S2. Ep21 - Google's Big AI Update: What It Means for Your Business

Katie (00:26)

Hello and welcome back to another episode. Hi, hello, I'm Katie and as always I've got Noel here with me today. Hi Noel, how you doing?

Noel (00:35)

Yeah, I'm doing great. Just trying to keep cool in this crazy weather.

Katie (00:39)

Yeah, we're currently experiencing a heatwave here in the UK and as everyone knows, us Brits love to moan about the weather. Hot, cold, dry, raining.

Noel (00:43)

It's never just right.

Katie (00:59)

So today's podcast episode we are covering lots and lots of updates and we're also not just telling you the updates but also how they would affect you and your business or how you're using AI or automations for your business. So Noel, shall we start with the very first update? Because there's quite a few, isn't there, this week?

Noel (01:28)

There is. There's lots from one particular AI provider, which I'm sure is probably gonna be in the podcast description. But I thought we'll go in with one of the little updates or little bits of news that came out last week. And that was basically OpenAI. One of their models solves an 80-year-old maths problem, which kind of baffled mathematician professors and PhD students.

It was quite funny watching them and their reaction to it. Two of them quoted that they'd lost multiple nights' sleep over this AI's ability to outdo them in a maths problem. It was quite funny watching them because they just couldn't get over it. But essentially what it did is AI kind of looked at it a bit outside the box. So it used multiple different maths problems and basically stuck them all together to come out with one solution. So it used lots of different things that as humans we hadn't even thought of. That's why no one solved it for 80 years because they hadn't really thought outside the box. But yeah, that's the first time AI has solved a big maths problem.

Katie (02:57)

Very cool. I mean, I don't think the mathematicians are probably saying that, but yeah, it's very cool.

Noel (03:04)

They did lots of checking, checking how it did it all and making sure it was all correct. And to their surprise it was. No real use cases in the real world, I don't think, for that particular maths problem just yet. I did do some research and it didn't really come back with much that was noteworthy. It's not gonna change our lives anytime soon. But who knows? The next maths problem it solves might.

Katie (03:34)

It might open up the key to something else. Good to know.

Noel (03:44)

Yeah. So the next set of updates are all from Google. I would say they've been very quiet this year up until now. They always come out with stuff in bulk. They're a bit like Apple. They don't drip-feed products, they release them like twice a year. So Google's big event was the other week and they released tons and tons of new stuff. Lots of it isn't related to AI or any sort of automation stuff, so we're gonna skip all over that. We'd be here all day otherwise.

But I've picked out four particular things I think would be really interesting for everyone in this sort of space. So to start off with is the new model, Gemini 3.5 Flash. Which sounds very fast.

Katie (04:52)

I love that they've called it that. I love that they've added Flash on the end. It always makes me chuckle. It makes me think of either Flash cleaning spray or Flash Gordon.

Noel (05:17)

No, there's no flying monkeys in this update. Showing our age a bit there, isn't it? But with 3.5 Flash, obviously it's smarter than the 3 Pro. It outdoes it on lots of benchmarks and stuff. But I find it very hard to get my head around benchmarks. I'm sure lots of people that listen to this look at the tables and think, well, who cares?

So I guess Noel's top tip for whenever a new model comes out, what I always look at is, is it a generational leap? Is it going up by a lot or just a tiny bit? Are there new features that have been added? And ultimately from an AI API cost perspective, how much is this gonna cost me if I plug it into an agent?

So with 3.5 Flash, I wouldn't say it's a generational leap. It's a decent jump from 3 Pro. It's not adding in any real new features, it's multimodal just like normal Gemini models. So you can put in images, video, audio, and it can get on and do stuff with it.

But in terms of cost, it is actually cheaper than Gemini 3 Pro on the API, which is interesting. But I did a bit more digging on this and found something quite interesting. So yes, it's cheaper, but actually in the long run it's more expensive. Because what they're finding when they're running these benchmark tests is that 3.5 Flash is actually using far more tokens than 3 Pro. So by the time you've added it all together, it's costing you more money than 3 Pro would. So that's something to bear in mind. Don't just look at the API costs and then think it's cheaper, because it is gonna chew through tokens far quicker.

I've been using it on Clyde actually, plugging it into my AI agent there. And I would say he's been quite thirsty, and now I've looked into it, it does make a bit more sense.

Katie (07:51)

So it is burning through tokens a lot more than the 3 Pro.

Noel (07:55)

Yes. In one benchmark the average is about 36 million tokens to do the full test. And then 3.5 Flash will go through 78 million. So it's a bit more than double. I would say if you're using the Gemini app, I'm sure you're getting lots of usage out of it and stuff like that. It would be very good, very quick, very useful. But yeah, worth noting the API costs.

Katie (08:33)

Shall we go on to talk about Gemini Spark?

Noel (08:33)

So this is an interesting one. This is their AI agent that they're bringing out. So the other week OpenAI brought out a couple of little agents that we talked about a couple of weeks ago. So this is kind of Google's take on this. It works on the cloud, so you don't need to have your laptop or your phone up and running at all times. You can shut the lid and walk off and it's just gonna keep carrying on working while you're away, which is great.

And one really cool feature is it will actually work within the Chrome browser. I mean, why wouldn't it? It's one of Google's products, so might as well put the two together. So in terms of computer use, it should be far better than anything that OpenAI or Claude could cobble together. They're very good, I give them that, but I'm pretty sure Spark will be able to use some sneaky little features in the back end that other people don't know about. So it should be a lot more efficient.

You can connect it to pretty much anything. It's mainly using the Model Context Protocol or MCP to connect to apps. So most apps that you're using in your business today will have not only API access but also MCP. So if you look within your settings, you'll be able to find how to get those codes and links, and you'll be able to connect that into Spark. So you can connect thousands of apps into it and you're not really doing a great deal either. You're just copying a link, putting in your password, and that's it. You're done. And then Spark can do whatever you need it to do.

The other thing, and it links to another release as well, they will be giving Spark access to the AI Agent Payment Protocol, which is coming soon on the Chrome browser. And what they also released, which is linked to this, is they're bringing out a common cart. So you could be shopping on many different websites, you could be shopping on three different sites, and then you could just check out using the Google cart, which will get everything together, and then you just do one payment instead of three or four or whatever.

So for my thinking, I thought, well, that'd be quite clever if there was an agent attached to this. And I did some research, and there is a payment protocol coming soon.

Katie (11:32)

So what does the Google cart mean for businesses who sell things on their website? So it could be courses, maybe it's one-to-one. Will they be able to be part of it or would you say it's mainly for shopping, like clothes, rather than for services?

Noel (12:11)

I think most of the things that we're talking about was general shopping. But I guess you're gonna need a cart system which is compliant and allowed to attach to it. There's apps like ThriveCart, which I know you've used in the past, Katie. So I don't know if that would be part of it, or there's many other different options.

Katie (12:38)

Or if it's gonna be things like Shopify and e-commerce.

Noel (12:45)

I would have thought Shopify would definitely have this nailed down. They'd be all over that. But I think if you're selling any products online, I think now we need to start thinking about, will an AI agent want to buy from me? Not will the person buy, but will the agent for that person want to buy?

Katie (13:17)

Yes, it's making sure that those sales pages have all the information. So that's going to be really interesting because obviously a lot of people currently buy products or maybe even services via ads. So maybe they've seen a video that is an ad for a product or a service, and then they go to buy. And I'm just wondering, without the ad, is the sales page going to have enough information for that AI agent to then go buy? Because at the moment obviously all of the adverts and our sales pages are set up for humans to buy. So a lot of it goes into the emotion and why they need it and why they can't live without it and how it's going to change their lives, how it's going to make them feel better. But does an AI agent care about those things?

Noel (14:38)

Yeah. I think it's also, and I know this is getting a bit nerdy, but it's almost like you need to make sure that the code that the agent can read is actually giving the information that would be helpful. So it's not going to look at the page. It doesn't care about the colours, it doesn't care about any images you could have with things like five bullet points, it can't see it. So those are forgotten.

It kind of links into the AI SEO stuff that we did with Milana last year. It kind of all interlinks into that. The AI is just looking at a code snippet. And I would go off and ask any AI model how to do it, it will tell you and give you the code snippet to make it better. But one of the big things it does like is FAQs. It loves an FAQ.

Katie (15:46)

And that was something that Milana was really saying in that episode, which we will link in the show notes. It was in series one. Milana was going into so much detail about that, wasn't she, saying about the FAQs and how important they were to have on every page. Not just the sales page, but every page of your website.

Noel (16:20)

It was. And you almost need to link the content that's on the page to those FAQs as well. So they can't be completely unrelated. It will soon spot if there's something amiss. So you can't quite fool it. But it's important to have those on there and you have to have them within the script code, which AI can produce for you. Ask it to produce FAQs script with the FAQs, good to go. But yeah, having fancy colours and images, if AI agents come to view your services or products, they're not that bothered about it. They just want to know the details and if it matches.

Katie (17:04)

Because at the moment we do spend so much time making sure it is visually appealing. It's got that beautiful brand, it's got those beautiful images. Some people have spent a lot of time, a lot of effort, even a lot of money having a creative director come on board for these products. If it's a big group programme that they're launching, you want to make sure that everything's coherent to the brand. I know people who have spent thousands and thousands on a creative director for the launch of their group programme.

So that's really interesting now, actually, how it is shifting. I still feel like that is so important having the right imagery and brand, but it's now also having that secondary part that is actually appealing to an AI agent to then report back and go, I think this is the course that would be best for you, or I think this is the product that I think would be best for you after doing all of my research.

Noel (18:28)

Yeah. I think actually where you mentioned secondary, I think secondary will actually become the primary point. The primary point is the AI one has to find it interesting. But then when you give it to the user, that's then what used to be the primary, the visuals, and then go, wow, okay, this looks really professional. I want to get in on this. I think we're starting to see that flip over, which is really interesting.

Katie (18:59)

Yeah. Maybe secondary wasn't the right word. I was thinking it's as important to have the visuals but also the information for the AI agent. There's no first or second, they're both as important to include. They're on par.

Noel (19:27)

Yes, definitely. But without the AI searchable, then yeah, forget it.

Katie (19:29)

Yeah, but then also without the visuals and the beautiful images and the right layout, the right copy, then again a human isn't going to click yes to buy. You have to have both equally. You have to have the AI agent that can find you and then report back to the human and say, go and check out this course. But then when the human goes to check out that web page, it has to have not only the right information but the right look.

Noel (20:22)

Yeah. But I guess there's also another thing that I just thought of. Is it gonna get to a point where you'll do your search chat with whichever AI, and then you'll just go, yeah, that's fine, just go and buy it. And then it'll use that Agent Payment Protocol to just go and pay for it. I think it'd be really interesting when it does come out. I would imagine they'd have it in the analytics for the website so you'd be able to see how that purchase was made. But I think it'd be really interesting to see who actually goes and visits from an AI search and then who just uses the AI to go and buy it.

Katie (21:05)

I mean, me right now, not actually going to check them out personally gives me anxiety. Like why wouldn't you? But I feel like this time next year we could be sat here saying, well, why would you go and check it out? Because the AI agents are so reliable.

Noel (21:22)

Exactly, yeah.

Katie (21:27)

But right now it gives me extreme anxiety to think that someone would just buy without even checking it out. I've had Claude come back and given me three different courses and I went and checked all of them out and I feel like it did an all right-ish job. I did not purchase any of them.

Noel (22:10)

Yeah. But that's just based on what it could find. So there could be the right course for you, but if AI can't find it, then it can't give it to you. What a crazy world to live in.

Katie (22:35)

Okay, so that was Gemini Spark we were talking about. What about Google Flow?

Noel (22:42)

So Flow has been around for a little while, but I thought it'd be really interesting to talk about this one because of what it could do. So it's a video editing and creation platform. That's all Flow is. But what they've recently added is a Gemini Omni model, which is quite clever. So what you can do is essentially give it a starting image or an end image that you've created somewhere else using like Nano or whatever. You could give it a voiceover or music and then you could give it text all in one go to say, this is how I want it to work, this is how the scene's gonna play out. But it's taking all of those bits of information in one big chunk to create that clip.

Previously you would have to create the clip, maybe have a couple of goes at it, and then you would have to put it in an editing platform to add the audio, and then there's all kinds of little things like that which took you away from the Flow platform. You would have to edit it somewhere else. Whereas now this Gemini Omni model is just going to do all of that for you within that platform. So I think that's gonna be pretty cool to see what videos and things businesses come out with. Hopefully it's not just gonna be yetis in the snow and stuff like that. Let's use it for something useful.

Katie (24:28)

And I feel like this is only gonna get better and better because at the moment isn't there a maximum clip of ten seconds?

Noel (24:36)

There is, but I believe most video models have a maximum of eight seconds. So it is a little bit more. But I guess if you've got your creative brain together, you could create a sequence of clips and then put them all together to get that 30-second or whatever video that you want. Especially for products and things, you could have a really cool transition to move from one bit to the next. I think that's gonna be really awesome to see how that works.

Katie (25:20)

I feel like this has to get better because it's such a short video. I feel like in six months' time or a year's time, Google Flow is gonna have way more updates. Hopefully. Or it's gonna not be around anymore. It always goes one way or the other, doesn't it?

Noel (25:38)

Well yeah, Sora died a death, didn't it, not long ago?

Katie (25:42)

Yeah, RIP. So hopefully it's gonna create longer videos and it will still be here in six to twelve months' time.

Noel (25:53)

Google have got the computing power and the money to do this sort of stuff. Whereas OpenAI I think were trying to do too much. Never mind. So yeah, interesting update that one.

And we can actually use Flow for free. If you've got a free Gemini account, I think you can do two videos a day. So you can't go crazy, but you can use it on a free account.

Katie (26:35)

I think that's pretty good. Two videos a day. I mean, not great when you're just trying to test it out and play around with it.

Noel (26:43)

Yeah, you have to come back to it tomorrow. But the usage links into the Gemini pricing structure. So as you pay more money into the Gemini ecosystem, you get more credits to use on Flow.

Katie (27:04)

The next update is AntiGravity 2.0 app builder.

Noel (27:10)

So to go back to AntiGravity 1.0, that's what I used to use every single day because it was a way you could build apps, websites, or whatever, but you could see the code. If you were so inclined, you could edit the code. You could fire up terminals and run the app and all this sort of great stuff.

I found it really useful. I mean, I'm not a developer, but I still found it useful, just to see things going on and being able to edit things manually if I wanted to. But what they decided to do with 2.0, and it has caused quite a bit of upset within the AntiGravity community, is all of that file editing and viewing has gone. So you can't open a terminal, you can't view the files, you can't edit files. All the developers were like, well, I really liked it, I wanted the best of both worlds.

AntiGravity 2.0 came out as a mandatory update. It wasn't optional. So lots of developers weren't too pleased about it. But essentially what it is now is a chat session. So you can talk about the app or website you're trying to build. And what it does is it uses a multi-agent system to create your apps and websites quicker. So it'll be able to do lots of tasks in parallel. It should therefore be more accurate. But yeah, unfortunately it just annoyed millions of people at the same time.

I haven't fully tried it out yet because I haven't quite got my head around the app as such. I need to go off and give it a go. But apparently it's better than what it used to be from an agent coding perspective.

Katie (29:20)

So again, this goes back to AI being all about communication, doesn't it, rather than tech? Do you think that's why they've done this update? Because more and more AI and automations are actually about how well you can communicate with the platform and the system. And it was becoming less and less about tech. I feel that's exactly the route that AntiGravity has gone down, and that's probably why it's upset developers because they are very techy and it's code-based. Whereas actually AntiGravity would be like, well, actually we want more people to use it. So we're gonna take away the tech bit and make it more accessible to people by just making it about communication and having a chat with the app.

Noel (29:57)

Yeah, because I think with the first version of it, it would have been very overwhelming. You wouldn't have enjoyed the experience, let's put it that way. Whereas now it's just like opening a normal chat session. But what I would say is a lot of people I've seen online using the earlier version of AntiGravity were using it a bit like me. I would never use the coding agent. I never touched it for that. I was just using it to edit stuff, use the terminal, maybe sometimes the agent, but really I was using Claude Code to do the bulk of the work. And that was just a chat session.

So really they've looked at what Codex from OpenAI and Claude Code from Anthropic are doing, and gone, well, actually maybe we should be more like those guys and move over to that. Because I'm just chatting with Claude Code like I would with any other AI agent. So I guess I should find AntiGravity 2.0 really easy, but I'll have to give it a go and see how I get on.

Katie (31:30)

It'll be really interesting to see if more people start using it or if there's a big decline. Because AntiGravity 1.0 has been around for quite some time, hasn't it?

Noel (31:42)

Well, in the AI world, yeah, probably about six months.

Katie (31:46)

That's a very long time in the AI world.

Noel (31:51)

It is, isn't it? It's like a decade.

Katie (31:53)

So I think that's the thing. People get used to using it a certain way and then when it completely changes everyone freaks out.

Noel (32:01)

Exactly. I opened it after the update and I was straight away like, where's the files gone? I've opened a project, now what do I do? I've just got a text box.

It also forces you to use the Gemini models instead. So before I would use AntiGravity and then load the terminal and then use Claude Code in the terminal. So I guess they're trying to get around that as well. More people are using their models to do the coding because selfishly that's where the money's at. We're spending lots of money on tokens, which is only a good thing if you're an AI provider.

Katie (33:04)

As a business model it makes complete sense.

Noel (33:09)

Definitely. It makes more sense now than when I was looking at it.

Katie (33:15)

Amazing. Well, I think that was all the updates. I say all the updates, that was a lot.

Noel (33:26)

There was lots of other stuff in the Google ecosystem that they updated. If you've got Android devices or anything Google-related, then they did tons of stuff.

Katie (33:41)

But let us know what you think of any of the updates that we've discussed today. Either email in at hello@makeautomations.ai or you can come over to our free group on LinkedIn, which is AI Automations for Business. Everyone is welcome to join, it's completely free. There's no upselling or anything, just come and have a chat with fellow people who love AI and automations. We always love hearing from you, knowing what you're doing with AI, building with AI or automations, how you're using it, how you're implementing it into your business. Come and let us know. We always love hearing about it, don't we, Noel?

Noel (34:22)

Absolutely. It's really awesome seeing what people do with it.

Katie (34:27)

Brilliant. Well, thank you so much for listening to this week's podcast episode. We hope you've enjoyed it and we'll catch you for another one very soon.