Main content

Next ‘24 - the rise of the AI agents! Google Cloud makes a compelling platform pitch for the enterprise

Derek du Preez Profile picture for user ddpreez April 9, 2024
Google Cloud CEO Thomas Kurian took to the stage at Google Cloud Next in Las Vegas today to deliver a pitch to enterprises that felt cogent and proficient.

An image of Thomas Kurian on stage at Google Cloud ‘Next 24
(Image sourced via Google Cloud)

Google Cloud’s annual user conference, Next, is taking place in Las Vegas this week and the company’s CEO, Thomas Kurian, took to the stage to outline the vendor’s strategic vision for generative AI. The keynote was full to the brim with product announcements and use case demos - and the vendor did a good job of delivering a compelling pitch for enterprise buyers. Why? It appears that Google Cloud is attempting to answer some of the more practical concerns that end users are currently facing, whilst also giving them a solution that allows them to differentiate through a balance of AI building/buying - with plenty of AI agents in the mix! 

Before we dive into the meat of what Kurian outlined, it’s worth considering the context of the message being delivered this week. Enterprises face a lot of choice at the moment, when assessing their options for generative AI. The hype of 2023 is fading (somewhat) and buyers are having to seriously consider which platforms and vendors will give them trustworthy results, guidance, and proficient tooling to experiment and test use cases. 

Almost every large-scale enterprise vendor in the market is telling buyers that they will answer their enterprise needs and ‘transform’ how employees and customers interact with their organization. Whilst a mix of chosen platforms and products is probably going to be the reality for many, it’s also true that these ‘data owner’ vendors, such as Google Cloud, could disrupt some of the traditional enterprise players. 

That’s because although the system of record vendors will likely integrate generative AI into their platforms, giving users solutions that interplay with existing workflows, Google Cloud is providing a platform that allows companies to build and tune AI solutions that could provide a differentiating experience.

I’ve written previously about how vendors last year were keen to point out that generative AI will progress in a way that allows them to customize models using their existing data, rather than relying on publicly available information. What Google Cloud is seeking to do is give them that opportunity. 

CEO Kurian said today: 

We last came together just eight months ago at Next 2023, but since then, we have made well over a year’s progress innovating and transforming with our customers and partners. We have introduced over a thousand product advances across Google Cloud and Workspace. 

We have expanded our planet-scale infrastructure to 40 regions and announced new subsea cable investments to connect the world to our Cloud with predictable low latency. We have introduced new, state-of-the-art models — including our Gemini models — and brought them to developers and enterprises. 

Last year, the world was just beginning to imagine how generative AI technology could transform businesses — and today, that transformation is well underway. 

A cogent platform

Before we dive into the product announcements, it’s worth noting the foundational elements that Google Cloud is providing to allow enterprises to create their own generative AI experiences. The below image depicts this, but the core components are Google Cloud’s AI Hypercomputer at the infrastructure level, coupled with Gemini models, the use of Vertex AI (a platform that provides users access to other models and tools to fine tune and ground models), with end user tools on top, which include Gemini for Google Cloud and for Workplace, as well a variety of ‘Agents’.

An image of Google Cloud’s AI Stack
(Image sourced via Google Cloud)

The AI Hypercomputer is the hardware and software system that runs an end user’s AI infrastructure in the cloud. Google Cloud announced a whole host of updates to this including its new GPU-based instance, developed with NVIDIA, called A3 Mega; the availability of NVIDIA’s Blackwell platform coming to Google CLoud in early 2025; AI-optimized storage options; TPU v5p, Google Cloud’s new AI accelerator for training and inference; as well as new options for Dynamic Workload Scheduler, which includes calendar mode for start time assurance and flex start for ‘optimized economics’. These announcements are important, but mostly only to note that Google Cloud is working closely with NVIDIA and is keeping pace with the infrastructure demands placed by heavy AI workloads. 

It gets more interesting further up the stack for enterprises, with Google’s Vertex AI. As Kurian noted: 

Vertex AI, our enterprise AI platform, sits on top of our world-class infrastructure. It is the only unified platform that lets customers discover, customize, augment, deploy, and manage gen AI models. We offer more than 130 models, including the latest versions of Gemini, partner models like Claude 3, and popular open models including Gemma, Llama 2, and Mistral.

Vertex AI allows you to tune the foundation model you have chosen with your data. We provide a variety of different techniques including fine tuning, Reinforcement Learning with Human Feedback (RLHF), distilling, supervised, adapter-based tuning techniques such as Low Rank Adaption (LORA) and more. Today we are announcing support supervised, adapter-based tuning, to customize Gemini models in an efficient, lower-cost way. 

Customers get far more from their models when they augment and ground them with enterprise data. Vertex AI helps you do this with managed tooling for extensions, function calling, and grounding. In addition, Retrieval Augmented Generation (RAG) connects your model to enterprise systems to retrieve information and take action, allowing you to get up-to-the-second billing and product data, update customers’ contact info or subscriptions, or even complete transactions. 

What this means in practice is that enterprises are being provided a lot of choice with Vertex, when it comes to the adoption of AI models, as well as a variety of tools to make those models useful. For instance, organizations can now ground their models not only using Google Search, which shows users where information has been pulled from if using publicly available information, but also from their own enterprise data, including applications sources such as Workday or Salesforce (or information stored in their own databases). 

For example, I spoke to a large financial services organization today that used Vertex to build an application that allows agents in its contact center to get a summary of customer records, as well as quickly source summarized information relating to complex policies and procedures. The organization used Vertex tooling to tune and nudge the models, alongside knowledge workers at the organization to ensure the results being delivered were accurate, prior to delivering the application to contact center agents. 

Sitting above Vertex is the use of Google Cloud’s Gemini model in its Workspace suite of tools. Some of the examples we saw today build on what we saw last year, where Gemini can be pointed at documents to quickly summarize information or build presentations that can later be edited by an end user. What was particularly interesting during the keynote was examples of how Google Cloud is integrating its image and video generation capabilities into Workspace, so that an organization can use text to video or text to image tools to quickly build out marketing or social media campaigns. 

Although the demonstrations were likely well selected and rehearsed, the results were particularly dramatic in terms of what it would mean for these functions. Instead of finding a location for a photo or video shoot, hiring a group of people, and editing the end results into something that is usable - Google Cloud seemed to demonstrate that this could all be done in a matter of minutes using its AI capabilities. Alarm bells will be ringing for those working in those areas, but it’s hard to deny that many enterprises will see opportunity here. 

The agents

The most compelling piece of Google Cloud’s AI platform for me, however, was the introduction of AI Agents. As Kurian noted: 

Finally, Vertex AI Agent Builder brings together foundation models, Google Search, and other developer tools, to make it easy for you to build and deploy agents. It provides the convenience of a no-code agent builder console alongside powerful grounding, orchestration and augmentation capabilities. With Vertex AI Agent Builder, you can now quickly create a range of gen AI agents, grounded with Google Search and your organization’s data. 

With our entire AI portfolio — infrastructure, Gemini, models, Vertex AI, and Workspace — many customers and partners are building increasingly sophisticated AI agents. We are excited to see organizations building AI agents that serve customers, support employees, and help them create content, in addition to the coding agents, data agents, and security agents mentioned earlier.

Google Cloud sees these agents adopting a number of roles, including customer agents, employee agents, creative agents, data agents, code agents, and security agents. 

For example, we saw a demonstration on stage of someone seeing a shirt that they wanted to buy, worn by someone in an image they’d seen online. They showcased how a retailer could use an ‘agent’ to allow a shopper to attach the image on their web page and write a prompt, something along the lines of: ‘show me what options you have similar to the person wearing this shirt’. The ‘agent’ then provides a number of options for the shopper, who then was able to select what they wanted to add to their cart. However, the demonstration then went further, showing the shopper calling the retailer to discuss their purchase, as they wanted to know where it was stocked nearby. An AI system was then able to converse with the shopper on the phone, explain where the retailer had it in stock, and carry out the purchase using the shopper’s card on file (whilst upselling them a bunch of other items that would go with the shirt in question). The entire process was carried out by an AI agent. 

Google Cloud provided a number of other examples. For instance, it highlighted how customers are using employee agents in the following ways: 

  • Dasa, the largest medical diagnostics company in Brazil, is helping physicians detect relevant findings in test results more quickly.

  • Etsy uses Vertex AI to improve search, provide more personalized recommendations to buyers, optimize their ads models, and increase the accuracy of delivery dates.

  • Pennymac, a leading US-based national mortgage lender, is using Gemini across several teams including HR, where Gemini in Docs, Sheets, Slides and Gmail is helping them speed up recruiting, hiring, and new employee onboarding.

  • Pepperdine University is using Gemini in Google Meet, enabling real-time translated captioning and notes for students and faculty who speak a wide range of languages.

In addition, Creative Agents, which help teams build and design creative output, were cited in the following examples: 

  • Canva is using Vertex AI to power its Magic Design for Video, helping users skip editing steps.

  • Carrefour is finding new ways to use Gen AI for marketing. Using Vertex AI, Google Cloud said it is creating campaigns across various social networks in weeks not months.  

  • Procter & Gamble is using Imagen to speed up the development of photo-realistic images and creative assets, with the aim of giving teams more time back to “focus on high-level plans”.

Kurian concluding by saying:

To date, we have helped more than a million developers get started with gen AI and our gen AI trainings have been taken millions of times. Looking back at this past year, it’s truly remarkable to see how quickly our customers have moved from enthusiasm and experimentation to implementing AI tools and launching early-stage products. 

Today, realizing the full potential of the cloud goes beyond infrastructure, network, and storage. It demands a new way of thinking. It's embracing possibilities to solve problems in boldly creative ways, and reimagining solutions to achieve the previously impossible. We're both inspired and amazed to see this mindset quickly materialize in our customers’ work as they pave new paths forward in the AI era — whether automating day-to-day tasks or tackling complex challenges.

My take

It’s been a very interesting first day with Google Cloud at its Next event. There has been a lot of information to digest and it’s clear that the company is pushing hard to give enterprises sophisticated tooling that can help them build customized AI solutions. Whilst a lot of this may seem like ‘future gazing’ given what’s being described, it’s worth noting that this has been grounded by many customers speaking about their generative AI deployments to back up what’s being pitched (more of those stories to come in the next few days on diginomica). Customer after customer was put in front of journalists to show us that this isn’t science fiction and that companies are doing this right now. 

What’s going to be interesting in the coming months and years is how buyers choose between their existing cloud platforms and systems of record, and the options provided by the likes of Google Cloud. I’ve had very interesting conversations with the vendor today about how AI provides an alternative path for enterprise buyers, where the nuts and bolts of process and workflow become less relevant as systems move towards a more ‘conversational’ approach. Knowing the ins and outs of a process becomes less relevant if you can simply ask a system for information and to execute an action for you. I’ve been impressed by Google Cloud’s grounding capabilities and the tuning tooling it provides, which gives enterprises the option to build their own AI solutions within a relatively easy framework.

We are likely entering a new era of ‘experience’, whereby employees and customers will get used to interacting with organizations in an entirely different way. Which vendors enable this the most proficiently remains to be seen, but Google Cloud does make a compelling pitch. That being said, it does seem like this approach will require some big changes on an organizational level for buyers, as departments have to work together to tune the models for the right outcomes. Equally, investment in internal capability will be needed to deliver on this ‘build’ vision being laid out by Google Cloud. Some enterprises will feel more comfortable simply relying on the generative AI capabilities being integrated into their existing platforms, which to some degree maintain the status quo, whilst delivering marginal benefits. Whether or not those using AI in the way Google Cloud lays out will force a new wave of competition in the market thanks to these new experiences being created, forcing a shift in technology decision making, remains to be seen (but I think could be likely). 

We will dive into these topics more deeply over the coming days as we speak to Google Cloud’s leadership. Keep your eyes on diginomica for further updates. 

A grey colored placeholder image