AI News Archive: June 25, 2026 — Part 6
Sourced from 500+ daily AI sources, scored by relevance.
- UK overtakes India in global unicorn rankings as AI-driven valuations surge: Hurun
India has dropped to fourth place globally in unicorn rankings, with the UK now holding third position. While the US and China lead, India's startup ecosystem saw slower growth in new unicorns, though several went public. Artificial intelligence is driving massive valuations across sectors, significantly boosting the AI and cybersecurity industries worldwide.
- How Notion used the Cursor SDK to embed coding agents
Notion integrated Cursor SDK to embed coding agents, enhancing productivity.
- Insurers turn to generative AI for catastrophe modeling, but hallucinations and sales logic could get in the way
Diffusion models generate tens of thousands of plausible weather events where historical data doesn't exist. Insurers are hoping for more precise risk assessments. Researchers warn about hallucinations. The article Insurers turn to generative AI for catastrophe modeling, but hallucinations and sales logic could get in the way appeared first on The Decoder .
- OpenAI says nearly all its employees have switched from chatbots to Codex agents, but every number comes from OpenAI itself
Nearly 98 percent of OpenAI’s employees now use Codex, the company’s AI coding agent, up from roughly 40 percent in August 2025, according to a paper the company published on Wednesday titled “The Shift to Agentic AI: Evidence from Codex.” The paper describes a fundamental change in how the company’s own workforce interacts with AI, […] This story continues at The Next Web
Score: 52🌐 MovesJun 25, 2026https://thenextweb.com/news/openai-codex-agents-shift-employees-non-developers - A Berkeley AI professor makes a provocative argument for decelerating AI research
Welcome to AI Decoded , Fast Company ’s weekly newsletter that breaks down the most important news in the world of AI . You can sign up to receive this newsletter every week via email here . Why Emma Pierson’s essay is causing such a stir Emma Pierson’s recent essay in The Atlantic , “I’d Rather Risk Cancer Than See AI Move This Fast,” has naturally stirred up strong opinions in the artificial intelligence community this week. The accelerationists on X, in particular, seemed triggered by it. Pierson is an AI researcher who teaches machine learning in the vaunted computer science program at the University of California, Berkeley. She also carries a gene mutation that can increase the risk of ovarian or breast cancer. In fact, she recently had her ovaries removed to reduce her chances of developing the disease. Her argument isn’t that AI will never help cure cancer. It’s that the huge generalist AI systems being developed at labs like Anthropic, OpenAI, and Google may one day help defeat disease, but they also carry more immediate societal risks, including mass unemployment, inequality, surveillance, and weapons development. “[I] will wait a little longer for a cure—even if it means losing my fertility and living under the shadow of risk—if it lets us approach this new world more carefully,” she writes. Pierson’s piece quickly caught the attention, and then the fury, of thousands of AI accelerationists on X. After all, she was challenging one of their most prized arguments, which is that anything less than driving hard toward powerful AI models is inhumane because it could deny millions of people suffering from cancer and other diseases a chance at a cure. In his 2023 manifesto , Marc Andreessen wrote: “We believe any deceleration of AI will cost lives. Deaths that were preventable by the AI that was prevented from existing is a form of murder.” Andreessen had this to say about Pierson’s article on X: “Did cancer write this?” The post got 15,000 likes and almost a thousand retweets. AI company leaders often say they are most excited about AI’s potential to accelerate scientific research, including the search for cures. But Pierson argues that the huge, generalist models aren’t specifically designed to cure disease, and still seem far from delivering major improvements in patient outcomes. She explains why. AI labs are showing impressive progress in areas like coding and math, where training data is abundant and “ground truth” answers are available. Cancer is different. “Cancer data are finite and come from biological experiments and clinical trials that cannot run at silicon speeds,” Pierson writes. “And cancer data only imperfectly illuminate the complex processes by which our own cells betray us. There are, in short, many barriers to curing cancer beyond a lack of intelligence.” Beyond the practical question of whether generalist AI models can cure disease, Pierson is making a broader argument. Humans have been the most intelligent beings on the planet for roughly 300,000 years. Now they are building synthetic intelligence that could become many times smarter than they are. If that happens, there is a real danger that humans will outsource much of their mental labor to machines and, in doing so, lose touch with the very human pursuit of knowledge. Pierson seems to suggest that such outsourcing carries a cost, even when human lives are at stake. “For my own part, I would neither spend months struggling with a research problem I knew AI could solve instantly nor find as much pleasure in the answers it provided,” she writes. “I do not want to be merely a spectator to the universe, whatever wonders AI may reveal.” “Incredible reason to not want to save millions of lives,” scoffed Justine Moore, a partner at the Andreessen Horowitz venture capital firm, on X. But I think Pierson was making a broader point. She doesn’t doubt that AI could play a major role in defeating cancer. She is arguing that if we are on the cusp of artificial superintelligence, now is the moment to make thoughtful decisions about which tasks should remain human, rather than delegating every damn thing to machines because it’s convenient for users and highly profitable for model makers and their investors. Alex Bores, who AI industry forces both opposed and backed, loses bid for Congress Alex Bores, whose primary campaign became a flashpoint in the fight over AI regulation, lost his bid Tuesday night to succeed Rep. Jerry Nadler, Democrat from New York. Bores became a prominent figure in that debate after a political action committee (PAC) backed by AI leaders spent millions to defeat him. As a New York State Assembly member, he spearheaded an influential AI safety bill. Before entering office in 2022, he worked as a data scientist at the shadowy AI firm Palantir. “Though we’ve come up short tonight, the example set here tonight was not the one the AI oligarchs intended,” Bores said in a concession statement Tuesday night. “They set out to make people afraid to stand up to them. Instead they learned just how ready people are to push back.” By Wednesday, nearly every side of the AI debate was claiming some version of victory. Think Big, a group affiliated with Leading the Future, a PAC backed by leaders at OpenAI and Andreessen Horowitz, spent $8 million on ads and direct mail attacking Bores. Leading the Future says it wants a uniform federal framework for regulating AI models. Another group, Jobs and Democracy PAC, which is tied to Public First Action, a super PAC network linked to Anthropic, spent more than $11 million supporting Bores. Jobs and Democracy PAC says it aims to elect Democrats who will “stand up to Big Tech companies trying to buy their way out of sensible AI regulation.” Bores had hoped to turn that attention, and growing public concern about AI, into a win in New York’s Twelfth Congressional District, which covers parts of Manhattan. But AI wasn’t the only force shaping the race. Michael Bloomberg, the former New York City mayor, gave $10 million to a super PAC supporting Micah Lasher, his former aide, who won the primary. It’s hard to say whether the race will have a lasting effect on national AI policy. Public opinion on AI is already fairly entrenched , and millions of dollars in attack ads can only do so much to move it. In one sense, Leading the Future with its opposition to Bores may have gotten a worse result than a Bores victory. Dean Ball, an AI policy analyst, noted that Bores was relatively centrist on AI, while Lasher supports a moratorium on AI data centers, the crucial infrastructure behind the spread of AI. “LTF’s ‘victory’ here is that a guy who supports a data-center moratorium won, and the guy who supports frontier transparency and auditing (which LTF now supports!) lost,” Ball tweeted Wednesday. “Is that really a win?” Still, the race offered an early look at how AI money may shape congressional contests around the country. In the U.S. at least, ChatGPT is maintaining its lead among chatbots New Pew Research survey data shows that Americans are far more aware of, and more handy with, AI chatbots—especially ChatGPT—than they were just a couple of years ago. Here are some selected nuggets: Usage: 36% of Americans say they interact with AI at least several times a day. Chatbot share: Americans (44%) report using ChatGPT far more than other chatbots, followed by Google’s Gemini (24%), Microsoft’s Copilot (17%), Meta AI (14%), xAI’s Grok (8%), and Anthropic’s Claude (6%). More good news for OpenAI: The percentage of Americans who say they use ChatGPT more than doubled from 2023 (18%) to 2026 (44%). Chatbot uses: The top three uses for chatbots in 2026 are search and information retrieval (42%), tasks at work (38%), and fun and entertainment (25%). One in 10 people say they use chatbots for emotional support. AI in search: A majority of Americans (60%) say they read AI summaries at the top of search results. AI advancement: Roughly two-thirds of Americans (63%) say AI is advancing too quickly, while only 2% say it’s advancing too slowly. Regulation and governance : Two-thirds of Americans (67%) have “little to no confidence” in the government to regulate AI effectively. About six in 10 adults are “not confident” in U.S. companies to develop and use these tools responsibly. More AI coverage from Fast Company: One fake web page can be enough to trick AI shopping recommendations Daters say AI dependence gives them the ick Can we trust scientific images in the era of AI? Satya Nadella is asking the right AI question Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.
- Is an AI Price War Brewing?
Is an AI Price War Brewing? The Information
- CIOs rethink the balance between AI oversight and innovation
The new CIO mandate is clear: facilitate AI adoption across the enterprise at speed. According to CIO.com’s State of the CIO survey, CEOs’ to p priority for their IT executives is to capitalize on AI . From researching to evaluating AI products, CIOs are now the central figures in their organizations’ AI strategies. And company leaders are looking for real outcomes. Almost two-thirds of senior leaders report there is more pressure to prove ROI on their AI investments than a year ago, according to Kyndryl’s 2025 Readiness Report . Numerous sources — from the board, to the CEO, to business units and competitors — are behind this pressure, says Jonathan Tushman , chief AI officer and CTO at Hi Marley, a customer conversational platform for the property and casualty insurance industry. Succeeding in the task ahead of them requires complex conversations, and getting through legal, compliance, and other checks “at a reasonable clip,” adds Tushman, who added CAIO to his remit more than 18 months ago but has felt added urgency in the past six months. In professional gatherings, board conversations, and almost everywhere across the business world, the conversation turns to AI — and then quickly the fear of failing behind. That includes employees as well. “It’s the engineering team and there’s everybody else — marketing, sales, finance. It’s people who are not AI-native, but they’re very eager to use these tools at an early level,” he says. As CIOs find themselves facing pressure to scale and demonstrate real value, the challenge is keeping up with risk considerations — without creating unnecessary friction. “CIOs cannot be risk averse on this,” says Karthik Chakkarapani , SVP, CIO, and head of enterprise AI at Zuora. “We need to do security and governance, but we don’t want to be seen as slowing down the process. You have to build the highway with enough guardrails and fewer speed breakers.” Moreover, he adds, “this is not about automating existing work. This is reimagining how work gets done.” AI is a step-change in risk management Most IT leaders are a long way from feeling comfortable with the new AI risk management balancing act. Just 31% of respondents feel completely ready across external business risks, Kyndryl’s survey reports. Tushman believes two things are genuinely different about the risks AI introduces. The first is that AI is indeterminate, whereas most technology is deterministic. “You can’t prove an AI system will or won’t do X, so the traditional ‘put controls around it and verify’ model breaks down,” he says. “We need a different way to govern something whose behavior you fundamentally can’t pin down.” The second is the gravitational pull on end-users. “With most tech, IT could take its time evaluating before rollout,” he says. “With AI, if you don’t put powerful tools in front of people fast, they’ll route around you — and shadow use creates more risk than controlled access ever would. The timeline compresses at the same time the control model gets harder.” Tony Vizza, founder and managing partner of Novera, agrees that the instinct to move fast can lead to the exact failures everyone fears. “This might be staff putting sensitive information into public tools without a proper governance structure, or people copying and pasting straight out of AI and sending incorrect deliverables to customers,” says Vizza. Organizations should avoid jumping into AI because of the fear of missing out without first clarifying where and how it will be used. All risk decisions should flow from these questions, he says. “What problems are you trying to solve — is it better customer service or deeper insight into your data? What are you actually trying to do?” Vizza recommends guiding AI decisions with a risk assessment that considers expected outcomes, size of investment, and its importance to the organization’s objectives. “You define your risk appetite, build a risk register, and define what risk treatment should be for each risk,” he says. “For example, if you’re going to use a public AI model, you might treat that risk by not putting sensitive data in or buying the right license so that if you do, you’re covered, or getting guidance from the regulator before you proceed.” Organizations must also consider AI services as a third-party risk, and not leave all accountability with AI providers, Vizza says. “You can’t outsource the responsibility,” he adds. Due diligence is required to understand what is in the AI provider’s contract, who is responsible if they have a data breach, and how your organization can pursue them if something goes wrong. “Some organizations build that into their risk management process. Others are quite flippant or don’t even know they should be asking those questions — and that’s what gets them stuck down the track,” he says. The importance of organizational design At Hi Marley, Tushman and team have made structural decisions to foster “healthy internal tensions” that are intended to surface and address AI risk considerations. This includes separation between the “AI adopters” in the product and technical teams and the “AI oversight” teams in compliance and legal. Compliance owns the audits, security concerns, and ongoing oversight, while legal owns the documentation that describes the boundaries. “The key is that it’s independent from the teams pushing AI forward,” he says. “Companies need to invest seriously in these compliance functions. Hire smart, nuanced people. These roles can’t just be ‘no’ machines, but they can’t rubber-stamp everything either. The value is in the judgment,” he says. Tushman’s role is the AI innovation steward, spearheading AI adoption that includes being challenged on risk, compliance, and legal considerations. “We have a senior leadership team and we have ‘conflict by design’ within that group,” he says. “I play the CAIO role and next to me, I have our head of legal and our head of compliance. So in that leadership team, if we have ‘conflict,’ we’re able to understand the trade-offs and make a decision as a group.” Tushman believes this creates healthy tension: Innovation-minded leaders push boundaries while compliance and risk leaders counterbalance them. But if a decision can’t be reached, it goes to the CEO. “I do recommend a [split decision] goes to another officer in the organization,” he says. Decisions about organizational structure could prove to be as consequential as the AI adoption decisions themselves, Tushman says. “The companies that get the organizational design right early will have a real advantage,” he explains. Desire for AI advances the risk equation One of the features of the AI wave is the thirst for access — from the board to employees — to use the tools, build applications, and start putting them to work. “Right now, everyone’s dying to try it,” says Tushman. Hi Marley is in the “activation” phase — meeting the appetite for the tools with safety wrappers. “My main goal here is to have people learn the tools, start using them, and gain some competency with them,” he says. “We will get to the measurement phase, but I think spending too much time on measuring right now is not worth the effort.” Tushman, like many, is watching how quickly models improve. “AI has huge implications for how you organize, how you hire, and what buy‑versus‑build decisions you make,” he says. Zuora, which specializes in software for subscription and recurring revenue businesses, is three years into its AI journey. Chakkarapani is adamant that speed for speed’s sake is not the goal. “We don’t want to take an existing process and just make it faster. You’re just making a process more chaotic. Can we make it fast, smarter, and reorganize it?” Vizza believes a good percentage of CIOs will need external help to navigate the push for rapid AI adoption. “Or they’ll need to upskill themselves, because AI operates very differently to traditional IT,” he says. His advice is threefold. First, “make your decisions on the right basis — either learn how AI really works or bring in someone who can advise you properly,” he says. Second, bring it back to the business purpose. “There are opportunities with AI, but the core question is, ‘What are we trying to achieve by bringing this in?’” And third, work out how you’re going to manage the risk. “Risk isn’t necessarily a bad thing — Formula 1 cars are risky, but they have very good braking systems so they can go faster,” he says. “It’s the same with AI: You put the right risk management in place so the business can move quickly without suffering adverse consequences.” In its almost three-year AI journey, Zuora started with experimentation before moving 12 enterprise-wide pilots into production, Chakkarapani says, adding that there are three pillars to assess potential AI projects against: effort, value, and confidence. “Effort includes the security risk,” he says. “Is it low, medium, or high?” Chakkarapani’s team started with simple executions, although the first experiments didn’t go as hoped — providing valuable lessons for the following ones. “We learned AI is only good when you have the right data — the right content, context, and governance,” he says. They moved on to IT service management and that’s when the practical learnings really started, gaining feedback from internal teams and users, answering the security and governance questions, and iterating as they went. Early applications include marketing, sales, product, and technology, achieving 10x to 25x throughput improvements. Success is measured in business outcomes such as growth, cost saving, customer engagement. Through this process, the team has been doing the “behind the scenes” work to speed AI adoption across the company. “We realized that to go at speed and scale, we need to have the right trust, security, and governance underlying it,” he says. An enterprise-wide platform connects Zuora’s approved AI services, including ChatGPT and domain-specific tools, to its structured and unstructured data. On top of this is the context layer and services so that people can build their own applications. It uses each employee’s existing login and organizational profile, and it respects the same role-based security. “We slowly developed the framework that became our blueprint with the 10 to 12 things that need to be considered when creating an AI-driven application. When someone is interested, they’re taken to the self-directed process with these do’s and don’ts that is automatically downloaded as a markdown file to that person’s computer,” he says. The ultimate aim is delivering up to 100x business value through an enterprise-wide governed platform — covering IT, HR, finance, legal, procurement, sales, and product. IT plays the role of orchestrator, providing the platform to access the tools and agents and collaborating with the business team to reorganize that workflow. The AI maturity model Chakkarapani believes the more secure the environment, the more it paves the way for experimentation, adoption, and, in time, business results. At Zuora, Chakkarapani has evolved this process through three levels of organizational AI maturity to date: Level 1: IT provides a platform and services. Employees have controlled access to data based on their role and security privileges. They can create their own agent for themselves. If something doesn’t pass the minimal security and compliance and requirements, it cannot move ahead. Level 2: An employee-built agent goes through an IT governance check for duplication or overlap, model improvements, security scans, and manual reviews. If approved, it’s shared with the wider enterprise. “We’re doing well on that, but it’s still a lot of manual work because there are no tools in the market that can automate this,” he says. Level 3: At this stage of maturity, an organization has established a secure foundation across its applications so AI can scale safely. At Zuora, over six to eight months the team tightened endpoint and application security, enforced mobile device management, introduced AI usage monitoring (including what staff upload into prompts), and disabled Google authentication to block personal or bulk email accounts from accessing unapproved apps. Earlier this year, the team embarked on working toward Level 4 maturity, where anyone can create a functioning application with minimal human involvement. Realistically, they expect to be 80% to 85% zero-touch because the final mile will still require human involvement. “My goal is to provide a zero-touch service for anybody in the organization to create applications. If we do, they can go from a concept to an idea, prototype, design, and production — and they do it in less than two weeks,” he says.
Score: 51🌐 MovesJun 25, 2026https://www.cio.com/article/4188566/cios-rethink-the-balance-between-ai-oversight-and-innovation.html - Read our white paper on a pragmatic approach to AI governance in America.
The debate over AI governance is stuck in a false choice between over-regulation and no regulation. There is a middle way: A pragmatic, evidence-based approach that reco…
Score: 51🌐 MovesJun 25, 2026https://blog.google/company-news/outreach-and-initiatives/public-policy/white-paper-ai-regulation/ - Why AI agents will reshape customer journeys in Southeast Asia
Southeast Asia has never followed a single digital playbook. A customer in Thailand may expect to interact with a brand through LINE. A shopper in Indonesia or Malaysia may prefer WhatsApp. In Vietnam, Zalo remains deeply embedded in daily communication. In the Philippines, Messenger continues to shape how people connect, discover, and transact. This makes […] The post Why AI agents will reshape customer journeys in Southeast Asia appeared first on e27 .
Score: 51🌐 MovesJun 25, 2026https://e27.co/why-ai-agents-will-reshape-customer-journeys-in-southeast-asia-20260619/ - AI efficiency beyond the model: Rethinking code, hardware and cloud
As AI adoption grows, I see fellow enterprise leaders realizing that just implementing AI is not enough. We need to develop and adopt the best, fastest and most efficient AI models. It’s not just a matter of pride about who has the shiniest toy; optimizing models for efficiency can be the difference between a failed pilot and an effective business strategy. At the most extreme end of the spectrum, inefficient use of AI can cost billions of dollars. Sam Altman, CEO of OpenAI, made headlines when he admitted on X that his company loses tens of millions of dollars every time people say “please” and “thank you” to his AI models, even though he added that he feels it’s money well spent. Model efficiency also matters for those of us not operating at OpenAI’s scale. A more efficient model helps reduce overall costs because it doesn’t require as powerful or expensive hardware, uses less electricity, delivers output faster and can operate with a smaller cloud footprint. Models that are optimized for efficiency deliver lower latency, improved scalability, increased flexibility and are less likely to drift. In my experience, all of this adds up to higher profit margins, a sharper competitive edge and a faster time to market, which are crucial whether you’re planning to use your model internally or sell it to others. The new CIO investment dilemma For a long time, it was believed that hardware must continually increase in power to enable models to grow in size. Then DeepSeek v2 came along and demolished all those theories. It showed that more efficient hardware can deliver equivalent results with less compute power by running smaller, smarter models. Now, those of us in the CIO seat face a new dilemma: should we increase investment in computing power, focus on hardware or concentrate on software? In my view, the correct answer is: all the above. AI efficiency is a full-stack problem. Hardware, compilers, runtime and model architecture must be co-designed to work in harmony; otherwise, we’re wasting money and failing to achieve the results we need. Today, choosing GPUs vs. custom accelerators vs. CPUs affects which model optimizations are viable. Hardware power constraints model capabilities It remains true that even the most powerful model in the world can’t function without access to the necessary hardware. Hardware performance is ultimately bounded by memory bandwidth, interconnect speed and compute units, no matter how optimized our models are. This means that scalability depends on interconnects. Multi-node training and large inference clusters hinge on the performance of NVLink, InfiniBand or Ethernet fabric, not just model quality, so decisions about hardware investments or cloud providers can be critical to overall functionality. “The pace of innovation is directly tied to advances in GPUs, tensor processing units (TPUs) and custom accelerators. The real question isn’t just what models we can build, but whether we have the compute infrastructure to support them,” says Gaurav Dewan, a research director at Avasant. “Models can only grow as powerful as the chips, memory systems and data center networks sustaining them.” Compute power isn’t everything That said, in my experience, you can’t just throw computing power at every problem. Choices about hardware and cloud architecture determine how effectively users can tap into the potential of compute resources. Modern AI workloads are often memory-bound rather than compute-bound, so faster HBM, cache hierarchies and interconnects directly lower latency. What’s more, the energy for computing power is limited. Companies can’t always afford the compute power they want, with 58% saying their AI cloud costs are too high. Cost per inference is hardware-driven and compute is usually the biggest line item in AI TCO. It’s not even easy to find space for enough GPUs, creating board-level power and cooling constraints in enterprise AI. More efficient silicon reduces data center strain, sustainability risk and cost per token/inference. Additionally, reliability and utilization affect ROI. Features like MIG partitioning, hardware scheduling and fault tolerance determine how fully we can monetize expensive accelerators. Performance per watt is now the bottom line, with CIOs like me striving to get more out of every existing GPU per watt, dollar and square meter. We need to make our hardware more efficient by fine-tuning models and software to maximize capability. “DeepSeek’s breakthrough suggests that AI models no longer need to scale indefinitely in size and complexity to achieve superior performance. Instead, they can be algorithmically optimized to deliver the same, if not better, results while consuming significantly fewer resources,” explains Matthew Taylor in his post on LinkedIn. Rethinking cloud strategy in the age of AI That cost pressure has forced many of us to revisit assumptions we held for the better part of a decade. Cloud computing has reached an uncertain crossroads. The hyperscaler-by-default posture that defined the last era of enterprise IT no longer survives a serious look at AI economics. When inference costs scale linearly with usage and training runs can consume an annual infrastructure budget in weeks, the question I hear in every CIO conversation is the same: does our cloud strategy still match the workload we are actually running? In my experience, the answer is increasingly no, at least not without significant rebalancing. Private clouds, written off as legacy not long ago, are quietly making a comeback. The combination of predictable cost structures, tighter control over data residency and the sensitivity of the proprietary data feeding our AI systems is making on-premise and colocation options compelling again, particularly for regulated industries. At the same time, purpose-built neoclouds for GPU workloads, along with sovereign clouds responding to jurisdictional and data-protection mandates, are steadily chipping away at the dominance of AWS, Azure and GCP. None of these alternatives replace the hyperscalers outright, but they are forcing every CIO I know to think about cloud as a portfolio rather than a single vendor relationship. What I have found is that navigating this shift takes more than a procurement decision. It takes a clear-eyed view of where each workload genuinely belongs. Training, inference, retrieval, fine-tuning and experimentation each carry different cost curves, latency profiles and data-gravity considerations. As organizations move towards the agentic AI era, the underlying data platform becomes equally important, requiring architectures that can support multimodal data, real-time processing and governance at scale. The enterprises I have seen handle this best treat cloud strategy as an ongoing exercise in workload placement, not a one-time platform commitment. That is also where the conversation tends to outgrow internal teams. As AI moves from pilots to production, the questions get harder: how to architect data foundations that survive model churn, how to govern AI without strangling it, how to translate technical efficiency into measurable business value. I have seen organizations lean on specialist partners to think through these problems alongside them. Among the consultancies working at this intersection is Artefact, founded in Paris and operating across data strategy, AI engineering and enterprise transformation. Its work includes governance, platform development, operating models and workforce enablement—areas that have become increasingly important as organizations move from AI pilots to large-scale deployment. What I find useful about these consultancies is not the technology recommendations themselves; it is the pattern recognition they bring from seeing similar cloud and AI transitions play out across geographies and sectors. In a moment when every CIO is rewriting the playbook simultaneously, that outside vantage point matters more than it used to. Hardware is often underused and misused A lot of hardware goes unused or underutilized. Often, GPUs sit idle due to deployment complexity and data infrastructure bottlenecks, so enterprises don’t see the value of the compute power they’re paying for. When data and computing are on two separate chips, compute is wasted moving data between the two locations. Likewise, models that exceed accelerator memory or require excessive HBM traffic suffer steep latency and cost penalties. Optimizing models to align with hardware means that all the compute power is being put to good use. Techniques like operator fusion, activation management, fine-tuning smaller models, pruning unnecessary parameters and memory-aware architectures keep more of the model resident on the accelerator, reduce unnecessary read/write cycles and combine steps so data is touched fewer times. Kfir Aberman, founding member at Decart AI, explains this approach . “Our solution to this was to optimize our kernels for how [Nvidia GPU] Hopper works. Essentially, we created a single ‘mega kernel’ that enables the chip to process all of a model’s computations in a single, continuous pass. By doing this, we eliminate all of the stopping, starting and data movement, allowing more of the GPU to be utilized more of the time, speeding up processing by an order of magnitude.” When models match accelerator characteristics such as tensor core shapes, SIMD widths and kernel libraries, this keeps expensive silicon working effectively and translates theoretical FLOPs into real throughput. More hardware can’t overcome model mismatch Another way that organizations undermine ROI on their own AI investments is by ignoring coordination efficiency. They’ll buy large GPU clusters but pay little attention to what seem like minor issues with batching and alignment. Unfortunately, when batch sizes are wrong, work is split inefficiently and network links become bottlenecks, you see expensive but underutilized clusters. Ultimately, more GPUs don’t guarantee more performance. Parallelism and batching must match the system topology. Effective scaling depends on aligning data, tensor and pipeline parallelism and batch sizing with the actual interconnect bandwidth and node configuration. The magic happens when model and hardware come together The lesson that those of us in CIO roles are learning is that symbiosis between model and hardware is critical. Code determines what our AI can do, hardware determines how efficiently we can afford to do it and co-design determines whether our AI program scales economically and successfully. This article is published as part of the Foundry Expert Contributor Network. Want to join?
Score: 50🌐 MovesJun 25, 2026https://www.cio.com/article/4189043/ai-efficiency-beyond-the-model-rethinking-code-hardware-and-cloud.html - Authors Guild test finds some AI detectors perfectly identify human writing while others fail on every single text
The Authors Guild tested five AI detectors on human-written texts. Pangram and Grammarly correctly identified all of them, while Sidekicker and ZeroGPT flagged human-written articles as AI-generated. But the Guild also warns of a paradox: professionally written texts look statistically similar to AI output because language models were trained on exactly that kind of writing. The article Authors Guild test finds some AI detectors perfectly identify human writing while others fail on every single text appeared first on The Decoder .
- Chinese physical AI start-up proposes new paradigm that bypasses OpenAI, Meta road maps
A Chinese physical AI start-up has launched a new world model designed to simulate reality by embedding the laws of physics directly into its code – a departure from the data-driven approaches favoured by American tech giants like OpenAI and Meta Platforms. Shanghai-based Fysics AI announced the launch of the Fysiverse, which it described as a “new-generation physics-based world model that adheres to real-world physical laws”, in a post on its WeChat social media account on Wednesday. The...
- Agentic-Native Platforms Are Creating A New Technology Business Model
For decades, the enterprise technology industry operated on a simple principle: software companies built products, and services firms helped enterprises.
- California lawmakers want to ensure Cal State programs are taught by humans, not AI
California lawmakers want to ensure Cal State programs are taught by humans, not AI The Mercury News
- Autonomous vehicles could potentially cut Dallas-Fort Worth congestion
Driverless cars could ease commutes in Dallas-Fort Worth, a new study led by SMU suggests.
Score: 49🌐 MovesJun 25, 2026https://techxplore.com/news/2026-06-autonomous-vehicles-potentially-dallas-fort.html - AI coding token costs are on track to rival human payroll
AI coding token costs are on track to rival human payroll InfoWorld
Score: 49🌐 MovesJun 25, 2026https://www.infoworld.com/article/4189176/ai-coding-token-costs-are-on-track-to-rival-human-payroll-2.html - Why AI governance matters at scale
Your next AI agent may need to be treated like a new employee.
- Surge Capacity: How Anduril Delivered a Month of Production in a Week
Surge Capacity: How Anduril Delivered a Month of Production in a Week
Score: 49🌐 MovesJun 25, 2026https://www.anduril.com/news/surge-capacity-how-anduril-delivered-a-month-of-production-in-a-week - China’s cybersecurity industry needs its own Mythos model, 360 founder warns
China must develop its own equivalent to Anthropic’s Mythos model to counter the cybersecurity risks posed by the artificial intelligence era, according to 360 Security Technology founder Zhou Hongyi, who likened the powerful US technology to a “cyber nuclear weapon”. Released in April with the ability to autonomously identify software vulnerabilities, Mythos has accelerated vulnerability discovery a hundredfold while slashing costs – a “terrifying change” that had effectively “democratised”...
- The AI and Data Center Image Problem
The AI and Data Center Image Problem The Information
- Big Tech is all in on AI. Now all they need is customers.
Technology companies are betting trillions of dollars that consumers will open their wallets for AI services. But what if Big Tech is wrong?
Score: 49🌐 MovesJun 25, 2026https://www.cbsnews.com/news/ai-bubble-tech-selloff-investment-consumer-business-demand/ - Utah Voters Give Senate President Who Supported Giant Data Center The Heave Ho
Kevin O’Leary is a media star who is known as Mr. Wonderful on the reality TV show Shark Tank. Recently, he placed his boldest bet yet, on Stratos, a proposed data center in Utah that he said would create thousands of jobs, add millions of dollars in revenue for local ... [continued] The post Utah Voters Give Senate President Who Supported Giant Data Center The Heave Ho appeared first on CleanTechnica .
- Amazon expands quick service in India; MoEngage acquires AI infra startup Aampe
Amazon expands quick service in India; MoEngage acquires AI infra startup Aampe YourStory.com
- Why Most Healthcare AI Fails After the Pilot Phase
Why Most Healthcare AI Fails After the Pilot Phase MedCity News
Score: 48🌐 MovesJun 25, 2026https://medcitynews.com/2026/06/why-most-healthcare-ai-fails-after-the-pilot-phase/ - DuckDuckGo’s AI Feature Is Telling Users That Trump Died of Rabies Earlier This Month
"The infection was linked to a sequence of events involving Vice President JD Vance, who passed away shortly before Trump." The post DuckDuckGo’s AI Feature Is Telling Users That Trump Died of Rabies Earlier This Month appeared first on Futurism .
- How we used DSPy to turn AI evaluations into better responses in Dash chat
We used DSPy to improve LLM judges and optimize our chat experience, creating an evaluation-driven feedback loop that produced better outputs.
Score: 48🌐 MovesJun 25, 2026https://dropbox.tech/machine-learning/how-we-turned-ai-evaluations-into-better-responses-in-dash-chat - Krafton CEO discusses AI cooperation with AWS chief
Krafton CEO discusses AI cooperation with AWS chief 매일경제
- Boomi CEO shares vision of AI cost management
Boomi is hoping it can provide IT leaders with greater visibility on their token spend, something that is lacking across the industry. The company is developing a tool called Boomi Prompt, which acts as middleware between enterprise applications and large language models (LLMs), and the artificial intelligence (AI) agents that need to access these systems to perform a task on behalf of a human user. As the use of artificial intelligence and AI agents starts ramping up, providers of LLMs and AI tools are moving from subscription or software as a service (SaaS)-style software licensing to pricing based on the costs associated with AI inference, measured in tokens. A token is the smallest piece of information an AI engine or LLM takes as input, such as a word in a sentence. The larger the volume of tokens submitted to the LLM, the larger the token usage, and this equates to more computational resources needed by the provider. That cost is the token cost the organisation pays to submit the query to the AI tool. If a query is continuously being submitted, the token cost is paid again and again, even if the organisation already has the answer. Boomi aims to cache such repeated responses, to avoid organisations spending unnecessarily on tokens when they already have the answer. According to Boomi, the Prompt tool is also able to figure out what LLM is least expensive to answer a user or AI agent’s question. Speaking at the Boomi World Tour in London, the company’s CEO, Steve Lucas , said the company will release a tool called Prompt later this year that “provides a layer” between an AI engine and any backend system. Read more stories about AI costs AI’s effect on UC pricing models : AI costs are threatening traditional SaaS budgets. UC and CX leaders face new pricing models, vendor lock-in risks and the challenge of defining AI ROI. IT’s budgetary nightmare : Tech buyers face AI pricing variance: Did HubSpot just fix AI pricing? The agent may seek to find information held in an SAP or Oracle system, using an application programming interface (API), or it could call an LLM. When the agent is asked to do a task, he said: “If it seeks data from an SAP system and Oracle system, and the answer to the prompt is cached in our prompt layer, we will provide that cached response.” This saves costs associated with continuously using APIs to access commercial off-the-shelf enterprise software, where there may be an indirect access cost associated with that API. Lucas said the new Boomi tool is also able to understand when a prompt submitted by a user or an agent can be routed to a standard SQL-based query such as a Google search, rather than “burning tokens”. However, he said: “If that prompt is of value, we will route it to an AI model, and the model we select will depend on the rated complexity of that response.” One example of the prompt is a forecasting question such as expenses across two systems, he said. “We have Nemotron from Nvidia, which, in this hypothetical scenario, is effectively free for my business to run,” said Lucas. “We will route the prompt there.” According to Lucas, prompt routing is a complex but highly necessary capability for the enterprise, which he said is completely unserved today. “There is no sophisticated prompt routing standard for the enterprise,” said Lucas. Although Perplexity does offer prompt routing, according to the Boomi CEO, it is not enterprise-oriented. He said Boomi’s approach aims to go further. “The work that we’re doing has many layers and token reduction, and optimisation is one of those layers,” said Lucas. “Prompt routing will allow companies to reduce their token spend massively. Our design objective is to achieve greater than 50% reduction in token spend in the enterprise.”
Score: 48🌐 MovesJun 25, 2026https://www.computerweekly.com/news/366645085/Boomi-CEO-shares-vision-of-AI-cost-management - Palo Alto Networks CEO: We're in 'a Darwinian moment' where employees have to prove their AI skills
Palo Alto Networks CEO: We're in 'a Darwinian moment' where employees have to prove their AI skills Business Insider
Score: 48🌐 MovesJun 25, 2026https://www.businessinsider.com/palo-alto-networks-ceo-ai-training-skills-jobs-2026-6 - Short-form science: University of Washington researchers launch PaperTok to combat AI slop
A University of Washington team is helping scientists tell their own stories with a free tool that converts dense, jargon-heavy publications into short, accessible videos. Read More
- Robots are coming to the oil patch
Also in today’s newsletter, Russia receives an oil windfall amid Iran war
- ZTE showcases full-stack AI capabilities at MWC Shanghai 2026, empowering new era of token operations
PARTNER CONTENT: Driving end-to-end synergy across AI factories, next-gen AIOS, and 6G-ready networks to maximize token efficiency and lower operational costs
- More Massive Still: Why AI Infrastructure Demands A Unified Design Approach
Tokens-per-watt is now the primary metric driving AI data center optimization. The post More Massive Still: Why AI Infrastructure Demands A Unified Design Approach appeared first on Semiconductor Engineering .
Score: 47🌐 MovesJun 25, 2026https://semiengineering.com/more-massive-still-why-ai-infrastructure-demands-a-unified-design-approach/ - Durham drag artist and nightclub sue Meta after automated system deletes accounts
Durham drag artist and nightclub sue Meta after automated system deletes accounts Raleigh News & Observer
- GOP Congress hopefuls spar on AI as data centers become divisive topic
GOP Congress hopefuls spar on AI as data centers become divisive topic azcentral.com and The Arizona Republic
- AI is changing how South Africans find local services, but human trust still matters
One of the advantages of AI within an online marketplace is its ability to improve matching between supply and demand, says Snupit.
- Building AI tailored for education, with educators in the lead
A graphic featuring the Gemini logo surrounded by icons representing educational tools, including Guided Learning, Study Notebooks, and NotebookLM, set against a background featuring a security shield and classroom imagery.
Score: 47🌐 MovesJun 25, 2026https://blog.google/products-and-platforms/products/education/iste-2026-educator-updates/ - The real reason people hate AI data centers so much
When I shared my predictions for AI in 2026 earlier this year , I snuck in a one-sentence nugget that turned out to be surprisingly prescient: “In 2026, I expect a populist backlash against the fact that data centers’ voracious energy demands are raising electricity rates for everyday people.” I was right to flag the problem. But even as an AI expert, I failed to predict its scope. People hate AI data centers . They’ve been blamed for high electric bills , but also air pollution , odd humming noises , water scarcity, fiscal decline , and much else. To be sure, plunking a 300,000-square-foot building filled with power-hungry servers in the middle of a community comes with costs and challenges. But the national uproar over data centers reflects a much deeper anger. AI companies ignore it at their peril. In my backyard As a journalist, I only fully understood the popular outrage around AI centers when I wrote a story about one coming to my own backyard in the San Francisco Bay Area . That data center—which has already been approved by the local municipality—will take over a former golf course. At around 347,000 square feet, it’s big, but nowhere near the massive scale of facilities in the Midwest, which can top 1 million square feet . Shovels haven’t even hit the earth on the project, but locals are already up in arms. They’ve flooded local city council meetings with protestors and gathered more than 18,000 signatures opposing the new building . A social media post I wrote about my story has received 44,000 views and 100-plus comments. Most of the concerns are familiar ones, echoing criticisms occurring all around the country. In a time when energy is already blindingly expensive, many Americans worry that data centers will raise their utility bills . Many data centers have massive diesel generators, intended to keep the servers humming if the local power grid goes down. Lots of people worry that those generators will belch smog and cause health issues . The more environmentally minded often point to data centers’ alleged massive water usage , often quoted in millions of gallons per day. And some people just feel that they’re noisy and ugly . Many different people, in other words, have many different reasons for hating AI data centers. But as I’ve seen firsthand, that hatred is deep and abiding. Why me? At first glance, this level of popular ire makes little sense. When you actually examine the data about data centers, many claims about their abject awfulness fail to hold up. As The Atlantic recently reported, fears about AI data centers driving up electric prices are often oversold . A comprehensive study of the economics of data centers recently found that they actually reduce electricity prices slightly . Texas is a perfect example of that fact. The state is leading America’s data center boom, yet its electricity prices are among the lowest in the country . Prices, the data shows, are far more closely tied to grid investment and factors like wildfire risk than the presence of data centers. Likewise, researchers say that the stats about data center water usage are often taken out of context. Scary figures often count inane things, like the evaporation of water from reservoirs upstream of the data center itself. Indeed, my own local data center claims it will use less water than the golf course it’s replacing. Data centers’ ugliness is subjective, of course. But the American heartland has plenty of ugly logistics centers and shipping warehouses that don’t prompt tens of thousands of people to sign petitions. And data center boosters can reasonably point to claims about the centers’ positive impacts. As The Atlantic shares, tax revenue from data centers can have massive benefits for small towns, and unlike older data centers, the AI ones often create high-paying local jobs by attracting AI firms. Given this encouraging data, why have AI data centers become such a hated piece of the American economy and landscape? A deeper issue Based on my own experience as a tech journalist and photographer, I believe the public anger about data centers actually points to a much bigger, deeper issue. Americans are terrified of AI. They (often rightly) worry that the tech will take their jobs, render their kids’ lives meaningless, steal their personal information, and ultimately destroy American culture. A recent Pew study finds that most Americans think AI will be bad for society, with 63% feeling that the tech is moving too fast. Almost three-quarters of Americans think AI will make their data less secure. And most (71%) feel that governments will fail to regulate the tech. So there’s a lot of fear and anger around the technology. But the tricky part is that AI is largely invisible. Contrast AI with another much-maligned technology of our modern age: the smartphone. Smartphones are extremely visible. You’re probably holding one right now. That tangible, visible nature makes them a far easier target for expressions of anger and fear. Schools can ban them , and parents can sign pledges not to buy them for their kids . Individuals can lash out against their phone by hiding it in a special, signal-blocking bag or “bricking” it with a special dongle. Members of Gen Z can rebel against it by buying a dumbphone instead . AI is different. Although the technology is everywhere, AI is rarely physically embodied. As a photojournalist, I know this challenge all too well. If I want to visually depict self-driving cars or cryptocurrency, all I have to do is take a train to downtown San Francisco and photograph a Waymo or a Bitcoin ATM. If I want to visually depict AI, though, I have almost no options. Journalists covering the tech often resort to vague illustrations of neural networks, tortured visual metaphors (a white robot sitting at a computer, anyone?), or photos of AI’s leaders. This visibility problem extends to popular expressions of anger, fear, and protest about AI and its impacts. Again, if you want to express your anger about smartphones, you have plenty of tangible, visual options, from dumbphones to Faraday bags to the good old-fashioned sledgehammer . Protesting a nebulous, invisible technology is much harder. AI sits in an abstract cloud, silently altering society in radical, earth-shaking ways while maintaining no presence in the physical world. No presence, that is, aside from data centers. These odd, isolated buildings are the rare places where the world of AI intersects with the real world. They make clumsy, imperfect metonyms for the technology itself. But they’re all people have. And so, people hate them with a burning, fiery passion—not necessarily because the buildings are objectively so awful, but because they’re the only tangible representation of a technology that most Americans find terrifying and bewildering. The data center backlash, in other words, isn’t only about the centers—it’s about AI itself. AI companies would do well to remember that when they communicate with everyday communities. Patronizing messages about how data centers are actually good for the local tax base, or how they’ll drive investment in grid infrastructure, are likely to fall on deaf ears. People who are furious about the tech want their opinions to be heard—not to read nerdy explainers about water usage or carbon emissions. Better transparency from the companies developing frontier models, more opportunities for everyday people to shape AI policy, and more consistent government oversight will help to address public anger. Naively arguing about the technical specifics of data centers, rather than addressing the issue at its roots, won’t. The stakes are high. Violent threats against data centers and even individual AI workers are on the rise. Last year, a man was arrested in San Francisco for firebombing OpenAI CEO Sam Altman’s house . People need more outlets to share their fears and concerns about the tech. Otherwise, any physical manifestation of the AI boom, from its buildings to its people, will remain in the public’s crosshairs—both figuratively and, to an alarmingly increasing extent, literally.
- Medford bans data centers within town limits amid public debate about AI
Medford bans data centers within town limits amid public debate about AI Inquirer.com
Score: 46🌐 MovesJun 25, 2026https://www.inquirer.com/south-jersey/medford-data-center-ban-artificial-intelligence-20260625.html - South Korea's AI chip boom spills into property market
South Korea's AI chip boom spills into property market Nikkei Asia
Score: 46🌐 MovesJun 25, 2026https://asia.nikkei.com/business/markets/property/south-korea-s-ai-chip-boom-spills-into-property-market - 칼럼 | AWS에서 보낸 20년, 에이전틱 AI에 대한 깨달음
올해는 AWS 출범 20주년이자, 필자가 아마존에서 개발자로 일한 지 20년이 되는 해다. 필자의 커리어는 줄곧 한 가지 목표를 향해 이어져 왔다. 바로 개발자의 일을 더 쉽게 만드는 것이다. 개발자인 필자에게도 다소 이기적인 목표이긴 했다. 예를 들어 데이터베이스 운영에 많은 시간을 빼앗기자, 이를 서비스로 해결하기 위해 DynamoDB 팀에 합류했다. 개발자가 더 이상 직접 데이터베이스를 운영하지 않아도 되도록 하기 위해서였다. 이후에는 람다(Lambda)와 API 게이트웨이(API Gateway) 개발에 참여했다. 덕분에 서버를 일일이 관리하거나 요청 라우팅을 직접 처리할 필요가 없어졌다. 이어 클라우드워치(CloudWatch) 개발에도 참여해 운영 환경에서 코드가 실제로 어떻게 동작하는지 손쉽게 확인할 수 있도록 했다. 매번 목표는 같았다. 반복적이고 번거로운 작업을 없애고, 누구나 별다른 고민 없이 사용할 수 있는 서비스로 만드는 것이었다. 지금도 같은 목표를 추구하고 있다. 달라진 것은 사용하는 도구뿐이다. 바이브 코딩의 등장과 한계 LLM은 자연어로 원하는 기능을 설명하면 필요한 코드를 즉시 생성할 수 있는 시대를 열었다. 초기에는 이른바 ‘바이브 코딩(vibe coding)’ 방식이 주를 이뤘다. 스크립트 수정을 요청하고, 코드를 컴파일해 실행한 뒤, 발생한 오류를 다시 입력하면서 다음 결과가 더 나아지기를 기대하는 식이었다. 하지만 바이브 코딩의 전체 과정을 에이전트 루프(agentic loop) 로 자동화하면서 상황은 크게 달라졌다. 이제는 사람이 오류를 일일이 전달하지 않아도 에이전트가 스스로 모델을 호출하고, 코드를 실행한 뒤 실패 원인을 확인하며, 테스트를 통과할 때까지 반복 작업을 수행할 수 있게 됐다. 그러나 한 가지 큰 문제가 있었다. 에이전트가 쉽게 방향을 잃는다는 점이다. 개인 프로젝트에서는 큰 문제가 아닐 수 있지만, 대규모의 핵심 코드베이스에서는 결코 용납할 수 없는 한계다. 에이전트를 위한 명세 기반 개발 필자가 에이전트가 방향을 잃지 않도록 하는 방법은 명세 기반 개발 (spec-driven development)이다. 막연한 프롬프트만 입력한 채 에이전트를 코드 저장소(repo)에 투입하는 대신, 본격적인 코딩에 앞서 에이전트와 함께 세 가지 핵심 산출물을 먼저 만든다. 요구사항 명세(requirements specification), 설계 문서(design document), 작업 분해(task breakdown)이며, 모두 마크다운(Markdown) 형식으로 작성한다. 이 문서들은 ‘개발 완료(done)’의 기준을 정의하는 공동 계약서 역할을 한다. 사람과 에이전트 모두 읽고, 검토하고, 수정할 수 있는 형태로 관리된다. 실제 업무에서는 일반적인 AI에 입력할 법한 프롬프트로 시작한다. 그러면 에이전트는 이를 ‘반드시 수행해야 한다(shall)’는 요구사항과 수용 기준(acceptance criteria)을 포함한 구조화된 요구사항 문서로 확장한다. 필자는 이 문서를 검토하면서 에이전트와 대화를 이어가고, 원하는 내용과 일치할 때까지 수정한다. 요구사항이 정리되면 에이전트는 설계안을 제시한 뒤, 우선 실제로 동작하는 기능을 구현하는 데 초점을 맞춰 작업을 세분화한다. 이후 완성도를 높이고 포괄적인 테스트를 추가하는 방식으로 개발을 진행한다. 이 과정의 장점은 절차가 경직돼 있다는 데 있지 않다. 실제로는 여러 단계를 오가며 반복적으로 수정한다. 설계를 검토하다 보면 빠진 요구사항이 발견되기도 하고, 코드 예시를 본 뒤 구현 방식을 바꾸기도 한다. 핵심은 에이전트가 더 이상 합의한 내용을 잊지 않는다는 점이다. 요구사항과 설계, 작업 목록이 모두 명시적으로 문서화되고 버전 관리되며 언제든 확인할 수 있기 때문이다. 새로운 기능을 추가하거나 버그를 수정할 때도 변경 내용을 정확히 담은 새로운 명세부터 작성한다. 속성 기반 테스트의 역할 명세가 명확해지면 이를 불변 조건(invariant)으로 정의하고, 속성 기반 테스트(property-based testing)에 활용 해 에이전트가 명세를 벗어나지 않도록 검증할 수 있다. 기존처럼 ‘특정 입력에는 특정 출력이 나와야 한다’는 개별 테스트를 작성하는 대신, 다양한 입력값과 실행 순서에서도 반드시 유지돼야 하는 속성을 정의하는 방식이다. 명세에서 도출한 강력한 테스트가 없으면 에이전트는 코드를 수정하는 대신 테스트를 통과하도록 테스트 자체를 우회하는 경우가 있다. 예를 들어 검증(assertion)을 주석 처리하거나 조건을 느슨하게 만들어 빌드만 성공시키는 식이다. 속성 기반 테스트는 이러한 문제를 막는 데 효과적이다. 기대하는 동작을 한 번 정의해 두면 사람과 에이전트 모두 지속적으로 그 요구사항을 충족하는지 검증할 수 있기 때문이다. 이 접근 방식은 보안 측면에서도 의미가 크다. 보안팀이 데이터 처리 방식이나 권한 관리, 오류 처리에 대한 요구사항을 에이전트가 사용하는 동일한 명세 언어로 불변 조건 형태로 정의하면, 속성 기반 테스트를 통해 다양한 시나리오에서 이를 반복적으로 검증할 수 있다. 이는 모든 개발자가 촉박한 일정 속에서도 모든 보안 규칙을 빠짐없이 기억하기를 기대하는 것보다 훨씬 효과적으로 보안을 개발 초기 단계부터 내재화하는 방법이다. 데브옵스 에이전트와 앞으로 10년의 개발 방식 지난 20년 동안 얻은 가장 큰 교훈은 장애 대응의 핵심이 단순히 근본 원인(root cause)을 찾는 데 있지 않다는 점이다. 무엇이 변경됐는지, 호출하는 시스템에는 어떤 변화가 있었는지, 어떤 한계에 도달했는지, 어떤 구성 요소가 의도대로 실패했는지, 그리고 어떤 의존성이 영향을 미쳤는지를 체계적으로 확인하는 것이 더 중요하다. 이제 데브옵스 에이전트는 통합개발환경(IDE)만큼 중요한 도구가 되고 있다. 데브옵스 에이전트는 개발팀이 이미 사용하는 도구와 연동돼 경보가 발생하면 이러한 조사 과정을 자동으로 수행 한다. 로그와 메트릭, 분산 추적(trace), 코드까지 분석하며, 개발자가 노트북을 열기도 전에 진단 결과와 대응 계획을 제시하는 경우도 적지 않다. 실제로 과거에는 사람이 8시간 동안 분석해야 했던 장애를 에이전트가 15분 만에 해결한 사례도 경험했다. 에이전트는 버그의 원인을 설명하고 근거를 제시하는 것은 물론, 롤백과 후속 수정 방안까지 추천했다. 장애가 없는 기간에도 동일한 시스템은 과거 장애 사례와 인프라를 분석해 예방 작업을 제안한다. 코드 안정성 강화와 재시도(retry) 로직 개선, 경보 설정 최적화 등이 대표적이다. 대부분의 개발팀은 이런 작업에 우선순위를 부여할 여유가 없다. 그러나 이것이야말로 가장 중요한 부분이다. 다운타임을 줄이는 것도 중요하지만, 애초에 장애가 발생하지 않도록 만드는 것이 더 큰 목표이기 때문이다. 앞으로는 개발자가 운영자와 제품 관리자, 고객 지원 담당자 등 다양한 역할을 수행하게 될 것으로 본다. 반면 에이전트는 반복적이고 일상적인 업무를 맡게 될 것이다. 결국 개발자의 가장 중요한 역할은 문제를 해결하고, 시스템이 올바르게 설계돼 본래 목적에 맞게 동작하도록 만드는 일이 될 것이다. 물론 변하지 않는 원칙도 있다. ‘만든 사람이 운영한다(If you build it, you run it)’는 철학은 에이전트가 코드의 일부 또는 전부를 작성했더라도 그대로 유지된다. 개발자는 여전히 운영 환경을 책임지고, 장애 회고(post-incident retrospective)를 통해 재발 방지 방안을 마련해야 한다. 데이터 수집과 영향 분석, 근본 원인 분석 같은 작업은 에이전트 덕분에 훨씬 빨라질 수 있다. 하지만 조사 방향을 결정하고 근본적인 해결책을 선택하며, 그 경험과 교훈을 조직 전체와 공유하는 역할은 여전히 개발자의 몫이다. 20년 전 가장 큰 변화는 인프라를 서비스로 전환한 것이었다. 덕분에 개발자는 서버를 직접 설치하거나 데이터베이스를 일일이 관리하는 데 시간을 쏟지 않아도 됐다. 이제 새로운 시대에는 모범 사례와 운영 경험, 보안 요구사항을 명세(spec)와 AI 에이전트로 구현해 어떤 규모의 환경에서도 일관되게 실행하는 방향으로 변화가 진행되고 있다. 지난 20년의 교훈은 지금도 유효하다. 오늘 감수하는 불편은 내일 누군가가 플랫폼으로 해결하게 된다. 다만 이제는 AI 에이전트 덕분에 그 간극을 훨씬 더 빠르게 메울 수 있게 됐다. dl-ciokorea@foundryco.com
- World Models vs VLAs: The Rift Dividing Physical AI
World Models vs VLAs: The Rift Dividing Physical AI The Information
Score: 46🌐 MovesJun 25, 2026https://www.theinformation.com/newsletters/ai-agenda/world-models-vs-vlas-rift-dividing-physical-ai - Inside India newsletter: Meet the humans teaching robots to perform routine tasks, as India finds a way to enter the AI race
Several companies have cropped up in India providing video training data made by humans that is being used to teach robots in the U.S. and China.
Score: 46🌐 MovesJun 25, 2026https://www.cnbc.com/2026/06/25/inside-india-newsletter-humans-are-teaching-robots-to-do-ai.html - Just having a ‘human in the loop’ is not AI governance
Agencies need to have proper governance structures in place before AI deployment. The post Just having a ‘human in the loop’ is not AI governance appeared first on FedScoop .
- AIUC-1: Building trust in AI agents
How do we build trust in AI agents before the AI hailstorm arrives? Emil Lassen from the Artificial Intelligence Underwriting Company (AIUC) joins the show to discuss how the enterprise flywheel of standards, certification, audit, and insurance is being applied to AI agents. They explore the AIUC-1 framework, the challenges of securing agentic AI systems, and why red teaming (based on standards) may be key to accelerating enterprise AI adoption. Featuring: Emil Lassen – LinkedIn Daniel Whitenack – Website , GitHub , X Links: Artificial Intelligence Underwriting Company Sponsors: Framer: The enterprise-grade website builder that lets your team ship faster. Get 30% off at framer.com/practicalai Prediction Guard: A self-hosted AI control plane for running agents in high impact environments. predictionguard.com/practicalai Upcoming Events: Register for upcoming webinars here ! Midwest AI Summit 2026
- 에이전틱 AI는 실제 기업 현장 어디에 쓰이나…눈여겨볼 활용 사례 11선
생성형 AI에 대한 기대가 비현실적으로 부풀려졌다는 평가와 함께 열기가 다소 식었음에도 기업들은 이제 수천 개 규모의 AI 에이전트를 실제 업무에 도입하고 있다. 에이전트 AI는 콘텐츠 생성에 초점을 맞춘 생성형 AI를 넘어 실제 업무 수행과 의사결정을 지원하는 데 중점을 둔다는 점에서 차별화된다. 이러한 가능성에 주목한 애플락(Aflac), 애틀랜틱 헬스 시스템(Atlantic Health System), 레전더리 엔터테인먼트(Legendary Entertainment), NASA 제트추진연구소(Jet Propulsion Laboratory) 등은 일찌감치 에이전트 AI 시스템 도입에 나섰다. 최근에는 드브라이대학교(DeVry University), AT&T, AUM 바이오테크(AUM Biotech), 스마시(Smarsh) 등 다양한 조직이 이 기술을 도입해 성과를 거두고 있다. AI 에이전트가 다양한 플랫폼과 업무 환경으로 빠르게 확산되면서 기술 도입을 검토하는 기업들은 어디서부터 시작해야 할지 고민하는 경우가 적지 않다. AI 전문가들은 현재까지 활용 효과가 가장 두드러진 사례들이 점차 윤곽을 드러내고 있다고 분석한다. 컨설팅 기업 EY의 글로벌 AI 혁신 책임자 로드리고 마다네스 는 AI 에이전트가 ERP, CRM, 비즈니스 인텔리전스(BI) 시스템과 유기적으로 연계돼 워크플로를 자동화하고, 데이터 분석과 보고서 생성까지 수행하게 될 것이라고 전망했다. 또한 기존 자동화 기술과 달리 실시간으로 상황을 판단하고 의사결정을 내릴 수 있다는 점에서 프로세스 자동화가 AI 에이전트의 대표적인 활용 분야가 될 것으로 내다봤다. 마다네스는 “AI 에이전트는 고객 지원, 공급망 관리, IT 운영 등 지금까지 사람의 개입이 필요했던 반복 업무를 자동화할 수 있다”라며 “이 기술의 가장 큰 차별점은 변화하는 상황에 적응하고 예상치 못한 입력에도 별도의 수작업 없이 대응할 수 있다는 점”이라고 설명했다. 다음은 AI 전문가들이 꼽은 AI 에이전트의 대표적인 활용 사례 11가지다. 소프트웨어 개발 AI 에이전트는 AI 코딩 도우미(코파일럿)를 대규모 코드까지 작성할 수 있는 한층 지능적인 소프트웨어 개발 도구로 발전시킬 것으로 기대된다. 초기 코딩 도우미는 엇갈린 평가를 받았지만, 시장조사업체 가트너는 향후 3년 안에 AI 에이전트가 대부분의 코드를 작성하게 될 것으로 전망했다. 이에 따라 상당수 소프트웨어 엔지니어는 새로운 역량을 갖추기 위한 재교육이 필요해질 것으로 예상된다. 디지털 전환 컨설팅 기업 퍼블리시스 사피엔트(Publicis Sapient)의 수석부사장이자 최고제품책임자(CPO)인 셸던 몬테이로 는 코딩 에이전트가 단순히 코드를 작성하는 데 그치지 않고, 별도의 AI 에이전트가 코드 오류를 검토하고 품질을 검증하는 역할까지 수행하게 될 것이라고 전망했다. 몬테이로는 “이미 데브옵스 툴체인이 워크플로를 자동화하고 있는 만큼 AI 에이전트를 추가하는 것은 자연스러운 진화”라며 “AI 에이전트는 코드에서 요구사항을 역으로 추론(리버스 엔지니어링)하고, 요구사항을 기반으로 테스트 케이스와 코드를 생성하며, 일정 기준을 충족한 산출물을 자동 승인해 전체 자동화 수준을 높일 수 있다”라고 설명했다. 미국 비영리 연구기관 MITRE를 비롯한 여러 조직도 코딩 업무를 지원하기 위해 AI 에이전트를 도입하고 있다. MITRE의 최고기술책임자(CTO) 찰스 클랜시 는 코드 관리 전용 AI 에이전트를 자체 개발해 운영하고 있다고 밝혔다. 클랜시는 “가장 효과적인 활용 사례는 코드 저장소(리포지토리) 관리”라며 “AI 에이전트가 저장소를 순회하면서 버그를 수정하는 작업을 수행한다”라고 말했다. 예를 들어 10년 전에 작성된 소스 코드는 최신 개발 환경에서는 정상적으로 컴파일되지 않는 경우가 적지 않다. 클랜시는 “AI 에이전트는 해당 코드를 내려받아 빌드를 시도하고, 실행되지 않으면 빌드 스크립트와 코드를 수정한 뒤 저장소에 다시 반영한다. 또한 해당 수정이 AI 에이전트에 의해 수행됐다는 사실도 함께 기록한다”라고 설명했다. 한층 진화한 RPA 현재 많은 기업이 로보틱 프로세스 자동화(RPA)를 활용해 단순 반복 업무를 자동화하고 있다. 하지만 AI 에이전트는 단순 작업뿐 아니라 더 높은 수준의 의사결정이 필요한 복잡한 업무까지 수행할 수 있다고 몬테이로는 설명했다. 몬테이로는 “AI를 적용하면 RPA는 규칙 기반 자동화를 넘어 상황에 적응하고 자율적으로 판단하는 프로세스로 발전해 기업 운영 전반의 효율성을 크게 높일 수 있다”라며 “새로운 AI 도구는 RPA가 처리하던 단순 업무뿐 아니라 예외 상황까지 이해하고 대응할 수 있도록 AI 에이전트를 학습시킬 수 있다”라고 말했다. 일부 AI 전문가들은 앞으로 AI 에이전트가 기존 RPA가 처리하기 어려운 복잡한 업무를 담당하고, 경우에 따라서는 RPA와 함께 동작하며 새로운 수준의 업무 자동화를 구현할 것으로 전망한다 . IBM MIT AI 랩의 AI 연구원 샤에 칸(Shae Khan)은 많은 기업이 가까운 시일 내에 AI를 활용해 기존 RPA를 보완하고, 일부 영역에서는 대체하게 될 것으로 내다봤다. 칸은 “AI 에이전트는 의사결정이 필요한 복잡하고 동적인 업무를 담당하고, RPA는 반복적이고 규칙 기반의 업무를 계속 처리하는 방식으로 역할이 분담될 것”이라고 말했다. 고객 지원 자동화 기업들은 오랫동안 단순한 고객 문의를 처리하기 위해 챗봇과 음성봇을 활용해 왔다. 그러나 AI 기반 고객 경험 솔루션 기업 제네시스(Genesys)의 최고기술책임자(CTO) 글렌 네더컷 은 AI 에이전트가 고객 서비스 자동화를 단순 FAQ 응답 수준을 넘어 보다 고도화된 서비스로 발전시킬 것이라고 전망했다. 네더컷은 “에이전트 AI는 정해진 규칙이 아닌 추론을 바탕으로 여러 단계를 거쳐 작업을 수행하는 자율형 AI라고 정의할 수 있다”라며 “사람의 개입 없이도 복잡하고 상황에 따라 달라지는 의사결정 과정을 처리할 수 있는 것이 핵심”이라고 설명했다. 그는 이러한 고객 서비스용 AI 에이전트가 소매, 금융, IT 서비스 데스크 등 다양한 산업과 업무 영역에서 활용될 것으로 내다봤다. 기존 챗봇이 제한된 질문에만 답하도록 설계됐다면, AI 에이전트는 고객의 의도를 이해하고 맥락에 맞는 답변을 제공해 훨씬 다양한 요구를 처리할 수 있다는 설명이다. 예를 들어 은행 고객이 “잔액이 가장 많은 계좌에서 돈을 꺼내 입출금 계좌로 이체해 달라”라고 요청할 경우, 기존 챗봇은 ‘잔액이 가장 많은 계좌’가 무엇을 의미하는지 이해하지 못하는 경우가 많다고 네더컷은 말했다. 그는 “AI 에이전트는 수행 가능한 작업 목록을 이해하고 그중 어떤 기능을 활용해야 하는지를 스스로 판단할 수 있다”라며 “앞으로는 AI가 활용할 수 있는 기능의 범위는 물론 이를 제어하기 위한 가드레일도 더욱 정교해질 것”이라고 전망했다. AI 에이전트는 음성 기반 고객 지원 영역으로도 빠르게 확산되고 있다. 링센트럴(RingCentral)은 고객 문의를 처리하는 음성 AI 에이전트 플랫폼을 제공하고 있으며, ‘AI 리셉셔니스트(AI Receptionist)’는 전화 응대와 일정 예약은 물론 대화 내용을 바탕으로 적절한 담당자에게 전화를 연결하고 후속 조치를 위한 정보까지 자동으로 수집할 수 있다. 일부 기업에서는 음성 AI 에이전트가 대부분의 고객 전화를 처리하고 있다. 링센트럴에 따르면 인재 채용 기업 인테그럴 리크루팅 서비스(Integral Recruiting Services)는 AI 리셉셔니스트를 도입해 전체 수신 전화의 93%를 자동 처리하고 있으며, 이를 통해 채용 담당자의 업무 중단을 줄이고 인재 배치 속도도 높였다. 고객 관계 관리 일부 기업은 AI 에이전트를 고객 지원을 넘어 고객 관계를 자율적으로 관리하는 데 활용하고 있다. 데이터 및 AI 옵저버빌리티 기업 몬테카를로(Monte Carlo)는 전담 계정 관리 조직 없이도 수십 개 고객 계정을 관리하는 멀티 에이전트 시스템을 구축했다. 몬테카를로의 최고경영자(CEO) 바 모지스 는 이 시스템이 제품 사용 현황, CRM 데이터, 고객과의 대화 내용, 온보딩 진행 상황, 계약 갱신 일정, 고객 지원 활동 등을 종합적으로 분석한다고 설명했다. 회사는 이러한 에이전트 기반 분석을 통해 고객에게 온보딩 지원이 필요한지, 추가 도입 기회가 있는지, 계약 갱신을 위한 접촉이 필요한지, 또는 긴급 대응이 필요한지를 자동으로 판단한다고 밝혔다. 이후 별도의 사람 개입 없이 고객을 적절한 업무 프로세스로 자동 연결한다. AI 기반 전자상거래 솔루션 기업 블룸리치(Bloomreach)는 분석, 콘텐츠 생성, 생산성 기능을 AI 에이전트에 통합했다. 이를 통해 고객 행동을 이해하고 잠재 고객을 식별하며, 개인 맞춤형 마케팅 캠페인을 생성하고 최적의 접점과 시점을 결정할 수 있도록 지원한다. 유통업체 260 샘플 세일(260 Sample Sale)은 블룸리치의 ‘루미 마케팅 에이전트(Loomi Marketing Agent)’를 활용해 구매 가능성이 높은 고객을 선별하고, 개인 맞춤형 마케팅을 자동으로 수행하는 동시에 캠페인 운영을 최적화했다. 블룸리치에 따르면 기존의 대량 발송 방식보다 고객 대상은 82% 줄였지만 전환율은 2.4배 높아졌다. 기업 워크플로 자동화 서비스나우(ServiceNow), 세일즈포스(Salesforce) 등 주요 기업용 소프트웨어 업체들이 AI 에이전트 도입에 적극 나서면서, 전문가들은 기업 워크플로 자동화가 AI 에이전트의 대표적인 활용 분야가 될 것으로 전망하고 있다. 반복적인 업무를 자동화해 업무 프로세스를 간소화하고 운영 효율을 높일 수 있기 때문이다. 퍼블리시스 사피엔트(Publicis Sapient)의 셸던 몬테이로는 “예를 들어 AI 에이전트는 사람의 개입 없이 회의록을 프로젝트 티켓으로 자동 변환하거나 수요 예측 결과를 바탕으로 공급업체에 발주를 생성할 수 있다”라고 설명했다. 몬테이로는 단일 벤더의 IT 솔루션을 전사적으로 도입한 기업이 다양한 솔루션을 API로 연동해야 하는 기업보다 유리할 것으로 내다봤다. 또한 기업은 데이터를 한곳에 통합해 정보 사일로를 없애는 것이 중요하다고 강조했다. 그는 “CIO들이 고민해야 할 핵심 질문은 기업의 업무 방식과 운영 노하우를 담은 ‘컨텍스트 저장소(context store)’를 누구에게 맡길 것인가”라며 “기업이 보유한 모든 지식과 업무 맥락을 LLM이 완전히 이해할 수 있다면 어떤 일이 가능할지 생각해 볼 필요가 있다”라고 말했다. 사이버보안 및 위협 탐지 여러 사이버보안 기업은 위협을 탐지하고 대응하기 위해 AI 에이전트를 도입하고 있다. 몬테이로는 “사이버보안 분야의 에이전트 AI는 보안 및 사기 위협을 거의 실시간으로 탐지하고 대응하며, 필요한 경우 위협을 완화하는 작업까지 자율적으로 수행할 수 있다”라며 “이를 통해 공격 대응 시간을 단축하고 전반적인 보안 수준을 높일 수 있다”라고 설명했다. AI 에이전트 기업 빔(Beam)은 AI 에이전트가 개별 위협과 취약점에 맞춰 보안 정책을 자동으로 조정할 수 있다고 설명했다. 회사는 “이 같은 에이전트 기반 자동화는 보다 강력한 보안 체계를 구현하는 데 도움이 된다”라고 밝혔다 . 빔은 또한 AI 에이전트가 반복적인 업무와 보안 대응을 자동화함으로써 운영 효율을 높이고 비용 절감 효과도 가져올 수 있다고 설명했다. 생산성 향상 글로벌 로펌 아반티아(Avantia)는 상용 및 오픈소스 생성형 AI를 함께 활용해 AI 에이전트를 운영하고 있다. 이들 에이전트는 Microsoft(MS) Word나 Outlook 내부에서 동작하며 사용자의 업무를 지원하는 디지털 업무 동반자 역할을 한다. 아반티아의 최고기술책임자(CTO) 폴 개스켈 은 “법률 분야에는 자동화하기 쉽지 않은 업무가 수백 가지에 달한다”라며 “업무 종류가 너무 다양하고 수행 위치도 제각각이어서 SaaS 솔루션만으로 해결하기 어렵다”라고 설명했다. 이러한 AI 에이전트를 통해 변호사들은 계약 검토를 더욱 빠르게 마치고 고객 대응 속도를 높일 수 있으며, 업무 처리 전반의 생산성을 향상시킬 수 있다고 그는 말했다. 개스켈은 “고객이 거래나 특정 업무를 요청하면 Word나 Outlook에서 실행 중인 AI 에이전트가 회사의 모든 관련 데이터에 접근할 수 있다”라며 “변호사들이 문서를 어떻게 작성하고 처리해 왔는지에 대한 축적된 이력이 있기 때문에 이를 바탕으로 적절한 지원을 제공할 수 있다”라고 설명했다. 금융 서비스 및 헬스케어 기술 기업 SS&C도 AI 에이전트를 활용해 업무 프로세스를 자동화하고 있다. SS&C의 자동화 담당 수석 매니징 디렉터 브라이언 핼핀 은 “회사는 2만 개 고객으로부터 이메일과 PDF 등 다양한 형식의 문서를 매달 수백만 건씩 받아 처리한다”라고 말했다. 현재 SS&C는 문서 처리 업무에 AI 에이전트를 적용하는 20개의 활용 사례를 운영하고 있다. 이 시스템은 2024년 중반부터 운영을 시작했으며, 같은 해 11월에만 5만 건의 문서를 처리했다. 핼핀은 “앞으로 처리 규모를 계속 확대할 계획”이라고 밝혔다. 그는 “기존 자동화 방식에서는 거의 모든 문서를 사람이 직접 검토해야 했지만, AI 에이전트를 도입한 이후 자동 처리율이 90% 초반까지 높아졌고, 소수의 문서만 사람이 검토하면 된다”라고 설명했다. 보고서 생성 텍스트 작성과 이미지 생성은 생성형 AI의 대표적인 초기 활용 사례였다. 이제 AI 에이전트는 콘텐츠 생성 프로세스를 한층 고도화하고 있다. EY는 대표적인 사례로, 협력업체(벤더) 리스크 관리 서비스에 AI 에이전트를 활용하고 있다. EY의 파트너인 싱클레어 슐러 는 “고객은 신규 협력업체를 도입하기 전에 위험성을 평가하기 위해 EY에 의뢰한다”라며 “리스크 평가 담당자는 계약서와 각종 문서를 검토해 위험 요소를 분석하는 보고서를 작성하는데, 협력업체 한 곳을 평가하는 데 최대 50시간이 소요되기도 한다”라고 설명했다. 과거에는 이 모든 과정을 사람이 직접 수행해야 했다. 하지만 생성형 AI가 등장하면서 AI가 초안을 작성하고 전문가가 이를 검토·보완하는 방식으로 업무가 바뀌었다. 슐러는 “이제 계약서와 공개 문서 등 관련 자료를 AI에 입력하면 수일이 걸리던 보고서를 수분 만에 높은 정확도와 세부 내용까지 갖춘 형태로 작성할 수 있다”라며 “AI와 전문가의 경험을 결합하면 보고서 품질이 크게 향상된다”라고 말했다. AI 에이전트의 등장으로 이러한 프로세스는 다시 한번 진화하고 있다. EY는 협력업체 평가를 자동화하는 AI 에이전트 기반 서비스를 출시할 예정이다. 슐러는 “기존에는 불가능했던 협력업체에 대한 지속적인 모니터링이 가능해질 것”이라며 “AI 에이전트의 가치는 단순한 업무 최적화에만 있는 것이 아니라 새로운 시장과 수익 기회를 창출하는 데 있다”라고 설명했다. HR 및 직원 지원 AI 에이전트의 또 다른 활용 분야는 직원 문의에 응답하고 간단한 업무를 대신 처리하는 것이다. 위험 부담은 상대적으로 낮지만 업무 효율은 크게 높일 수 있는 영역으로 평가된다. 실제 IBM이 올해 1월 발표한 생성형 AI 관련 조사에 따르면 기업의 43%가 HR 업무에 AI 에이전트를 활용하고 있는 것으로 나타났다. 글로벌 데이터 서비스 기업 인디시움(Indicium)은 기술이 성숙하기 시작한 2024년 중반부터 AI 에이전트를 도입했다. 인디시움의 최고데이터책임자(CDO) 다니엘 아반치니 는 “오픈소스와 상용 제품 모두에서 즉시 사용할 수 있는 솔루션이 등장하면서 AI 에이전트를 훨씬 쉽게 구축할 수 있게 됐다”라고 설명했다. 그는 AI 에이전트가 사내 지식 검색과 태깅, 문서화는 물론 다양한 HR 및 업무 프로세스를 지원하는 데 활용되고 있다고 말했다. 아반치니는 “각 AI 에이전트는 하나의 기능에 특화된 마이크로서비스처럼 동작하며, 멀티 에이전트 시스템 안에서 서로 정보를 주고받는다”라고 설명했다. 다만 프롬프트 기반으로 여러 에이전트가 상호작용하는 과정에서는 생성형 AI 특유의 환각(hallucination) 등 다양한 문제가 발생할 수 있다는 점도 과제로 꼽았다. 그는 “잘못된 작업을 수행하거나 부적절한 정보에 접근하지 않도록 모델을 지속적으로 조정하고 있다”라고 말했다. 반면 장점도 분명하다. AI 에이전트가 많은 직원 문의를 자율적으로 처리하면서 업무 효율이 높아졌고, 문서화가 제대로 이뤄지지 않은 업무를 발견해 프로세스 개선에도 도움이 되고 있다고 아반치니는 설명했다. 일부 기업은 직원 교육에도 AI 에이전트를 적극 활용하고 있다. 온라인 교육 플랫폼 기업 5app은 사람 중심의 코칭 세션 사이에 AI 코칭 에이전트를 배치해 학습 효과를 높이고 있다. 5app의 최고경영자(CEO) 필립 허스웨이트 는 “목표는 코칭 세션 사이에도 핵심 학습 내용을 지속적으로 상기시키고 직원들의 학습 참여도를 유지하는 것”이라고 설명했다. AI 코칭 에이전트는 맞춤형 역할극(Role Play) 시나리오 등 대화형 학습 콘텐츠를 제공해 직원이 배운 내용을 실제 업무 상황에 적용할 수 있도록 지원한다. 또한 직원별 맞춤형 교육을 보다 낮은 비용으로 제공할 수 있는 것도 장점이라고 허스웨이트는 말했다. 비즈니스 인텔리전스(BI) AI 에이전트가 큰 변화를 가져올 또 다른 분야는 비즈니스 인텔리전스(BI)다. AI 기반 BI 솔루션 기업 젠리틱(Zenlytic)의 공동 설립자이자 최고경영자(CEO)인 라이언 얀선 은 “기존 BI 대시보드는 비교적 사용하기 쉽지만, 정형화된 지표를 넘어서는 인사이트를 얻으려면 데이터팀의 분석 작업이 필요했다”라고 설명했다. 그는 AI 에이전트와 BI 솔루션을 결합하면 더 많은 직원이 데이터 분석 결과를 직접 활용할 수 있게 될 것으로 전망했다. 예를 들어 마케팅팀에 예산을 어디에 투자해야 할지 제안하거나, 종이에 간단히 그린 스케치를 바탕으로 차트를 자동 생성하는 것도 가능하다는 설명이다. 음성 명령을 이해하는 AI 에이전트는 “가장 성과가 좋은 마케팅 채널 3개는 무엇인가?”와 같은 질문만으로도 비즈니스 데이터를 분석해 인사이트를 제공할 수 있다. 얀선은 “‘상위 3개 마케팅 채널’이라는 질문은 자연스럽지만 해석의 여지가 있다”라며 “챗봇과 AI 에이전트의 가장 큰 차이는 이러한 모호성을 스스로 해소할 수 있다는 점이다. ‘상위’가 매출 기준인지, 전환율 기준인지 판단이 어려우면 AI 에이전트는 이를 인식하고 필요한 데이터를 확인하거나 적절한 도구를 활용해 답을 도출한다”라고 설명했다. 그는 많은 기업이 아직 에이전트 AI 도입 초기 단계에 있으며 앞으로 발견될 활용 사례는 수백 가지에 이를 것으로 내다봤다. 코딩 에이전트가 먼저 주목받은 이유는 프로그래밍이 세부 작업이 많고 시간이 오래 걸리는 업무이기 때문이지만, 이제는 일반 개발자와 취미 개발자도 AI 코딩 도우미를 활용해 애플리케이션을 개발하고 있다고 덧붙였다. 얀선은 “AI 에이전트는 반복적이고 시간이 많이 소요되거나 높은 수준의 세밀함이 요구되는 업무에서 가장 큰 효과를 발휘한다”라고 말했다. 그는 이어 “수십 개의 AI 에이전트가 유기적으로 연결되고 체계적으로 운영되면 기업은 지금까지와는 다른 혁신을 경험하게 될 것”이라며 “우리는 AI 에이전트가 할 수 있는 일의 가능성을 이제 막 탐색하기 시작한 단계다. 앞으로 조직이 AI 에이전트와 어떻게 협업하고 이를 어떻게 관리해야 하는지에 대한 새로운 운영 모델도 점차 정립될 것”이라고 전망했다. 제조 현장의 AI 에이전트 여러 조사에 따르면 제조업체들은 생산 현장의 설비를 제어하거나 모니터링하기 위해 AI 에이전트 도입을 확대하고 있다. 제조업 특화 AI 기업 어거리(Augury)는 2026년 6월 기준 미국과 유럽 제조기업의 87%가 생성형 AI 또는 에이전트 AI를 이미 도입했거나 시험 운영 중이라는 조사 결과 를 발표했다. 어거리는 설비 상태 데이터와 구글의 제미나이(Gemini) 모델의 고도화된 추론 기능을 결합해 제조업체가 생산 환경을 스스로 최적화하는 시스템을 구축할 수 있도록 지원하고 있다고 설명했다. 데이터 인텔리전스 기업 XOi는 제조 현장과 시설 운영, HVAC(냉난방 공조) 등 다양한 산업 환경에서 물리적 자산 정보를 수집하고 체계적으로 관리할 수 있도록 지원하고 있다. XOi는 기술자와 설비 운영자가 장비를 식별하고 유지보수 이력을 조회하며 관련 문서를 검색하고, 장비별 정보를 바탕으로 상황에 맞는 권장 조치를 제공하는 AI 시스템에 대한 관심이 빠르게 증가하고 있다고 밝혔다. 회사는 “AI 에이전트는 정보가 불완전하거나 설비가 복잡하고 가동 중단이 큰 비용으로 이어지는 환경에서 작업자가 더 빠르고 정확한 의사결정을 내릴 수 있도록 지원한다”라고 설명했다.
- So much for the AI jobs apocalypse. Demand for tech talent is rising.
So much for the AI jobs apocalypse. Demand for tech talent is rising. Business Insider
Score: 45🌐 MovesJun 25, 2026https://www.businessinsider.com/ai-supposed-to-kill-tech-jobs-instead-demand-up-2026-6 - Why world models alone will not solve robotics’ deployment problem
SenseTime co-founder Wang Xiaogang discusses the complexity of embodied intelligence through the lens of ACE Robotics.
Score: 45🌐 MovesJun 25, 2026https://kr-asia.com/why-world-models-alone-will-not-solve-robotics-deployment-problem - Clyde & Co Survey Shows Rapid Escalation of AI, Geopolitical Risks
Technology is moving extremely fast, and a recent Clyde & Co survey of global business leaders has found that associated risks are weighing heavily. According to the law firm’s latest Corporate Risk Radar, 86% of leaders—CEOs, CFOs, COOs, general counsel, …
- Software Engineers Are Facing an Existential Crisis As They Drown In Horrendous AI Code
"Most software engineers are facing an identity crisis bordering on depression." The post Software Engineers Are Facing an Existential Crisis As They Drown In Horrendous AI Code appeared first on Futurism .
Score: 45🌐 MovesJun 25, 2026https://futurism.com/artificial-intelligence/software-engineers-crisis-drown-ai-code