The CIO’s Guide To Agentic AI: How To Move From Experiments To ROI |
The latest must-have tech isn’t just AI. It’s AI agents. And while it sounds great in principle to have AI automate routine tasks, many companies are stuck on what that actually looks like. It’s far too powerful a technology to be used for sending personalized emails on your day off, but it’s probably not going to replace swaths of your employees on its own.
Agentic AI as a technology is becoming more formalized—the National Institute of Standards and Technology has even started drafting standards to govern its use—and it’s time for companies to formalize how they use it. I spoke with David Brudenell, co-CEO and executive director of global agentic AI firm Decidr, about how a CIO can move AI agents from the testing stage to driving efficiency at scale. An excerpt from our conversation is later in this newsletter.
This is the published version of Forbes’ CIO newsletter, which offers the latest news for chief innovation officers and other technology-focused leaders. Click here to get it delivered to your inbox every Thursday.
The announcement that the tech world has been buzzing about for some time officially came this week: Tim Cook is stepping down as Apple CEO and will become its executive chairman on September 1. Apple’s next CEO will be John Ternus, who has been the company’s senior vice president of hardware engineering since 2013. Ternus joined Apple in 2001, and the company said he was instrumental in several of Apple’s signature products, including the iPad, AirPods, several iPhone models and the company’s new MacBook Neo.
Cook, who was personally recruited by Steve Jobs, has led Apple for 15 years and is best known for turning the company into a global business juggernaut. Under Cook, Apple’s market cap skyrocketed above $4 trillion—a 1,932% stock value increase, notes Forbes senior contributor Peter Cohan—and it’s currently the world’s third-largest company. Cook oversaw releases of new devices that people were excited about, as well as the move to make Apple a successful media company through streaming and news services, writes Forbes senior contributor Andy Meek.
But Ternus has a big challenge in front of him as he takes the helm of Apple: AI. Forbes senior contributor Rachel Wells writes that the company’s more measured approach to consumer AI puts it at risk of losing relevance in tomorrow’s tech sphere. And Ternus is a hardware-and-product leader, which gives him grounding in some areas Apple could focus on to outpace competitors. But the big question is how well Ternus can drive the kind of groundbreaking innovation for which Apple has been known.
However, Forbes contributor Sandy Carter writes that Apple is far ahead in one area of AI development: it builds its own chips and owns its compute space, unlike every other frontier AI company. So while it doesn’t have the applications just yet, Apple controls its own silicon—and is not competing with other companies for bandwidth or chips. That capability, Carter writes, was largely built by Johny Srouji, who has been named Apple’s chief hardware officer in the transition. This role didn’t exist previously, but it brings together the all-important areas of hardware engineering and technologies under one executive. It also cements the partnership between Ternus and Srouji—meaning the underlying hardware and tech is likely to be strong as Ternus takes the helm.
Everyone knows that vibe coding is a huge business, but with its recent deal with Cursor, SpaceX is putting a value on it. Announced this week, Elon Musk’s space exploration and AI company now has the option of buying the vibe coding startup outright for $60 billion later this year, or paying it $10 billion to continue working together. Cursor, founded in 2022 by four friends from MIT, evolved from a popular code editor into an AI coding business, Forbes’ Anna Tong and Rashi Shrivastava wrote in a feature on the company last month. Cursor is striving to be the code writing application of choice by training on specialized data sets and building its platform on several open source models. More than 1 million developers use Cursor daily.
The partnership with SpaceX gives Cursor an infusion of cash and a huge vote of confidence from one of technology’s most powerful leaders. But Forbes contributor Sandy Carter writes there is a bigger win for both SpaceX and Cursor here. When SpaceX merged with Musk’s xAI earlier this year, it gained the company’s Colossus supercomputer—which needs a high-level product to route traffic through it. Cursor fits the bill nicely. The loyalty of developers who use Cursor and the business acumen of its founders also bring more value to SpaceX. And with SpaceX preparing for a huge IPO later this year, the company’s valuation—and power—could be destined for the stratosphere.
ARTIFICIAL INTELLIGENCE
Google is advancing its AI chip strategy, releasing two distinct TPU chips for different functions at its Google Cloud Next event, writes Forbes senior contributor Maribel Lopez. The TPU-8t is a chip designed to be a training workhorse, while the TPU-8i is an inference and agent engine. The differentiated chips, Lopez writes, shows that connectivity between chips and specialized performance is becoming a variable in how to design AI infrastructure. And while enterprises don’t buy TPU chips, Google’s announcement shows how system design is changing to handle evolving technologies—bringing different variables into play when making decisions about AI, like energy consumption, reliability and value.
How To Set Your Agentic AI Strategy
As agentic AI capabilities expand and more big SaaS providers add agents to their platforms, more companies are adopting them. But are they using them in a way that actually harnesses the technology’s power, or are they bringing in the hot new tech to do things that aren’t really worth the compute?
I spoke with David Brudenell, co-CEO and executive director of agentic AI firm Decidr, about how companies should be thinking about and deploying AI agents. This conversation has been edited for length, clarity and continuity.
Where are companies now in terms of AI agent adoption? Are they looking at simple tasks or deploying them to things that could actually make a difference to business?
Brudenell: Of the CIOs and businesses I’ve talked to, boards are saying, ‘You must use AI,’ and their current solution is, ‘Let’s get a corporate account of Claude or OpenAI.’ I spoke to a multi-billion-dollar company based out of New York. On the surface, they claim that their staff have created 1,800 agents. Of the agents that have been created, only 50 are actually used, and of the 50, only four are actually used with any regularity.
Most businesses are experimenting. It’s the probabilistic, such as summarizing documents, creating images. To really start to use AI properly, you need it to be deterministic. Enterprise businesses are deterministic. You need to do the same thing every single time in the same way, and foundation models cannot produce that. Interoperability exposes enterprise businesses more, because that means AI is using probabilistic and transacting at a higher rate. It increases the number of holes in your boat, so to speak.
Businesses are huge workflow engines that go through process steps. And tacit knowledge is the most valuable part of a business, but it’s not well understood. You can’t put an AI in it; you don't really know how work actually gets done. Microsoft said enterprise sovereignty is the most important thing that businesses need to focus on: How you do the thing inside your business. That’s the difference between one law firm deciding to litigate versus another law firm deciding to settle: Same case, but their tacit knowledge is how they figure out which way to go. That’s also pretty critical for understanding where to start with AI.
How should enterprises be figuring out the best way to use AI agents?
AI does not replace jobs. AI replaces tasks in workflows. Businesses are starting to understand that, and that’s the process they’re going through now: Starting to look at, ‘How does my organization look underneath the hood?’ Google codified the click and turned attention into a monetizable thing. AI is going to monetize the task, because we’re literally productizing knowledge.
Enterprises will start to understand what tasks are done in their organization. Then they’re going to say, ‘What is the price for those tasks?’ We’ve been through this journey before. Before AI, it was robotic process automation. Before robotic process automation, it was BPOs [business process outsourcing]. Before BPOs, it was management consultants. Before management consultants, it was the professional manager. We’ve been on the curve of labor cost reduction. AI is another way of reducing labor costs.
As they start to understand where tasks leverage in their organization, there will start to be more pronounced, coordinated activity to replace with AI. That’s why we’re seeing some of the tech titans do mass layoffs in their engineering department because engineering is all language. Writing code is language; literally artifacts of tasks coming out. That’s why it’s one of the first that's being replaced.
Also among the first ones that will be replaced will be in sales and marketing, because those are lowest risk, highest return. When you think about risk, compliance, payroll, no CIO is going to put a startup agent in their payroll process—If the employees aren’t paid, that’s like a nuclear bomb going off in your org. You are not going to replace your accounting system. Versus: Will you change your project management system? Yeah, that’s easy to replace.
What advice would you give to a CIO who is trying to chart the best path for their company to be able to adopt agentic AI in a smart and useful way?
Before you know where you want to go, you need to know who you are. Start to capture the tacit knowledge in your organization because that is the asset of businesses in the future: how you do things.
Tacit knowledge capture leads to task understanding. Task understanding begins the quantitative exercise of assessing where the opportunity is, because tasks have cost and you can attach them to your P&L.
Now you have a quantitative foundation to assess risk, look at ROI and see how your workforce may be reshaped based on the agentic decisions that you make in your business. That’s better to take to a board than, ‘Our sales team now has automated emails, but they’re not actually making more sales.’
Related to their CEO handoff that we covered earlier, consumer technology firm Apple named Johny Srouji as chief hardware officer, effective April 20. Srouji began at Apple in 2008 and is moving into the expanded role after serving as senior vice president of hardware technologies. He will succeed John Ternus, who was tapped for the CEO role.
HR platform provider Alight will appoint Naveen Baweja as its chief technology officer, effective April 29. Baweja previously worked as vice president of technology at The Walt Disney Company and he succeed Deepika Duggirala.
Healthcare staffing firm Cross Country has hired Chris Tyrell as chief information officer, effective April 21. Tyrell joins the firm from Eclipse Advantage where he worked as chief technology officer.
Nobody will argue that clean data makes systems run more smoothly. But before investing time and money in data cleaning, you should figure out how valuable the data is in the first place. If it’s not data that can actually make a difference, it may not be worth the effort.
Stress isn’t just an issue for you. Leaders set the emotional tone for their teams, meaning your own trip toward burnout could be creating a toxic environment for your employees. Here are five ways to manage stress—both your own and your team’s.
Google announced it’s creating a dataset based on an unconventional metric it will collect from its offices. What is it?
B. Shoes worn by employees
See if you got the answer right here.