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.
Signal Capture
Signals from the frontlines of AI adoption
Splitit CEO Says Checkout Is AI Shopping’s Next Test
...
By Adam Mills
AI shopping is getting better at helping consumers find things. Nandan Sheth says actually purchasing them is where the harder work begins.
Sheth, CEO of Splitit, a payments company that offers card-linked installment technology, said much of the current AI shopping conversation is still centered on discovery and recommendations. That part of the experience is improving as consumers get better at prompting and large language models get better at understanding what shoppers want. Checkout is different.
The problem is that completing a purchase requires more than matching a shopper with the right product. The order has to get back to the merchant, the payment has to work, the merchant’s systems have to close the sale and someone has to handle disputes if the customer later says the agent was not authorized to make the purchase.
“There’s a huge repudiation problem,” Sheth told Newsweek. In conventional card payments, when a customer disputes a transaction, that can lead to “a chargeback,” and there is “a very defined process around that,” he said. With an autonomous agent, the issue becomes harder because the dispute may turn on whether the shopper clearly gave the agent permission to make the transaction.
Card networks including Mastercard and Visa, he said, are developing standards around liability and certainty for agent-driven transactions.
For consumer commerce, Sheth sees early progress in more complex purchases, where an agent can prepare the work but still leaves the final decision to the shopper.
“Imagine you’ve got an assistant that can get everything ready for you but won’t buy anything till you say yes,” he said.
That could apply to travel, where an agent might research a trip, assemble tours, flights and hotels, then present the package for approval. In business-to-business commerce, Sheth sees a stronger case for autonomous transactions, especially when companies are buying commoditized items with clear pricing, quality standards and delivery terms.
“I think there is a lot more need there than there is on the consumer side,” he said.
Consumer checkout brings all manner of complications. A human shopper can switch payment methods, pick another seller, change an address, request a refund or call about a chargeback. An agent has to know how to handle those moments without creating more friction or risk.
“The difficulty will be not in the happy path, but in the more exception-orientated path,” Sheth said. “Payment fails. I need to do a refund. How does that work?”
Merchants also have their own reasons to move carefully. Sheth said retailers learn from the signals shoppers provide while browsing and buying: how long they spend with a product, how many clicks it takes to reach checkout, whether they read reviews and where they drop off. If an agent becomes the main front door, many of those signals could become harder to see.
“The signals that they pick up from the commerce interaction, human to machine, in the majority are going to be lost,” he said.
That is why some merchants may try to pull shoppers back into their own checkout experiences or build systems that let their own agents communicate with consumer-facing agents. The goal is to avoid being reduced to a price, delivery time and rating inside someone else’s interface.
Payment certainty is another unresolved piece. In an agent-driven transaction, the merchant needs confidence that the payment will work at checkout and hold up after the sale.
“There are two parts, in my opinion, to certainty,” Sheth said. “Certainty of the payment going through from a consumer or shopper’s standpoint, and the certainty that that payment is not going to be repudiated from a merchant’s standpoint.”
Sheth said card-linked installments could fit that environment because they rely on a shopper’s existing credit card and available credit, avoiding the extra underwriting and account-registration steps that can come with traditional buy-now-pay-later products.
“The reduction in friction becomes really important because in many cases, these are automated transactions, either machine to machine or human to machine,” he said.
Even then, Sheth said consumers will need clear controls before agents can spend on their behalf. Those controls could include spending limits, ranges around those limits, merchant preferences and other restrictions based on the shopper’s values or goals.
“Consent [and] spending limits should absolutely be established upfront,” he said.
The milestone, in Sheth’s view, is when AI-driven commerce becomes a measurable share of online buying rather than a demo or experiment. He put the threshold at roughly 1 percent to 1.5 percent of e-commerce happening end-to-end inside an AI environment.
“That’s when it becomes real,” he said. “That’s when the inflection point happens.”
Core Intelligence
Wall Street’s AI Org Chart Still Needs a Human Boss
...
By Adam Mills
AI agents can handle pieces of analyst work before financial firms are ready to take accountability out of human hands.
Rogo’s pitch to financial institutions starts there. The company’s Felix workflow can help with work that has long filled junior bankers’ days: pitch materials, spreadsheets, company analysis, scenario work and acquisition research. Kevin Buehler, Rogo’s chief innovation officer and senior partner emeritus at McKinsey & Company, said the effect is already visible.
“I think this elevates the work that folks can do,” Buehler said during Newsweek’s recent “AI Impact Forum” webinar, “AI in Finance: From Individual Adoption to Enterprise Transformation.”
The impact reaches beyond faster analysts. If agents can handle more first-pass work, financial firms may be able to rethink the organizational design that work once supported. Buehler described a future where senior leaders remain at the top, a smaller layer of AI-capable professionals manages complex workflows and specialized agents handle more of the work at the base.
“Right now most firms look like pyramids because that’s what’s required to get the work done,” Buehler said. “We think we’re going to see a shift towards skyscrapers.”
The skyscraper image is useful because it keeps the promise and the danger in the same frame. AI could reduce the need for large teams built around manual production, while firms still need people who understand clients, markets, trade-offs and risk. Leaders have to decide whether the capacity created by AI becomes cost reduction, broader client coverage, better training or more time spent on judgment-heavy work.
Buehler pointed to DBS, the Singapore bank, as one example of the more ambitious path. He said the bank improved its technology, data and AI capabilities, then retrained people who were freed up and sent them after mid-market commercial banking customers. The result, he said, was revenue and share growth rather than a pure cost-cutting exercise.
Workflow redesign adds another layer. Buehler described an M&A sale process where AI could help gather documents, build models, prepare a confidential information memorandum, create a virtual data room, respond during due diligence and run parts of the bidding process. Connecting those steps is where the operating challenge begins.
Financial firms still need visibility into how AI-assisted work is produced. Buehler said Rogo documents include citations so every number can be traced back to its source, whether from public filings, market data or internal files. Bankers still have to decide whether the work makes sense.
“I think we’re all going to have to understand how to do that with AI as the input to the system, but not the final output,” he said.
The skyscraper model still carries a governance problem. If agents take on more of the junior work at the base, senior people will not only have fewer people beneath them but also more responsibility for what reaches clients. A flawed recommendation is unlikely to be treated as the agent’s mistake alone.
Wall Street may get a new shape from AI, but the work still needs someone willing to sign their name to the result.
You can watch the discussion and read the full article here: Rogo’s Kevin Buehler: AI Could Reshape Wall Street’s Pyramid.
Upcoming Webinars
Is India on the Right Side of the AI Trade?
India’s role in the global technology economy has long been shaped by scale: a massive IT services sector, fast-growing capital markets, digital public infrastructure and a young workforce. AI could raise the stakes even further by changing where value is created, who captures it and how countries compete.
In an upcoming “AI Impact Forum” session, Dr. Ranjit Tinaikar speaks with Shri Ashishkumar Chauhan, CEO of the National Stock Exchange of India, about whether India is positioned to benefit from the AI revolution—and what the technology could mean for the country’s economy, financial markets, market infrastructure and next stage of growth.
Join the live discussion on Thursday, July 23, at 10 a.m. Eastern. Register for free.
Prompt Injection
What’s one recent insight you’ve learned about AI?
...
"AI accountability is top of mind as enterprises wrestle with the human element of these evolving solutions. The questions of ‘who owns the system’ or ‘who is responsible for this AI’ are becoming just as critical as the technology itself, and businesses are unsure if they need entirely new executive roles to manage AI oversight.
As a CTO, I believe that my peers and I hold the ultimate responsibility for AI development and deployment. When it comes to any AI use case, the first question we need to answer is whether AI is the best, and responsible, solution. That said, the execution of AI deployment must be a collective effort, involving everyone from the C-suite to the day-to-day users.
We must collectively prioritize integration over isolation and expand our existing IT, security and legal frameworks to encompass generative AI solutions. This, in turn, will accelerate the development of innovative frameworks and prevent misalignment, ensuring AI usage remains a natural extension of our company standards. The key isn't just more AI rules and barriers—it’s education and upskilling.
When the entire organization understands AI and the ‘why’ behind it, guidelines become a facilitator of innovation rather than a bottleneck."
Have your own lesson to share? Email us at: a.mills@newsweek.com
Run Log
AI use case of the week
By Adam Mills
DXC Technology’s AI use case started before the product reached customers.
The IT services company was building DXC OASIS, a managed-services platform that uses AI agents to monitor enterprise IT operations, surface issues and coordinate responses. Its customers often run large technology environments built across years of vendors, tools and systems, which can make it hard for IT teams to see what is happening in real time.
Holly Grant, senior vice president of strategy and innovation at DXC and head of LabX, the company’s AI product incubator, said engineers began testing Anthropic’s Claude Code during development and quickly made it part of the build process.
According to Grant, Claude generated 95 percent of the DXC OASIS code, which then went through human review, helping accelerate software delivery nearly 10-fold. DXC launched OASIS in April 2026, and Grant said it is now deployed across more than 50 customers.
The product itself uses AI agents to read operational signals continuously, identify patterns and support human operators as they respond. Grant said DXC studied how operators and customers actually worked before making product decisions, including where friction showed up and where human judgment was most needed. Each agent is designed around a human operator role, with defined responsibilities and domain knowledge, so work can be routed across agents while people remain involved in higher-stakes decisions.
“They’re built the same way we train our best people,” Grant said.
The use case is unusual because AI’s fingerprints are on both sides of the work. Claude helped DXC build OASIS faster, while OASIS uses agents to help operators detect problems earlier, coordinate responses and keep human judgment attached to mission-critical IT work.
Have an AI use case to share with us? Email us at: a.mills@newsweek.com
Context Window
■ California launched a public dashboard to track AI-related job loss trends, with monthly early indicators for policymakers and initial data that shows no statewide rise in unemployment claims among AI-exposed occupations. [California Governor’s Office]
■ OpenAI is reportedly leaning toward delaying its IPO until next year after SpaceX’s post-debut stock slide and choppy tech markets raised doubts about investor appetite for AI companies at trillion-dollar valuations. [The New York Times]
■ Chinese AI startup Z.ai plans to use listing proceeds to fund its AGI ambitions after its open-source GLM-5.2 model narrowly trailed leading U.S. systems on public benchmarks, a fresh sign of China’s push to close the frontier-model gap. [Reuters]
■ JPMorgan says AI is now both a primary threat and defense mechanism in enterprise cybersecurity, with non-human identities, API vulnerabilities and automated remediation now key priorities for buyers and investors. [J.P. Morgan]
■ The Department of War launched Agent Network, an AI-enabled battle-management system designed to scan defense intelligence and operational systems, present commanders with faster options and keep humans responsible for targeting decisions. [Department of War]
■ BMW Group will deploy Figure 03 humanoid robots at its Spartanburg plant for logistics sequencing after an earlier Figure 02 pilot supported production of more than 30,000 BMW X3 vehicles, part of a broader push to use physical AI in manufacturing. [BMW Group]
Transfer Protocol
Tracking executive moves across the AI landscape
James Black, who spent 15 years at Google across customer service, cloud global offerings and trust and safety for Android, has joined TaskUs as vice president of AI services operations, where he will help build practices around AI safety, LLM model quality, robotics and other high-growth opportunities.
Christian Barnard, previously chief operating officer and chief growth officer at KUNGFU.AI, has been appointed chief executive officer at OneSix, where he will lead the data and AI services firm as it helps enterprises move AI from pilot to production.
Jason Gu, previously vice president and head of AI at Visa, has joined BRG as a managing director in its AI and Decision Intelligence practice, advising clients on AI strategy, implementation, delivery and governance.
Dimple Shah, previously head of strategy and client experience at Osaic Wealth, is moving to Humanity Labs as managing director of wealth, where she will introduce the company’s AI workforce tools to wealth management firms.
Vu Tran, a tech founder and engineer who most recently worked in Meta’s AI division, has been named Opendoor’s first chief AI officer, a role focused on bringing AI into how people buy and sell homes.
Usman Javaid, currently chief product and marketing officer at Orange Business, will become Orange’s group chief AI officer on Sept. 1, with responsibility for scaling AI and agentic AI across networks, customer relations, operations and business services.
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?
...
“The most fun shift for me recently is that AI has moved from answering questions to producing artifacts. A year ago, the interaction was ‘ask a question, get an answer.’ Today I ask a question and get back a Python file, a PowerPoint deck, and a working prototype. The output of the conversation is increasingly the deliverable itself, not a description of it.
As a PM, we've started taking business requirements documents and turning them directly into working code, with remarkable accuracy. But the moment that really stood out happened this week. We're building a network discovery tool that scans for SNMP-capable devices — not an area I'm personally an expert in. I asked Claude to explain how SNMP discovery actually works, and it didn't just explain the concepts. It then wrote me a Python script I could run against my own network to find devices, with vendor classification and a CSV export. I went from ‘curious about the problem’ to ‘have a working tool’ in the span of one conversation.
What stands out isn't the code itself — it's that the gap between understanding something and having a usable artifact for it has collapsed to almost nothing. That changes what it means to be a generalist leader. I can go deeper into technical areas I'd previously have had to delegate, which makes me a sharper partner to my engineering team.”
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