed page that was like an open-ended question that popped up immediately as you landed and just asked them what triggered you to purchase today.

30:31And they'd used that to validate a lot. They'd used that to just dig into, well, the triggers firstly, but it uncovered a lot of insights. Was there like an incentive attached? No, it was just, well, I actually didn't ask that question thinking about it. I think you probably could, but I think it's because it's so like direct and so quick after purchase. I imagine you just get quite a decent completion rate at scale. yeah yeah they'd use that to rebuild an idea of what's the reason why someone to just it just changed the perception on the trigger of the of the purchase like they had an idea about like the the person who who bought their products and it was quite high price point it was often an upgrade but the trigger was often experiencing specific faults or problem with the like uh the alternative cheaper product and that was coming up so regularly that they then were able to kind of call out those really niche nuanced examples rather than just the bigger overall problem.

31:37I thought it was just a really interesting like simple quick hack to get new fresh data points. I think triggers are quite important. They're quite like, it's a bit different to like pain point. It's like you can have a pain point but then there's something that almost triggers it to be greater than the the cost of the product that you're selling if in in some ways from like an impulse perspective um so i think triggers are quite an intentional word i think you'll probably get quite a lot of learnings there if you if you're implementing that on most brands yeah it's so true i think yeah we don't we don't speak about triggers enough but as you say it's like there's always some form of trigger whether that's emotional um whether whether that is based off of some form of action failed solution like and and capturing that in the moment is really valuable because as you say like a a lot of the time that's where you know the the longer that you extend that feedback window like the less accurate i guess that yeah yeah it's because it's like yeah it's like something's driven them to that website in that moment to buy yeah example being i was procrastinating on buying a water bottle for like four months because i just kept but i just kept buying plastic bottles at the gym and then i went one day and they had no they had no plastic water bottles so i bought i bought a water bottle that's a stupid like example top of mind yeah i don't think that's gonna do it but no you know what i mean it's like people other people will see something procrastinate around purchasing it and then something will trigger them to go and make that action that day and they got a lot of data points around that and it shifted this the strategy quite a bit yeah worth doing second one is customer interviews i actually saw i shared it in our chat today that 11 labs have released customer interviews at scale through ai which i think is a very interesting it's a great idea really good idea yeah i mean it's it it was expected to come because if you think about how it's being used as like a sales mechanism now for like even just sales calls and things like that yeah and we haven't tested this it literally came out yesterday but i we've seen the impacts of getting customers on on the phone um we probably don't internally do it as much as we ideally should because it's quite time consuming it's and we're moving at a lot of pace for a lot of clients um but i think that opens the door to maybe being able to just script those out and just doing like and you could even then think about trying to do it across different segments of customers you could do it across different personas if you can get that detailed in how you segment them in the back end of the business yeah um and you could tailor the script towards that so i think that's a really interesting it's a great idea brand of of research that we're to see come to the forefront.

34:17Yeah, we'll definitely test that and we'll let everyone know how we get on with it. I remembered what I was going to say earlier that I forgot. Just back, let's reverse back into the ad account. One thing that we big piece of advice, even if you're don't turn off winning ads. I just don't believe that you should ever turn off an ad that has been a significant winner. Even in times when CPA may start to go above target, we've had a couple of cases recently where we've seen this behavior in accounts and you come in without context and you turn off those winning performers because they are above CPA because we're going through a period where we need to do work, we need to improve the performance of the account as a whole.

35:10And it can be tempting without that context just to turn the ad off because it objectively looks poor, but a lot of the time they can be supporting the whole ad accounts. And it's always a view of let's beat that, let's succeed that. We may need to tape a spend off whilst we do that to make sure CPA comes down and we have that under control. But I just don't believe in turning off significant winning ads. It's always a beat them process. Back to Persona. Any more bits on Persona? No, I think we'll definitely do a full episode going much deeper into this. It's just some stuff that I'm working on at the moment so we can do a full episode breakdown of that.

35:49Should we get on to AI? Yeah, let's get on to AI. You want to kick us off here with, I know you've been spending a lot of the Christmas break just experimenting, playing. What have you been doing? Well, I feel like we've gone through, as I said, another acceleration of just narrative around AI's pace of progression. I'm not very technical, but over like December mid December, late December there was a lot of narrative on Twitter from big influencers the Claude team themselves around the quality of output people were getting from Claude code and how it had almost reached this inflection point where it is so good that you just do not need any technical background to build really significant pieces of software so just went down a rabbit hole I saw a tweet actually that can reference that I think really crystallizes what we're going to experience over the next 12 months.

36:47And it said those people who experiment with and are curious about the frontier models of AI are going to start to enter a fundamentally different reality to those who don't use them because they're going to be able to produce seven days worth of work in one day. And then it's going to be 30 days work in one day. and then that's like the that's the sort of cycle we're in i've also been reading stripe presses oral history of ai scaling and like how that's been going which is quite i recommend people going and kind of read some of the fundamentals about how the models work how how scaling model training and like the what what that means and some of the security stuff because that's mental like some of the um you know they don't have any idea how like a model actually works yeah I thought that's crazy.

37:39Which is bonkers. And then that leads to like a lot of questions around security. And yeah, anyway, that's very philosophical work done that word. But yeah, I thought that was just, and so that led me to just doing more and more research and just spending more and more time trying to just interface and build ways of working with the tools. And just think Claude right now is just so, so good. like what they're putting out feature wise see them as almost like the apple of of ai like i enjoy using the tool way more than the others even though maybe gemini is better at certain things like they've got it's got like a taste aspect to it which i think is really enjoyable i agree the ux for me is is definitely the most enjoyable and i i feel like they also are doing that whole like steve jobs talks about that intersection of humanities and technology that's like one of his big principles and how you humanize tech.

38:35And I feel like the way that they're talking about skills, co-worker, rather than gem projects, it really humanizes the experience and makes it really relatable and makes it really easy to understand, which I think then makes you want to go and explore it more. Exactly. It's that whole thing of jobs as well, like simplification. Yeah. And I agree, like especially with Claude's skills, like you know you can you can technically do all the same stuff across a lot of these different models but the way how accessible Claude make Claude skills and how you can literally have a conversation with it and it will direct you in in the right uh yeah in the right way to upload the right context at the right point and then help you actually layer um its logic flow is is like just just so it's beautifully done and it makes it so much easier yeah and i think the context window for Claude is just so, so much great, so much, so strong relative to others.

39:35So I think one thing that a lot of people on our team and probably in your businesses is like, you want to centralize everyone to a plan because, and then you need to make sure everyone's got the connectors in there set up on like Macs. So Notion for us, Google, Google Calendar, all of these like foundational things. I think it's easy to think that everyone is doing that, but often they aren't and you can go and its retrieval system is amazing so if you're working on a problem and you and you ask it to go and reference anything that it thinks is relevant to that problem in notion it will go and retrieve those files and download it into its context so you need to do that foundational piece but then yeah i think it's been been experimenting with uh all of the different areas i'll let you talk about skills i think i'm a massive fan of deep research still i used i've started to use deep research a lot more for um just most problems because you got one you can also like have about seven running and just disappear and it'll come back it'll just be a loads of information it doesn't always need to be on um on like massive problems or research in its in its truest form it can often just be like anything really.

40:51I wouldn't think of it as just a research agent. I just think of it as something that is able to spin up loads of sequential agents and come back with a lot more depth around the topic. We'll put in the links below a guide, a really in-depth guide on how to leverage deep research to its maximum, which includes how to prompt properly, how to get context in. but one of the big things that i loved about deep research is it always asks for clarifying questions and you notice it does it there but then i was like well let's you should pull that thought process into every trap so at the end of every prompt i just put ask me as many clarifying questions as you need to ask until you are confident you have absolute max context on the problem and that that'll often spit out like 16 to 18 questions and then i'll have to whisper flow or type them through but the quality of output is so much greater than when you don't and i think for many i've kind of the prompt engineering is like a really big topic but i think prompt engineering is often about minimizing differentiation of output like minimizing variance of output in a system so if you're putting in a prompt and you need the output to be consistent every time prompt engineering like narrows variance like you're tweaking prompts i think when you're doing a lot of qualitative work it's more you're you're better focusing on context i actually think the prompt has very little impact yeah i just don't even like the um the phrase prompt engineering like i think it should be context engineering yeah it is gathering the the right context to be able to to to create a better output yeah i think i think the actual prompt obviously i'm not saying it doesn't have there's best practices the guide i've will share in the comments has a lot of those but you better spend it you bet eight out of ten context three out of ten prompt is going to be the opposite every day and i think you only need to really obsess around prompts if you need if it's like data if it's like you really need certainty that every time you run that like an agent question yeah um but i think the prompt is the vehicle for context yeah um yeah and i i yeah i fully agree i think um deep research is is is great i've been playing a lot with claude skills which as i said is really like in in any it's really true the point you mentioned just about like curiosity and like how the gap it's going to create with those that are like 10 steps ahead versus not and i think one of those is that i see just create creating a huge uh gap in in efficiency and also quality of output is is is and a vehicle to get there is called skills um i think there's there's there's so much that can be done utilizing this tool um and thinking through the lens of like what are the skills required within your business to get to what your output is and how that we isolate those different skills.

44:06You know, for example, creative strategy is too broad, but then if we distill that down to script writing, there's so many opportunities that we can build there. So we built a Claude skill around script writing and we have this logic flow where essentially it starts with the skill logic, which essentially includes core script writing principles. So it's more thinking universally around script writing, funnel position logic, format detection and routing logic. So essentially it knows through that prompt, it can detect, okay, what is the right format to push us into and why? So it's almost like a, it becomes like this, a skill that becomes a master for almost prompting other skills within it.

45:00So we have a format-specific skill set. We have a funnel position skill set, and it's able to retrieve the right level of context of each of these to be able to get the right output. And then once we get through the initial skill layer, we then go into smart framework selection. So it's able to actually pull out once it has that macro understanding of the actual skill itself being script writing. It can then give us more specific context on the areas that we are discussing. So, for example, if we're creating a high production ad, it's able to pull the right context from high production theory and also high production best performers.

45:46and this is where what you mentioned there about like the connectors which are great with claude so because we've got notion connected and we've got a whole database of best performing frameworks that are split between hyper and ugc and static claude is able to then understand when to go into that best performing framework database pull the context of what is working right now with high production and use that as leverage to drive a better output of the script that it's going to it's about to create yeah nice um and yeah you can start to stack these can't you you can start to have it so it pulls in multiple skills throughout a workflow exactly like an agent network but but slightly different um so i think it becomes it's just again leveraging that context that unfair advantage which is so powerful yeah and it's thinking of it in layers so you know you have your score your core script writing dna which is the same across everything that you're doing so thinking about when you're building skills what that means from from a layer perspective and then how that feeds into like specific pieces of context um and that kind of just shows like it's why it's so important that every every skill that you have as an individual and you have within the business like document it it's the same as like sops like we document a process for everything that we do but i think we're in a world now where we need to be documenting every single skill set and subset of those skills as well 100 agree like we are using it for non uh like tasks that aren't just execution either so presentations um management approach like you can codify a business's approach to management in a skill and that can become a soundboard for for difficult conversations or managing people i think you can do it for general general communication there's so many different ways to ladder this up there's a lot there's a huge skills library on anthropic which you can which you can pull a lot of pre-built skills from and i think that'll only expand but i agree uh it's just such a powerful tool that people need to be leveraging the one thing i would just to to cap this off from a script writing claude skill um perspective like the context documents that you will want to build um is a master script writing um principles context document now a little hack um if you remember my storytelling in scripting mastermind that i did on ddc diaries um take that take the transcript and create a context stock off the back of that because a lot of the inputs that we use in ours are from that video um the other then you just need specific to the different formats right so take take all the transcripts of your best performing high prod ads ugc ads and static ads from the last 90 days and put them into individual documents and do the same with a hook theory um principles document and then really it has everything it needs to looking at script writing more holistically and then from a context specific point of view the only addition that we have is like a creative strategy one-on-one context document that again is almost like the layer up from the script writing principles document um which is more just about like fundamental creative strategy principles that then feed into script writing as an element of that nice very comprehensive that's like a three years of work right into into the context so it is but i think if you're listening to this and you don't have that context yourself within the business like go to the dara denny's of the world's the dc dara's of the world's the sarah levengers there's so much content out there that you can literally just go to youtube show transcripts download transcripts put put them all into a chatbot and ask them to consolidate that information and put it into a context document like that in its simplest form will still do a great job 100 um that's how matt built his uh he built creative strategy agent he said it just downloaded all of your all of our episodes and uploaded into into a bot um one other thing on like prompting that i've found to one thing i'm prompting that i have found to deliver like a lot better results if people aren't doing this already is like chain of thought and asking the AI to like go through step by step.