Cloud Subsequent This week Google joined a throng of tech distributors pushing the idea of “agentic AI” on an unsuspecting and maybe unreceptive assortment of enterprise customers. Questions stay about how efficient this tranche of instruments will likely be at fixing enterprise issues and the way a lot it would all price.
At its annual Google Cloud Subsequent bash in Las Vegas this week, the Chocolate Manufacturing unit opened the floodgates to a slew of product information.
Among the many bulletins was a promise to introduce an Agent Improvement Equipment (ADK), an open-source framework Google mentioned would simplify the method of constructing enterprise software program that integrates AI brokers – question-answering and task-performing software program that makes selections and types its output utilizing giant language fashions. Brokers will be seen as software program that talks to different software program.
Google claims its ADK will permit clients to construct an AI agent in below 100 strains of hopefully intuitive code. The ADK additionally comes with an “agent backyard” of pre-built bots and instruments, in addition to greater than 100 pre-built connectors to widespread knowledge sources.
In the meantime, the search and adverts large is about to introduce a brand new protocol designed to assist brokers from totally different distributors discuss to 1 one other. The Agent2Agent (A2A) protocol has attracted 50 companions signing up, comparable to Accenture, Field, Deloitte, Salesforce, SAP, ServiceNow, and TCS, which it mentioned had been actively contributing to it.
That is giving enterprises false hope in accelerating human alternative, and we’re not there but
Digging into the main points, Google’s agent backyard is a pre-packaged assortment of instruments inside the ADK permits customers to entry 100+ pre-built connectors, customized APIs, integration workflows, or knowledge saved inside Google cloud programs like BigQuery and AlloyDB. Google cited regulation agency Freshfields as an adopter of its agent imaginative and prescient. The smaller biz mentioned it could use Google’s enterprise knowledge platform for AI, Vertex AI, to create bespoke AI brokers for its authorized and enterprise processes, whereas additionally utilizing search-bots to deliver collectively company-held info.
To energy language processing and knowledge gathering for customers, Google introduced the seventh technology of its Tensor Processing Unit, aka TPU, specialist AI-accelerating matrix-math {hardware} first used internally at Google.
The chips code-named Ironwood – which you’ll learn extra about on The Reg right here and The Subsequent Platform right here – have greater than 10 occasions the efficiency of the sooner Trillium TPU. A completely decked out pod of Ironwood TPUs, rented from Google Cloud, can hit 42.5 exaFLOPS of FP8 compute, “assembly the exponentially rising calls for” of generative fashions comparable to Google’s just lately introduced Gemini 2.5.
Google isn’t alone in seeing in AI brokers a chance to extract extra income from clients. Salesforce, for instance, has been extraordinarily bullish on the thought of enterprise interactions being powered by its imaginative and prescient of gen-AI-powered automatons. CEO Marc Benioff advised buyers final 12 months the corporate would possibly overcome a fall in buyer license numbers by charging customers for every dialog they’ve with a bot, which he mentioned was “a really excessive margin alternative” for Salesforce.
SaaS enterprise utility vendor Workday has additionally pushed AI brokers on its new platform, even seeing them as a method to reduce its personal human headcount.
In the meantime, large omni-vendor Microsoft can be within the agent sport with its 365 Copilot vary, powered by its $10-billion-plus alliance with ChatGPT maker OpenAI.
However what about customers, what are they getting out of it? In a latest analysis word, Gartner mentioned of us ought to negotiate Microsoft Copilot Studio merchandise with care as there’s potential for additional prices to come up on prime of direct licensing.
“Microsoft 365 Copilot Chat may have a major affect on the software program and cloud spend of Microsoft purchasers. Governance of utilization and price is required,” it mentioned.
Throughout all AI agent deplyments, customers wanted to concentrate on the complexity of calls for they may make on their cloud infrastructure.
Yrieix Garnier, merchandise veep at monitoring and observability vendor Datadog, mentioned brokers usually function throughout a number of platforms, comparable to Microsoft and Google. “Their habits can range considerably primarily based on enter, context, and chaining logic,” he mentioned. “A single consumer immediate can provoke a multi-step reasoning course of, set off API calls, or spawn different brokers — making utilization and price extremely dynamic and tough to foretell.
“To handle this complexity, organizations have to implement observability practices that transcend conventional monitoring. This consists of capturing and analyzing inputs, intermediate reasoning steps, and outputs for each agentic activity.
“Token utilization, latency, mannequin responses, and error charges should be tracked at a granular stage. Simply as with microservices, groups want traceability and context to know the place inefficiencies or errors are occurring — and what they’re costing. Guardrails are additionally vital: With out limits, brokers can enter loops or generate extreme downstream duties, resulting in uncontrolled spend.”
Google’s agent interface protocol is sort of compelling … Chances are high that that is extra advertising hype
Chirag Dekate, VP analyst at Gartner, mentioned Google was differentiating itself available in the market with its funding in TPUs, its potential to work together with knowledge sources, and the A2A protocol, which guarantees collaboration throughout brokers from totally different distributors.
“For those who’re attempting to create an agented enterprise of the longer term, you want to have brokers working with different brokers to unravel complicated duties and sophisticated actions. Right here, Google’s agent interface protocol is sort of compelling, as a result of it lets you have a number of brokers work collectively or constructively with each other,” he mentioned.
Nonetheless, limitations within the present generative-AI fashions means they’re removed from the extent of intelligence wanted to displace a human workforce. “Chances are high that that is extra advertising hype. That is giving enterprises false hope in accelerating human alternative, and we’re not there but,” he mentioned.
On the identical time, organizations could also be postpone investing within the imaginative and prescient of agent AI provided by Google, Salesforce, Microsoft, and others due to the financial uncertainty created by America’s international tariff struggle, which has wiped trillions of {dollars} off international inventory valuations and prompted warnings of worldwide recessions.
“We’re beginning to see early indicators of enterprises shifting in the direction of price optimization, versus final 12 months’s focus round gen-AI. Companies will proceed to put money into gen-AI however price optimization will probably take a way more severe tone this 12 months spherical, as a result of all the pieces is altering. AI adoption charges will endure, however AI innovation won’t endure as a result of the cloud and hyperscalers should not slowing down. They’re doubling down on investments,” Dekate mentioned. ®
Cloud Subsequent This week Google joined a throng of tech distributors pushing the idea of “agentic AI” on an unsuspecting and maybe unreceptive assortment of enterprise customers. Questions stay about how efficient this tranche of instruments will likely be at fixing enterprise issues and the way a lot it would all price.
At its annual Google Cloud Subsequent bash in Las Vegas this week, the Chocolate Manufacturing unit opened the floodgates to a slew of product information.
Among the many bulletins was a promise to introduce an Agent Improvement Equipment (ADK), an open-source framework Google mentioned would simplify the method of constructing enterprise software program that integrates AI brokers – question-answering and task-performing software program that makes selections and types its output utilizing giant language fashions. Brokers will be seen as software program that talks to different software program.
Google claims its ADK will permit clients to construct an AI agent in below 100 strains of hopefully intuitive code. The ADK additionally comes with an “agent backyard” of pre-built bots and instruments, in addition to greater than 100 pre-built connectors to widespread knowledge sources.
In the meantime, the search and adverts large is about to introduce a brand new protocol designed to assist brokers from totally different distributors discuss to 1 one other. The Agent2Agent (A2A) protocol has attracted 50 companions signing up, comparable to Accenture, Field, Deloitte, Salesforce, SAP, ServiceNow, and TCS, which it mentioned had been actively contributing to it.
That is giving enterprises false hope in accelerating human alternative, and we’re not there but
Digging into the main points, Google’s agent backyard is a pre-packaged assortment of instruments inside the ADK permits customers to entry 100+ pre-built connectors, customized APIs, integration workflows, or knowledge saved inside Google cloud programs like BigQuery and AlloyDB. Google cited regulation agency Freshfields as an adopter of its agent imaginative and prescient. The smaller biz mentioned it could use Google’s enterprise knowledge platform for AI, Vertex AI, to create bespoke AI brokers for its authorized and enterprise processes, whereas additionally utilizing search-bots to deliver collectively company-held info.
To energy language processing and knowledge gathering for customers, Google introduced the seventh technology of its Tensor Processing Unit, aka TPU, specialist AI-accelerating matrix-math {hardware} first used internally at Google.
The chips code-named Ironwood – which you’ll learn extra about on The Reg right here and The Subsequent Platform right here – have greater than 10 occasions the efficiency of the sooner Trillium TPU. A completely decked out pod of Ironwood TPUs, rented from Google Cloud, can hit 42.5 exaFLOPS of FP8 compute, “assembly the exponentially rising calls for” of generative fashions comparable to Google’s just lately introduced Gemini 2.5.
Google isn’t alone in seeing in AI brokers a chance to extract extra income from clients. Salesforce, for instance, has been extraordinarily bullish on the thought of enterprise interactions being powered by its imaginative and prescient of gen-AI-powered automatons. CEO Marc Benioff advised buyers final 12 months the corporate would possibly overcome a fall in buyer license numbers by charging customers for every dialog they’ve with a bot, which he mentioned was “a really excessive margin alternative” for Salesforce.
SaaS enterprise utility vendor Workday has additionally pushed AI brokers on its new platform, even seeing them as a method to reduce its personal human headcount.
In the meantime, large omni-vendor Microsoft can be within the agent sport with its 365 Copilot vary, powered by its $10-billion-plus alliance with ChatGPT maker OpenAI.
However what about customers, what are they getting out of it? In a latest analysis word, Gartner mentioned of us ought to negotiate Microsoft Copilot Studio merchandise with care as there’s potential for additional prices to come up on prime of direct licensing.
“Microsoft 365 Copilot Chat may have a major affect on the software program and cloud spend of Microsoft purchasers. Governance of utilization and price is required,” it mentioned.
Throughout all AI agent deplyments, customers wanted to concentrate on the complexity of calls for they may make on their cloud infrastructure.
Yrieix Garnier, merchandise veep at monitoring and observability vendor Datadog, mentioned brokers usually function throughout a number of platforms, comparable to Microsoft and Google. “Their habits can range considerably primarily based on enter, context, and chaining logic,” he mentioned. “A single consumer immediate can provoke a multi-step reasoning course of, set off API calls, or spawn different brokers — making utilization and price extremely dynamic and tough to foretell.
“To handle this complexity, organizations have to implement observability practices that transcend conventional monitoring. This consists of capturing and analyzing inputs, intermediate reasoning steps, and outputs for each agentic activity.
“Token utilization, latency, mannequin responses, and error charges should be tracked at a granular stage. Simply as with microservices, groups want traceability and context to know the place inefficiencies or errors are occurring — and what they’re costing. Guardrails are additionally vital: With out limits, brokers can enter loops or generate extreme downstream duties, resulting in uncontrolled spend.”
Google’s agent interface protocol is sort of compelling … Chances are high that that is extra advertising hype
Chirag Dekate, VP analyst at Gartner, mentioned Google was differentiating itself available in the market with its funding in TPUs, its potential to work together with knowledge sources, and the A2A protocol, which guarantees collaboration throughout brokers from totally different distributors.
“For those who’re attempting to create an agented enterprise of the longer term, you want to have brokers working with different brokers to unravel complicated duties and sophisticated actions. Right here, Google’s agent interface protocol is sort of compelling, as a result of it lets you have a number of brokers work collectively or constructively with each other,” he mentioned.
Nonetheless, limitations within the present generative-AI fashions means they’re removed from the extent of intelligence wanted to displace a human workforce. “Chances are high that that is extra advertising hype. That is giving enterprises false hope in accelerating human alternative, and we’re not there but,” he mentioned.
On the identical time, organizations could also be postpone investing within the imaginative and prescient of agent AI provided by Google, Salesforce, Microsoft, and others due to the financial uncertainty created by America’s international tariff struggle, which has wiped trillions of {dollars} off international inventory valuations and prompted warnings of worldwide recessions.
“We’re beginning to see early indicators of enterprises shifting in the direction of price optimization, versus final 12 months’s focus round gen-AI. Companies will proceed to put money into gen-AI however price optimization will probably take a way more severe tone this 12 months spherical, as a result of all the pieces is altering. AI adoption charges will endure, however AI innovation won’t endure as a result of the cloud and hyperscalers should not slowing down. They’re doubling down on investments,” Dekate mentioned. ®
