Latest / In The Loop / EP17 - Google I/O & Microsoft Build In 10 Minutes: What We Learned From The Two Biggest AI Conferences

Transcript

EP17 - Google I/O & Microsoft Build In 10 Minutes: What We Learned From The Two Biggest AI Conferences · In The Loop

• 0:00
  In seven days, Google and Microsoft launch 150 new AI
• 0:04
  features between them at the two biggest AI conferences of the
• 0:07
  entire year, Microsoft alone mention agents 300 times in just
• 0:11
  a few days. And while both companies are
• 0:13
  clearly obsessed with AI, they've both taken different
• 0:16
  approaches. It's been highlighted quite
• 0:18
  clearly at their conference. Google has gone all in when it
• 0:21
  comes to interweaving AI into consumer products, and Microsoft
• 0:25
  has painted a picture of an open agentic web that takes over the
• 0:28
  world of work. Into today's episode, I'm going
• 0:30
  to explore what I think is the more interesting parts of their
• 0:33
  announcements and what it means for the future direction of AI.
• 0:37
  This is in the loop with Jack Horton.
• 0:39
  I hope you enjoy the show. For
• 0:51
  today's episode, I'm going to try and do my best to do a round
• 0:54
  up of all the big announcements that happened at the conference.
• 0:58
  As I said, there's over 150 of them, so I don't want to go
• 1:00
  through all of them and I'm going to try and cherry pick
• 1:03
  what I think are the more interesting parts of the
• 1:05
  announcements and really what it means for the future direction
• 1:08
  of AI over the next probably 3-6 or nine months.
• 1:11
  So I'm going to start with Microsoft because actually I
• 1:13
  think their build conference, that's their big developer
• 1:16
  conference of the year was actually more interesting
• 1:19
  because they painted this new picture, which they call the
• 1:22
  Frontier Firm. So this is an organization that
• 1:32
  embeds AI intelligence from the ground up, where humans suddenly
• 1:36
  have teams of agents around them and everyone becomes an agent
• 1:39
  manager, and we basically scrap traditional organizational
• 1:42
  charts and where people have to become AI bosses.
• 1:46
  And Microsoft CEO Satya Nadella said that they're going to go
• 1:49
  much further than just provide case studies to customers about
• 1:54
  what this could look like for them.
• 1:55
  And they're actually aiming to have 50% of all code written by
• 1:59
  AI alone within the next year. To put that in perspective,
• 2:02
  they've got 20 to 30% of all code now written by AI.
• 2:05
  And they've just announced a series of layoffs.
• 2:08
  So 3% of their entire workforce is being laid off, which equates
• 2:11
  to about 6000 people. And to be honest, it wouldn't
• 2:14
  surprise me if this is also them eating their own dog food and
• 2:17
  reinvesting some of that cash into trying to grow the business
• 2:21
  from an AI perspective. So Microsoft really is penning
• 2:25
  this picture of them being customer 0 for all the same AI
• 2:28
  ideas that they're selling to enterprise.
• 2:30
  And the open agentic web is a key concept to that vision.
• 2:34
  And it basically took over the entire of the first day for
• 2:36
  Microsoft's conference. So what is the Agentic Web?
• 2:39
  It's called? It's a idea or a series of ideas
• 2:43
  that are not new. It's things that were discussed
• 2:45
  on this podcast a number of times, but really the backing of
• 2:49
  Microsoft and creating a narrative around it is going to
• 2:52
  bring it all into being much faster.
• 2:55
  So this involves AI agents that interact with products and
• 2:59
  services on the Internet, from booking appointments to
• 3:02
  collaborating with other agents to buying products from other
• 3:05
  agents on behalf of me or other people.
• 3:09
  And the open part is what is critical because Microsoft
• 3:12
  doesn't want to build this ecosystem that's closed.
• 3:15
  They want to have an interoperable, open standard
• 3:19
  ecosystem that allows any company to put their agents
• 3:24
  inside it to collaborate, to buy things and for their ecosystem
• 3:29
  or Microsoft ecosystem to facilitate all of this.
• 3:32
  I've said this before, these big companies really want to own
• 3:36
  everything when it comes to AI. And I think this is another core
• 3:39
  part of that strategy for Microsoft.
• 3:41
  And to make this vision a reality, they've introduced a
• 3:43
  number of important standards and components that I'll go
• 3:46
  through because none of them, again, are that new.
• 3:49
  But bringing them all together in a shared vision and bringing
• 3:52
  other companies on board is what's going to make this
• 3:54
  happen. So first, there's model context
• 3:56
  protocol. So MCP, which is that universal
• 3:59
  USBC connector between agents and other platforms.
• 4:03
  So it let's agents use other systems to share data so my
• 4:07
  agent can speak to Amazon and make an order, for example.
• 4:11
  Now, I did a deep dive into MCP not long ago, so I recommend go
• 4:15
  listen to that, but their support for this is a huge leap
• 4:19
  forward. Second, we've got agent to agent
• 4:21
  protocol. This is another open protocol
• 4:23
  that allows agents from different vendors or ecosystems
• 4:27
  collaborate directly. So imagine your Microsoft based
• 4:29
  agent working with Google's agent or the Amazon agent
• 4:33
  without any incompatibility issues.
• 4:35
  Again, this is about allowing people like Mindset and others
• 4:38
  that are building technology in this space to all work together
• 4:43
  and connect with each other's ecosystems.
• 4:46
  Again, I did an entire episode in A to a protocol.
• 4:49
  So I again, I recommend go listen to that if you want to
• 4:51
  learn. And what's key about all of this
• 4:52
  is they're actively supporting these protocols and protocols
• 4:57
  are only successful if you, if you listen to those episodes
• 5:00
  when lots and lots of different people all use them, bake them
• 5:03
  into their strategies, their products and how they build
• 5:05
  things. And then we turn to all the
• 5:07
  infrastructure and the technology support that
• 5:09
  Microsoft have now launched to make those protocols a reality.
• 5:13
  Again, they're positioning themselves as a neutral vendor
• 5:17
  that allows people to work together in a shared ecosystem.
• 5:20
  So they're upgrading their copilot studio for multi agent
• 5:24
  systems. So that's enabling people to
• 5:26
  start thinking about teams of agents.
• 5:28
  They're going to launch the Azure AI foundry, which is
• 5:31
  basically a service to monitor agents and then they're
• 5:34
  launching what they call agent ID.
• 5:36
  So this is like an identity management system for agents.
• 5:39
  So just like a human employee has, you know, permissions and
• 5:43
  roles and what those permission and roles mean for access of
• 5:46
  information within a company, that's what that agent ID system
• 5:50
  is going to allow. These tools will form the
• 5:51
  backbone of that vision for the Openagentic Web.
• 5:55
  Again, nothing new. We've discussed all of them in
• 5:57
  this podcast. But Microsoft support makes all
• 5:59
  of this really critical to happen.
• 6:01
  A part of that vision for the Openagentic web is called the NL
• 6:05
  Web Initiative. Now, this is short for Natural
• 6:08
  Language Web. Now the NL Web wants to make
• 6:10
  websites more agent friendly. I did discuss this a while ago
• 6:14
  in terms of multi agent systems and what they are and the
• 6:16
  challenges of agents going shopping for me and accessing
• 6:19
  websites for example. Because instead of what we now
• 6:22
  have which is just human readable web pages, this
• 6:26
  initiative is trying to make websites something that can
• 6:29
  expose certain functionality data to agents.
• 6:33
  So through APIs and protocols. So basically it would make it
• 6:36
  really easy for websites to allow, say, Jack's agent to go
• 6:41
  buy tickets from another company site.
• 6:44
  Because again, if the websites aren't agent friendly, an agent
• 6:47
  can't come and buy things. And companies like Shopify,
• 6:50
  Snowflake, and even some institutions like the Chicago
• 6:52
  Public Library are already on board, which again, is quite
• 6:56
  interesting because as I've said before, for these types of
• 6:58
  protocols and initiatives to work, they're collaborative.
• 7:01
  They require people to opt in and embed these ideas into how
• 7:06
  they build things. So that's the open agentic web.
• 7:08
  It's quite an interesting concept.
• 7:21
  Now if I turn to the Google IO conference, which is again their
• 7:23
  big developer conference and these happened just days apart.
• 7:26
  What struck me about their conference was it told you a lot
• 7:31
  about the difference in strategies here.
• 7:33
  So Google is certainly the consumer AI strategy and
• 7:36
  Microsoft is the enterprise AI strategy because the Google
• 7:39
  conference had a huge amount of Consumer Focus, AI in search for
• 7:43
  everyone, AI features in Gmail, Photos, Android Shopping, even a
• 7:48
  lot of its announcements when it pertains to developers were
• 7:52
  related to consumer products. So Android or Firebase for
• 7:57
  consumer apps, and Google's biggest splashy demos were
• 8:01
  always, again, in that consumer space.
• 8:03
  So like the AR glasses for translating speech in real time
• 8:07
  or their magic photo editor. Whereas, as I said, Microsoft's
• 8:10
  Build conference was very much on that enterprise and business
• 8:13
  use case. Now, what's very clear here is
• 8:15
  that Google really hasn't settled on a vision for the
• 8:19
  eugenic web, and instead they're kind of throwing spaghetti at
• 8:22
  the wall with different experiments, which makes sense.
• 8:25
  They're trying to test what resonates most with users.
• 8:28
  For example, one of these experiments is called Project
• 8:30
  Marina, which is a initiative to enable AI to control computer
• 8:35
  screens. So this is often called Computer
• 8:36
  use agents. Again, nothing terribly new.
• 8:39
  Think of it as AI acting like a real user so it can click
• 8:42
  buttons, fill out forms on a website, and again, without
• 8:46
  human assistance there. They've also got agent mode
• 8:49
  launch, which basically uses model context protocol to allow
• 8:53
  users to say, describe, plan my weekend trip.
• 8:56
  And the AI can then create and execute a plan and then use MCP
• 9:02
  to access different platforms and websites to book you a say,
• 9:06
  Airbnb or whatever it might be. And they've also got a lot of
• 9:09
  different experiments with AI search and shopping.
• 9:11
  For example, they've said that their Google AI overview, so
• 9:14
  anyone that uses Google these days would have probably spotted
• 9:16
  them, now serves over 1 1/2 billion people in over 200
• 9:20
  countries, which technically makes Google search one of the
• 9:22
  most used generative AI products in the entire world.
• 9:25
  And internal data from Google has showed that AI overviews
• 9:29
  have driven a 10% increase in search usage.
• 9:32
  So for them, they're trying to claim to Wall Street and to
• 9:35
  everyone around the world that AI is not going to cannibalize
• 9:38
  their big search market, which represents most of their
• 9:40
  revenue, but it's actually going to make it bigger, faster,
• 9:44
  stronger. And they've now actually
• 9:45
  launched products specifically for AI.
• 9:47
  So there's a £200 a month license to use like Notebook LM,
• 9:51
  Gemini's best model, kind of like what Open AI have been
• 9:53
  doing, which is actually one of the few big companies in the
• 9:57
  world to have actually done something like that.
• 10:00
  Most of them are hiding AI revenue inside other parts of
• 10:03
  their business. So it's really difficult to
• 10:04
  understand what people are paying for when it comes to AI
• 10:08
  versus their traditional normal business models.
• 10:10
  They've also launched new AI driven search shopping
• 10:13
  experience products. So they've got a conversational
• 10:15
  shopping assistant to help basically people refine their
• 10:19
  product searches. And they've even got virtual try
• 10:21
  ONS so you can let customers, you know, upload photos to see
• 10:24
  exactly what clothing would look like on them.
• 10:28
  Again, nothing terribly new, but they're throwing a lot of
• 10:30
  different experiments at the wall about what this opener
• 10:32
  Gentic Web means for them. And they've even launched a new
• 10:35
  consumer hardware product. So they previewed their Android
• 10:39
  XR glasses built with Samsung. And they're working on something
• 10:43
  called Project Muhan, which is an ARVR headset.
• 10:46
  And they're going to be partnering with Warby Parker on
• 10:49
  all of this. So hopefully they've learnt from
• 10:51
  their last experiment, which was their Google Glasses failure
• 10:54
  when they tried to build hardware themselves and realize
• 10:57
  we suck at this. And they're probably copying
• 10:59
  what Meta did by partnering with Ray Ban as well.
• 11:01
  And if you've been listening to every episode, you might notice
• 11:04
  a bit of a trend that I'm hedging my bets.
• 11:06
  But I do wonder whether a mobile phone could be completely
• 11:10
  replaced over time by some form of glasses or AI technology,
• 11:15
  because the phone certainly probably won't be the main
• 11:18
  consumer device forever. So to conclude, as I said, I I
• 11:21
  think they're taking very different strategies, as you can
• 11:24
  tell with the different types of announcements and Doug and Rung,
• 11:26
  there are some crossovers when it comes to developers and there
• 11:30
  are some crossovers in a few other places.
• 11:31
  They're both huge companies launching a crazy amount of
• 11:34
  stuff, but there is a big, big separation here.
• 11:37
  Now, who's going to win? Who's going to be the best?
• 11:39
  Who knows? I hate to say because I've never
• 11:42
  been a massive Microsoft fan, but I think Microsoft are making
• 11:45
  massive leaps when it comes to the enterprise side and it's a
• 11:49
  particularly interesting space. Anyway, that's it for today.
• 11:51
  I hope you enjoyed the episode and I'm going to see you next
• 11:54
  week.