AI alone won\u2019t transform your business. It has the potential to speed up workflows, improve CX and sales, perform useful research in seconds, and ease employees\u2019 workloads\u2014but only if it\u2019s thoughtfully designed and implemented. <\/p>\n
Many workplaces use enterprise tools\u2014AI and otherwise\u2014provided by Google Cloud in some capacity: They’re largely accessible and ubiquitous, and the company designs and refines them to perform the functions users need. I spoke with Karthik Narain, Google Cloud\u2019s chief product and business officer, about how he works with customers, data and technology to design the enterprise software they need. I also got some advice on figuring out ways to work with data, AI and software within the enterprise, creating pathways for employees to get the most out of technology. An excerpt from our conversation is later in this newsletter.<\/p>\n
Until next time. <\/p>\n
This is the published version of Forbes\u2019 CIO newsletter, which offers the latest news for chief innovation officers and other technology-focused leaders. Click here<\/a> to get it delivered to your inbox every Thursday.Artificial Intelligence<\/p>\n Photo Illustration by Avishek Das\/SOPA Images\/LightRocket via Getty Images<\/p>\n Two new AI models from China have been announced or launched in the last week, promising users inexpensive and accurate performance of tasks. DeepSeek\u2019s V4 Flash<\/a> became available last Friday, writes Forbes senior contributor Rachel Wells, highlighting its low cost and intelligence. V4 Flash has an intelligence score of 50\u2014equivalent to Google Gemini 3.6 Flash\u2014and a cost of 3 cents per benchmark task\u2014105 times cheaper than Anthropic\u2019s Claude Fable 5, which is $3.15 for a similar task. However, Wells writes, this doesn\u2019t mean the new DeepSeek is the solution for budget-conscious IT departments. It has a 37% accuracy rate and an 84% hallucination rate.<\/p>\n Alibaba\u2019s Qwen3.8-Max AI model<\/a>, which will be available next week, sounds more promising on paper. It will be one of the world\u2019s largest open AI models at 2.4 trillion parameters, and Alibaba says it will be able to perform long, complex tasks. Alibaba highlighted Qwen3.8-Max\u2019s coding ability, saying it can spend hours on code without needing human input, and that it spent 16 days writing and improving a coding tool. Forbes contributor Jon Markman writes<\/a> this model\u2019s price undercuts competitors, at $2 per million input tokens and $6 per million output\u2014about 40% of the input price for Anthropic\u2019s Claude and a quarter of its output price.<\/p>\n The big question: Can Qwen3.8 Max deliver for its lower price? Accuracy studies have not yet been published\u2014though Alibaba has said it is comparable to Anthropic Fable 5 and OpenAI\u2019s GPT5.6-Sol. However, unlike Anthropic and OpenAI, Alibaba\u2019s models are open\u2014meaning developers can download, run and modify them. <\/p>\n Meanwhile, other leading AI companies are responding to the cost pinch. Last week, Wells writes<\/a> OpenAI cut the cost of GPT-5.6 Luna by 80%, while GPT-5.6 Terra is now 20% less. Prices for the top-tier GPT5.6-Sol stayed the same. But those numbers may not mean much to many companies, which may not be doing a great job of tracking AI costs. Wells notes that a recent study shows about 29% of organizations attribute more than a quarter of their cloud spend to AI, while more than two in five only review AI costs quarterly. As AI price wars begin, it is past time for tech and finance departments to figure out just what\u2019s being spent\u2014and which vendors are needed to provide necessary services. <\/p>\n Stock Market News<\/p>\n Big tech and AI are once again taking the stock market to new heights\u2014sort of. The S&P 500 and Dow Jones Industrial Average both hit record highs this week. Part of that boost came from optimistic talk out of Washington, D.C. about the war in Iran, but part of it came from strong tech earnings.<\/p>\n Palantir\u2019s stock<\/a> surged nearly 30% this week, after the company reported \u201cotherworldly\u201d gains. These include a 93% year-over-year increase in overall revenue\u2014with a 149% boost in commercial revenue and a 90% surge in government revenue. Amazon\u2019s stock saw one of its best days in a decade last Thursday, after a blockbuster earnings report<\/a> showed net sales up 20% year-over-year to $200.6 billion. The biggest growth sector was its AWS business<\/a>, writes Forbes senior contributor Peter Cohan, with 37.6% revenue growth\u2014the fastest in 18 quarters. <\/p>\n But this doesn\u2019t mean investors are past fears of AI\u2019s high capital expenditure costs. In its first-ever quarterly report on Wednesday, SpaceX reported spending $15.8 billion in AI-related capital expenditures\u2014far above the $13.2 billion estimate and marking a near sixfold increase in this year\u2019s capex compared to last year\u2019s. Elon Musk\u2019s space exploration and AI company saw its stock take a 13% hit following its earnings report. Forbes senior contributor Peter Cohan writes<\/a> SpaceX saw AI-related revenue increase 247%, which didn\u2019t seem to faze investors. But Nvidia got a 3% stock boost<\/a> after SpaceX said it plans to only use Nvidia chips on its planned future orbiting AI data centers.<\/p>\n Bits + BytesLessons In AI Product And Application Design From Google Cloud<\/p>\n Google Cloud Chief Product and Business Officer Karthik Narain.<\/p>\n Google Cloud<\/p>\n Karthik Narain\u2019s title is Google Cloud chief product and business officer\u2014something that sounds simple, but encompasses a vast amount. He is in charge of product and engineering at Google Cloud, as well as customer-facing go-to-market applications\u2014basically building products for customers and ensuring their success. I talked to him about the product design function and how to use AI to actually improve what employees do.<\/p>\n This conversation has been edited for length, clarity and continuity.<\/p>\n How do you use customer feedback and AI-generated insights to improve your products and services? <\/p>\n Narain: I\u2019ve always started with: What is the customer problem we\u2019re trying to solve? I\u2019m loosely using the word problem: There\u2019s always a hammer and there\u2019s always a nail. The nail is the problem or the opportunity. The hammer is the solution for the opportunity. We need to look at it from both perspectives:<\/p>\n It\u2019s very important that we don\u2019t find one problem for one customer and build a technology solution just to solve that. Customizing it for one customer is not reusable. We look at a lot of these inputs, find the signal from the noise, and find a solution. <\/p>\n [Everyone is worried about] the risk for AI to be used for cyber attacks. There was this perception there is one universal \u201csuper set\u201d of all models to find cyber vulnerabilities. When we did our test, we found out there\u2019s no one \u201csuper set.\u201d Most importantly, it\u2019s not about finding the problems. It\u2019s about fixing it. How do you fix it at the same pace at which you are finding it? We created a solution called CodeMender. It does what the name says: mends the code when it sees the vulnerability. <\/p>\n We launched our AI agentic defense platform. Using Wiz, it created a harness where it will use multiple models to find the postures of attack. You can use CodeMender to fix the code and then you use Wiz’s red agents to simulate attacks. Using Google SecOps, you\u2019re constantly monitoring what is happening in the environment. While all of this noise was going on about the digital world, we tuned those products and created an end-to-end platform to solve that.<\/p>\n When you are releasing new AI products, how do you launch them to enterprises in a way that they can actually use them?<\/p>\n We take a three-part approach. The real world is brownfields, and you need to build solutions that work in the existing setup. Waymo is running in a city or town that was built [before driverless cars]. How do you work to build a product that can integrate with existing architecture? <\/p>\n Number two is that you [solve problems for] companies pushing the envelope on what needs to be done. Our forward-deployed engineers take our product that we believe is 80% ready and extend it\u2014not just customize. Extensions are things we would bring back to the core of the product, and are maintainable over a long period of time. Our customers need not spend additional money to maintain that; it\u2019s part of the product they are buying from us. We are using those FDEs to help customers ground our products in their environments and make them work. Sometimes we recognize that there is change management needed, so we will use partners to do change management within our customer\u2019s organization. <\/p>\n The third element is we try to take a platform approach to everything so it\u2019s not a single-project solution. How do you bring agility? At the same time, how do you create stability? <\/p>\n That is how the Gemini enterprise agent platform got created. The starting problem was, \u2018I need to be able to create agents to do X, Y and Z.\u2019 Within 60 days, they came back saying, \u2018We are having an agent jungle. I don\u2019t know how we are going to maintain who\u2019s accessing what data, and how many of these are duplicate agents?\u2019 <\/p>\n Our platform is built for that. When somebody is building an agent, it can go into production only if it goes through Agent Registry. There\u2019s Agent Gateway, where your organization policies, governance and security guidelines are adopted. And now we\u2019ve created Agent Discoverability, where you can say, \u2018I\u2019m trying to do this using an agent,\u2019 and it\u2019ll immediately say, \u2018One of your colleagues already built an agent that sounds like what you\u2019re trying to do.\u2019 <\/p>\n What advice would you give a CIO that is trying to take enterprise needs and technology to create systems at their workplace that works for them?<\/p>\n You need a top-down vision mandate from the organization. The good news is AI is the first technology wave that business execs and technology execs are equally interested in. This technology can truly unlock reinvention, value beyond saving cost. Identify your top priorities, and balance between growth versus efficiency priorities. I would say do it top down and prioritize growth and efficiency in a 60 \/ 40 ratio.Know where you are starting and make a goal. Take a platform approach so your second, third and 13th project are not equally hard and equally expensive. You are putting miles behind you so that each incremental project is shorter, more value creating, less expensive, and so on. You also need a bottom-up approach because there is a change by management element. Change is not going to happen by a strategy provider doing a workshop or an exercise. People will have to start using AI themselves. Having access to it can automate their daily routine or start using agentic workflows created at an enterprise level to take their daily job into new ways of doing things. Strategies + Advice<\/p>\n While AI is changing how we do everything, there\u2019s one thing that needs to keep a human touch: Communication. Here\u2019s how to walk the line<\/a> between using AI to help with writing speeches and emails and preserving your personality.<\/p>\n As business becomes more complicated, so does decision-making. Here are some tips<\/a> to help you make the right decisions faster.<\/p>\n Quiz<\/p>\n Microsoft paid out a record amount to people who found system vulnerabilities in the last year. How much was the company\u2019s total bug bounty payout?<\/p>\n A. $17 million<\/p>\n B. $19 million<\/p>\n C. $20 million<\/p>\n D. $22 million<\/p>\n