By Adam Mills
AI Editor
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.

Newsweek
Editor's note: This is AI Impact, Newsweek's weekly newsletter where each week, we will explore how business leaders are unlocking real value through artificial intelligence.
Read More on AI
Tap or click here to get this newsletter delivered to your inbox.
Good morning and thanks for joining me.
In the weeks ahead, several Newsweek webinars will take on a question I keep hearing from folks in different forms: What has to change inside an organization once AI moves beyond experimentation?
On September 17, Dr. Ranjit Tinaikar will sit down with Genpact President and CEO Balkrishan “BK” Kalra for “The Inflection Point: How Agentic Operations Compound Enterprise Value.” Kalra’s argument is that companies will increasingly move from work that is processed and validated by people toward work that is processed by machines and validated by people. They’ll explore what that means for operating models, jobs, controls and the economics of how work gets delivered.
Then, on October 21, Gabriel Snyder, Newsweek’s executive editor, enterprise, will moderate “Overcoming Barriers to AI Transformation in Legacy Industries,” presented by Cognizant. Matt Sanchez, chief operating officer at Yahoo, will join a panel examining how established businesses connect AI to existing systems and workflows, manage governance and risk and turn promising pilots into measurable results.
We also have a session outside the enterprise series that I think many of you will find interesting. On September 3, my colleague Alexis Kayser, Newsweek’s health care editor, will moderate “What Makes a Smart Hospital?” with leaders from Statista, NYU Langone Health, Samsung Medical Center and Sheba Medical Center. They’ll look at what separates individual technology deployments from a genuinely integrated digital strategy, drawing on Statista’s Smart Hospitals Maturity Survey.
I’d love to have you join us for any—or all—of them.
The three sessions start from different places, but they all come back to a similar challenge: Organizations can make AI available much faster than they can develop the skills, processes and judgment needed to use it well.
This week’s Signal Capture looks at just how wide the gap between access and proficiency can be.
Signal Capture
Signals from the frontlines of AI adoption
KPMG Tax Vice Chair: Only 5 Percent of Users Show Sophisticated AI Behaviors

Newsweek/Getty Images/Photo-illustration by Newsweek/Getty
KPMG is evolving how its tax teams are organized to include three kinds of AI expertise: power users, “makers” who create reusable prompts, workflows and agents and “builders” who turn those ideas into broader solutions.
Rema Serafi, vice chair for tax at KPMG LLP, told Newsweek the structure is part of how the firm is preparing for AI to become embedded across tax workflows. She drew a distinction between widespread AI adoption and actual proficiency.
KPMG research conducted with the University of Texas at Austin found sophisticated use was uncommon: only 5 percent of users consistently demonstrated highly sophisticated AI behaviors.
“Routine AI is typically task-based. It looks like prompting a chatbot to summarize a document, answer a question, or generate content outside the workflow. Sophisticated AI use is workflow-based,” Serafi said.
The more sophisticated users described in the research treat AI as a thought partner, using it to explore ideas, test hypotheses and refine their thinking. They give it the context, constraints and objectives needed to delegate complex, multi-step assignments effectively and apply it across analysis, problem-solving, technical guidance and innovation.
Serafi said AI can take on more repetitive, process-driven tax tasks, giving professionals more room to interpret information, challenge assumptions, manage risk and identify opportunities for the business. As access to the technology becomes more common, organizations will increasingly distinguish themselves by how effectively they combine it with curiosity, critical thinking and expertise.
KPMG’s Digital Gateway platform, powered by Anthropic’s Claude, is designed to bring AI into the same environment where KPMG professionals and clients already work with data, knowledge and governance. Serafi said the platform can support multiple steps in a workflow, including accessing approved data, using connected tools, identifying issues and routing work for review.
“AI can process huge volumes of data and handle many of the mechanics of tax work, but it still struggles with the nuanced, context-heavy aspects,” Serafi said.
Tax professionals may have to interpret complex or conflicting regulations in light of a client’s particular circumstances, recognize edge cases and gray areas that do not fit neatly within established rules. They may also have to account for context, constraints and concerns that emerge through client relationships but never appear in a document or data set.
Responsibility for those decisions remains with the person delivering the work.
“Accountability can't be delegated to technology. Clients trust people, not technology,” Serafi said. “The professional delivering the work remains responsible for the quality of the analysis and advice.”
KPMG is using simulation to give professionals more opportunities to practice making difficult decisions. Its TaxSIM program uses high-volume, real-world simulations designed to accelerate the development of judgment and decision-making skills that can otherwise take years to build.
“Traditional training is incredibly valuable, but judgment often develops through experience – and experience takes time,” Serafi said.
The simulations can expose professionals to client situations, unexpected developments and difficult choices before they encounter comparable circumstances in practice. Serafi described the approach as one part of a broader learning strategy that develops technical expertise, critical thinking and client advisory skills together.
The range of backgrounds leading into tax and accounting could widen as the work changes. Serafi expects people to enter from data science, computer science and other nontraditional fields as career paths increasingly cross tax, law, technology and data analytics. A strong technical foundation remains important, alongside the ability to use AI effectively and exercise judgment over how it is applied.
Serafi said the technology will also raise the baseline for knowledge work, increasing the importance of judgment, critical thinking, communication, relationship building and the ability to navigate complexity.
“Clients already have access to more information than ever,” she said. “What they value is perspective, context, and confidence in the decisions they make.”
Upcoming Webinars
Overcoming Barriers to AI Transformation in Legacy Industries

Newsweek
AI investment is accelerating, and long-established companies are finding ways to bring the technology into organizations shaped by years of customer relationships, complex operations, critical systems and regulatory responsibilities. The challenge is turning promising pilots into lasting business results without disrupting the strengths that have made those companies successful.
In an upcoming Newsweek AI webinar presented by Cognizant, Gabriel Snyder, Newsweek’s executive editor, enterprise, will moderate a discussion titled “Overcoming Barriers to AI Transformation in Legacy Industries.” The conversation will delve into how established organizations are connecting AI to existing systems and workflows, clarifying ownership and governance and aligning technology investments with people, processes and measurable business outcomes.
Matt Sanchez, chief operating officer at Yahoo, will join the conversation, with additional speakers to be announced.
Join the live discussion on October 21. Register for free.
The Inflection Point: How Agentic Operations Compound Enterprise Value

The next phase of enterprise automation is moving beyond tools that assist employees and toward systems that can execute parts of a business process themselves. That raises bigger questions for leaders about where people remain accountable, how operating models need to change and what happens to the economics of work when machines take on more of the execution.
On September 17, Dr. Ranjit Tinaikar, host of Newsweek’s “AI Impact Forum,” will sit down with Balkrishan “BK” Kalra, president and CEO of Genpact, for “The Inflection Point: How Agentic Operations Compound Enterprise Value.”
Kalra will bring Genpact’s perspective on what it calls “Agentic Operations,” including lessons from its own implementation and research into why some enterprise investments produce value while others stall.
They’ll get into the implications for jobs and roles, responsible AI controls, operating models and commercial structures as companies give autonomous systems a larger role in how work gets done.
Join the live discussion on September 17. Register for Free.
Prompt Injection
What’s one recent insight you’ve learned about AI?

...
“One of my biggest realizations about AI is that it fundamentally changes the relationship between humans and technology. For decades, we've asked people to learn the language of machines. We learned operating systems, programming languages, menus, commands, and increasingly complex interfaces to get things done.
AI flips that model. For the first time, technology is learning to understand us.
When I look at the history of computing, every major breakthrough has been an abstraction layer that hides complexity and makes technology accessible to more people. We moved from machine code to programming languages, from command lines to graphical user interfaces. AI agents are the next step in that evolution because they remove the need to learn the interface itself.
Instead of asking, ‘Which application do I need?’ users can simply express what they want to accomplish. Technology then understands their intent and takes action. To me, that's a profound shift and one that will make computing more intuitive, accessible, and valuable for everyone.
It's reinforced my belief that the most important role for AI is not replacing human creativity or decision-making. It's removing friction between intent and execution.
I increasingly think of AI as a powerful intermediary that can help people focus on outcomes rather than mechanics. Whether you're writing software, analyzing data, planning a trip, or creating content, too much effort is often spent translating an idea into the specific steps required by technology. AI can dramatically reduce that burden.
I see this clearly in software development. Historically, developers spent significant time translating concepts into syntax, frameworks, and implementation details. AI allows them to spend more time thinking about the problem they're solving and less time on the mechanics of expressing it.
More broadly, I believe conversational interaction will become a primary computing interface. The real opportunity isn't making technology more sophisticated. It's making sophisticated technology easier for more people to use. That's the lens through which I think about AI today.”
Have your own lesson to share? Email us at: a.mills@newsweek.com
Run Log
AI use case of the week
Neurometric Cut Resume-Screening Costs With a Smaller AI Model

...
A customer came to Neurometric with a resume-screening workflow that cost about $40,000 a year and was roughly 70 percent accurate, which the company said still required substantial human review.
Rob May, co-founder and chief executive officer of Neurometric, a company that helps businesses evaluate AI workloads and run them on task-specific language models, said the bill was also growing 30 to 40 percent month over month.
May’s team ran the customer’s actual production workload through a range of smaller models built for narrower tasks. The company compared how well each one screened resumes against the accuracy level the customer required.
“I always like to compare this to hiring someone with three PhDs to work a cash register,” May said.
The evaluation found that a smaller language model could handle the screening work. After the switch, May said the monthly cost fell to $250 and accuracy rose to 98.5 percent.
May attributed the improvement to using a model trained specifically for this kind of screening rather than a general-purpose model built to handle a much wider range of tasks. He did not identify the customer or provide the size of the evaluation set or details on how accuracy was measured.
Getting the first version of an AI workflow running is still necessary, May said. Companies can run into trouble when they keep adding work to the model they chose at the beginning without testing whether a narrower one can meet the same requirements for less.
Have an interesting AI use case to share with us? Email us at: a.mills@newsweek.com
Context Window
■ OpenAI is awarding $1 million in grants and up to $1 million in API credits to 14 independent projects examining how AI could reshape work, taxation, energy, health care, scientific research and societal resilience. [OpenAI]
■ Economists remain divided over AI’s role in the weak entry-level job market, with Stanford research finding a 16 percent relative employment decline among young workers in AI-exposed roles while other studies point to remote work and broader hiring trends as bigger factors. [NPR]
■ Gartner predicts AI inference costs per agentic workflow will increase more than fivefold through 2028 as increasingly complex agents consume more tokens, even as the cost of individual tokens continues to fall. [Gartner]
■ A survey of more than 1,500 developers found that 54 percent of senior developers think AI is making junior roles less relevant, raising concerns that short-term productivity gains could weaken the pipeline for developing future senior technical talent. [World Economic Forum]
■ An Intel-commissioned survey of 800 senior respondents found that 60 percent expect their organizations to operate robot fleets within five years, but only 40 percent currently have a formal strategy for managing a mixed human-robot workforce. [Intel]
Transfer Protocol
Tracking executive moves across the AI landscape
Anthony Jules, co-founder and CEO of Robust.AI, has moved into the chief innovation officer role, where he will lead strategic physical AI initiatives around the company’s warehouse robotics platform.
Sam Liu, previously an associate partner in Bain & Company’s AI practice, has joined Great Hill Partners as director of AI, helping develop AI capabilities across the firm’s investment process and portfolio companies.
Brian Schaad, previously an engineering leader at Consensus, has been appointed chief technology officer at ServiceTrade, where he will lead engineering and scale the architecture behind the company’s AI products.
Jason Forget, most recently president and chief revenue officer at Cockroach Labs, has joined Resolve AI as president and founding chief revenue officer, overseeing the company’s global commercial organization as it expands its agentic AI platform for production software.
Manish Kalia, formerly vice president of product management at Ping Identity, has joined Cyberhaven as chief product officer, shaping the company’s product strategy for the agentic enterprise and its approach to using AI to secure AI.
Know someone on the move in AI? Send job change info to a.mills@newsweek.com
Magic Moment
What’s the most fun or unexpected way you’ve used AI lately?

...
“During a recent Regard implementation, I was building out a service line's documentation workflow. We were rapidly iterating on her requested adjustment and tweaking multiple things in real time.
She said, ‘I wish my Epic IS tickets were resolved this quickly.’ So I asked her to walk me through her pending requests, to see how many we could solve on the spot with our AI.
We finished her entire list (and the documentation workflow customizations) before the meeting was over. Some of those requests had been sitting in a queue for more than a year. Typically, in the hospital, the answer to a good idea has been a ticket, a queue, and a wait measured in quarters. In one hour, with a little AI prompting, that wait was gone. She was ecstatic.”
Experience some AI magic? Tell us about it at a.mills@newsweek.com
Tap or click here to get this newsletter delivered to your inbox.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines