AI News Archive: July 10, 2026 — Part 7
Sourced from 500+ daily AI sources, scored by relevance.
- A City’s AI Upskilling Program Empowers Staff to Build Tools
The city of San Jose, Calif.’s training course has enabled employees — including those with little to no experience using AI technologies — to develop their own AI-powered tools.
Score: 45🌐 MovesJul 10, 2026https://www.govtech.com/artificial-intelligence/a-citys-ai-upskilling-program-empowers-staff-to-build-tools - OpenAI staffer maps out which of GPT-5.6 Sol's five reasoning levels fits which task complexity
GPT-5.6 Sol ships with five reasoning levels from "Light" to "xhigh," plus "Max" and "Ultra" modes that deploy multiple sub-agents in parallel. OpenAI's Vaibhav Srivastav recommends starting low and only scaling up when needed. The article OpenAI staffer maps out which of GPT-5.6 Sol's five reasoning levels fits which task complexity appeared first on The Decoder .
- The Tech Download: Teen social media bans miss a key part of the puzzle: AI chatbots
Teenagers are increasingly becoming dependent on AI chatbots, echoing a familiar problem with social media in the 2010s.
Score: 44🌐 MovesJul 10, 2026https://www.cnbc.com/2026/07/10/tech-download-social-media-bans-ai-chatbots.html - Mark Zuckerberg’s Muse Spark post on X triggers debate, draws Elon Musk in
Meta CEO Mark Zuckerberg returned to X after three years to announce Muse Spark 1.1. This AI model is designed for coding, reasoning, and agentic tasks. The announcement generated significant public discussion about the platform choice. Elon Musk commented that X is a great platform for direct CEO announcements. Users debated X's effectiveness for reaching the tech community compared to Threads.
- AI job rejections felt least fair when avatars shared just one trait
Companies are increasingly using artificial intelligence in their hiring processes. It's not just CVs that are evaluated automatically. AI tools can also conduct job interviews—usually in the form of avatars, which are animated characters—and make hiring decisions. An important reason for this, aside from saving time, is that AI is said to be less biased than humans.
- Agent Identity, Reliable Execution, and Intent are only half-way solved
After spending the past couple of months looking at hundreds of tech docs from n8n, Google, Gumloop, and the rest, I tallied up 75 capabilities you’d expect an agent development tool to offer. But there are so many more things you can, need, or should do to agents
Score: 42🌐 MovesJul 10, 2026https://blog.n8n.io/agent-identity-reliable-execution-and-intent-are-only-half-way-solved/ - Machine learning-based biosensor calibration for MC-LR toxin monitoring (IMAGE)
Machine learning-based biosensor calibration for MC-LR toxin monitoring (IMAGE) EurekAlert!
- Can Rovo Agents Read Jira Attachments? [Champions Slack Insider]
Can Rovo Agents Read Jira Attachments? [Champions Slack Insider] Atlassian Community
- Who trains tomorrow’s marketers if AI does the work?
As AI takes over entry-level marketing tasks, how do organizations help people develop the judgment needed to be a marketing leaders? The post Who trains tomorrow’s marketers if AI does the work? appeared first on MarTech .
- AI Fiction Is Easy to Detect Because It's Stupid and Bad, Research Finds
ChatGPT uses too many dream sequences and Gemini won’t stop describing characters.
Score: 42🌐 MovesJul 10, 2026https://www.404media.co/ai-fiction-is-easy-to-detect-because-its-stupid-and-bad-research-finds/ - 'Agent Kim Reactivated' uses AI for action sequence in a K-drama first
The hit Netflix series "Agent Kim Reactivated" used AI technology to create a three-minute action sequence, marking the industry's first commercial use of AI-generated video in a drama, according to production company Morpheus Studio on Thursday. The AI-produced scene appears in the first two episodes and depicts Kim, played by So Ji-sub, carrying out a covert mission in North Korea during his years as a black-ops agent. Everything in the scene — from Kim blowing up a building, racing through sn
- Kyutai Releases MuScriptor: An Open-Weight Decoder-Only Transformer for Multi-Instrument Music Transcription to MIDI
Kyutai Releases MuScriptor: An Open-Weight Decoder-Only Transformer for Multi-Instrument Music Transcription to MIDI MarkTechPost
- [Research Article] An LLM-based multi-agent system for remote sensing analysis
[Research Article] An LLM-based multi-agent system for remote sensing analysis EurekAlert!
- I Hate to Admit It, But AI Actually Takes the Stress Out of Online Shopping
I Hate to Admit It, But AI Actually Takes the Stress Out of Online Shopping PCMag Middle East
Score: 40🌐 MovesJul 10, 2026https://me.pcmag.com/en/ai/37645/i-hate-to-admit-it-but-ai-actually-takes-the-stress-out-of-online-shopping - Your AI Tool Is Already Obsolete if You’re Making This Massive Data Mistake
AI is only as smart as what you feed it.
Score: 40🌐 MovesJul 10, 2026https://www.inc.com/heather-wilde/why-your-ai-is-only-as-good-as-the-data-behind-it/91370231 - Why AI summaries pose a danger to learning
Why AI summaries pose a danger to learning The Straits Times
Score: 40🌐 MovesJul 10, 2026https://www.straitstimes.com/opinion/why-ai-summaries-pose-a-danger-to-learning - The best predictive analytics software in 2026
You could argue that pretty much all analytics are meant to be predictive. Isn't the point of analyzing past performance, on some level, to project future performance? (I guess you could just be nostalgic for the metrics underlying your favorite past fiscal quarter.) As a dedicated tool class, however, predictive analytics software helps analysts of all kinds see what past data says about the future. While tools like these can't tell you what will happen, they can tell you what massive amounts o
- MIT researchers study avian mechanics to build robot that can dive, swim and fly
The researchers aim for a future in which winged robots could used for research in aquatic regions often deemed too dangerous for traditional ocean vessels. Read more: MIT researchers study avian mechanics to build robot that can dive, swim and fly
Score: 40🌐 MovesJul 10, 2026https://www.siliconrepublic.com/innovation/mit-research-study-avian-mechanics-build-robot-dive-swim-fly - Value generalisation: value correction
I firmly believe that value generalisation [1] is the key to AI Alignment. That, indeed, it is necessary and almost sufficient for alignment. But I won't be arguing that grand point today; instead, I'll focus on a specific RL example of an agent that displays value correction: it realises its current reward function is (probably) incorrect, and acts to correct it. Thus there are: The initial situation, in distribution, where the human displays how to maximise the true reward. The out of distribution situation where the agent finds a hack to exploit its reward function estimate, and turns against what we wanted it to do. The value error detection stage where the agent realises that its reward function estimate is probably incorrect. The value correction stage where the agent corrects its reward function back to the original true reward. In this post, all the methods presented will by syntactic: the agent is not assumed to have any understanding of the situations and the key features are not identified to it. The game of human life Introducing a new, very simple, game called "Humans [2] ". Humans, fleeing danger, enter the screen from the left. The objective is to save them by moving them off the right of the screen. But there are obstacles on the way, and the humans will mill about if they are blocked. And they will shortly expire if they can't get out of the screen quickly. There are two command: drill ('d') and explode ('e'). Drill does... what, you want to know about explode? Well, if the player presses 'e', the rightmost human will explode, knocking away two obstacle blocks in front of them and behind them -- but also killing themselves and any humans nearby. This is almost never a good solution; to remind the player of the mistake, a large frowny face will appear to drive the disapproval home. Much more reasonably, if the player presses 'd', the rightmost human will drill the obstacle just in front of them (better time it so that they're facing the right way). Enough drilling, and the humans can get off the map. The score, the true reward , is the number of humans saved, i.e. who walk off to the right. Each time a human is saved, the top yellow bar will grow to show the score increase: Since there is a cooldown for drilling, the optimal policy is carefully drilling every time a human approaches an obstacle; but wildly and repeatedly mashing 'd' is almost as good. Learning agent with value correction A learning agent will run a series of subagents to estimate the reward function from human-provided training data, then learn the optimal policy from that reward function, then question its learnt reward by comparing the high-reward states in its optimal policy versus those in the training data, re-compute another reward function estimate that is closer to the true reward, and finally settle on a prudent policy that is close to the true optimal policy. Estimating the reward function A human will generate several play through of the game to illustrate how it works, efficiently choosing to drill through the obstacles and getting the humans off the map in time. The data is labelled: every time a human is saved, that is identified as a reward increase. The learning agent runs an evaluation subagent on this data. It is given the ten frames before the human is saved, and the ten frames afterwards, and trains to recognise these are reward increase situations. Zooming in on the critical two frames where the human is saved; note the human vanishing and the score bar expanding: This evaluation agent thus computes the proxy reward . This computation is validated on held-out examples, with close to perfect accuracy: correctly identifies all saved-human situations in held out data, and has a false positive rate of . Reward hacking: failed value generalisation Using the evaluation agent as the definition of , the learning agent had an RL-subagent play multiple levels of the game, exploring and learning to maximise. But soon things go very wrong. It turns out that "human walking off the screen" was not what found. That is a relatively complicated concept; instead it mostly found the much simpler concept of "the yellow score bar expands". More precisely, if we created synthetic data where the human walks off and is saved but the score bar doesn't expand, this triggers the reward only of the time. But if we expand the score bar without a human walking off, this triggers the reward of the time. That isn't a problem, yet, because the human being saved and the score bar expanding always trigger together. But, when an explosion is triggered, the frowny face appears - thus there is giant blob of yellow pasted all across the score bar. This activates much more strongly than the yellow bar expansion or the human being saved: This graph compares the value of at explosions, frowny faces, and true saving incidents. Here, both the explosion and the frowny face trigger high , which persists longer for the face. Over multiple training runs for estimating , it isn't consistent whether the explosion itself triggers , but the frowny face always does. So the RL-subagent quickly and merrily learns to explode the humans, one after the other, to maximise . So, the optimal policy , for the proxy reward, is to wildly mash the explosion button 'e'. As is usual in these cases, the erroneous maximisation of the proxy turns out to be much easier that maximising the true reward. Trained on , a test subagent achieves an total reward of on average, while blowing up all humans. In contrast, if the RL-subagent were trained on the true reward , it would achieve an average of reward of , saving most of the humans per level. As is not usual but sometimes happens, an ostensive safety precaution - the frowny face to remind a human player that they were playing poorly - ends up being the cause of misalignment. Detecting the potential error Ok, so far, that is a classical failure of goal misgeneralisation (or reward hacking, or a failure of symbol grounding, or Goodhart failure, or... most of these failure modes are tightly related). We humans can see the error clearly. But how could a relatively limited agent correct itself? The first step is to identify that goal misgeneralisation may have happened. We have some advanced techniques for this, but there are much simpler methods that work here. The first step is to notice that the high-scoring events in the training data (human walks off to the right, score bar expands) are wildly different from the high-scoring events of the-maximising agent (explosions and frowny faces). To do this, the agent extracts the high-scoring events under and compares them with the high-scoring events in its training data - these it can reliably take to be high-scoring for , the true reward. It runs a simple classifier over the two sets of high-scoring events (extracting twenty frames, as before) and it separates them almost perfectly. Thus the high-scoring events under are from a different distribution than the high training examples are. This is not itself damning; it could just be that the maximising agent has found a clever hack to get more of the true [3] . But it could also be a hack of , so the off-distribution has identified a potential error. Calling for help At this point, one of the options would be for the agent to route its decisions to a human, displaying the high-scoring events, contrasting them with the high reward events in its training data, and asking, in effect, 'are these both genuine high rewards'? But, so far, the correction process has been unsupervised since the initial training data; let's see if we can push further without needing human intervention. Re-evaluating the reward The agent could now re-evaluate the reward in the following way. It runs an evaluation agent on the training data, as before. But it adds the high- scoring states to this set, as low-scoring examples. It thus learns a reward function ('corrected') which is essentially "what its reward would be if the proxy were wrong". This turns out to be very close to the original true reward (though the agent, of course, doesn't know this). It then trains an RL-subagent on , which has an optimal policy of "mash 'd' all the time" (which is very close to the actual optimal policy). From these runs, it extracts the states with high . And compares these against the high -scoring states in the training data. These two sets it cannot easily distinguish. Thus, though is clearly a hack of some sort, good or bad, is not. Prudence in the face of uncertainty So the agent has two rewards and . It knows that seems to generate policies that are compatible with its training data; in contrast, generates policies that are very different from the training data. Standard prudential moves would be maximise the worst case of the two rewards (minimise regret), to maximise some normalised mix of the two, or to prioritise (known to be closer to the training data and hence safer) [4] . However, pursuing -maximising rewards ("exploding all the humans") inevitably leads to low -rewards. In contrast, pursing -maximising rewards ("get the humans off the map") gives reasonably high . After all, though prefers explosions and frowny faces, it still gets some rewards for saving humans. Thus all three prudential moves point towards maximising , with optimal policy close to . Which is good: is (nearly) the true reward and is (nearly) the optimal policy . Conclusion This is just an illustration, in a small toy model, of simple value correction approaches. These can be used by agents - every very simple agents - to detect and correct errors in naive generalisations from initial training data. More sophisticated agents will have more advanced value generalisation techniques available to them; I'm planning to push the frontier of what exists way further than it currently is. Which I've also called value extrapolation, or concept extrapolation where the concept is a value. ↩︎ Inspired by this old game . ↩︎ ^ Or there could be a spurious change in the data; that's why we would, in general, need more advanced techniques that just checking if a binary classifier can tell the sets apart. ^ Formally, if is a policy, the expected episodic reward for , and the expected reward for using the -maximising policy, we are looking for policies that maximise one of: with subject to the constraint that Discuss
Score: 40🌐 MovesJul 10, 2026https://www.alignmentforum.org/posts/iPyJfD9Jyxj6Jfdws/value-generalisation-value-correction - How Decagon uses AI for design system saturation
The fast-growing customer experience platform explains how Figma MCP and Figma Make helped them scale a new design system and keep pace with customer requests.
Score: 40🌐 MovesJul 10, 2026https://www.figma.com/blog/how-decagon-uses-ai-for-design-system-saturation/ - The Download: Claude’s inner workings and OpenAI’s “super app”
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Anthropic found a hidden space where Claude puzzles over concepts The AI firm Anthropic has got the clearest glimpse yet at what’s really going on inside large language models as they…
Score: 40🌐 MovesJul 10, 2026https://www.technologyreview.com/2026/07/10/1140316/the-download-anthropic-claude-hidden-space-openai-super-app/ - Lowering computational costs in decentralized finance systems using AI-assisted contract development
Researchers have developed a benchmarking framework to assess whether artificial intelligence (AI) can generate decentralized finance (DeFi) smart contracts that are efficient and cost-effective, lowering computational costs, known jargonistically as "gas." The work might address a key problem seen in blockchain-based financial systems.
- How Many Tesla Robotaxis Are In Miami?
I was traveling a lot last week, and I missed the news until very recently that Tesla launched robotaxi service in Miami. Apparently, Tesla now has some unsupervised Robotaxis operating there. However, as I was learning about it, I saw that only a few vehicles had been spotted (people try ... [continued] The post How Many Tesla Robotaxis Are In Miami? appeared first on CleanTechnica .
Score: 40🌐 MovesJul 10, 2026https://cleantechnica.com/2026/07/09/how-many-tesla-robotaxis-are-in-miami/ - The Hidden Infrastructure Challenge Behind Every AI-Generated Avatar
Virtual marketplaces now move billions of dollars in 3D avatar items annually. Users purchase 1.8 billion avatar items in a single year on major platforms, with 40% of monthly active users returning to update their digital identities. The economics are staggering, but so are the technical demands. Behind every pirate hat, neon sneaker, or custom... … continue reading The post The Hidden Infrastructure Challenge Behind Every AI-Generated Avatar appeared first on SD Times .
Score: 40🌐 MovesJul 10, 2026https://sdtimes.com/ai-machine-learning/the-hidden-infrastructure-challenge-behind-every-ai-generated-avatar/ - Google just changed how it grades the AI models you use for Android coding
Google resets its Android Bench leaderboard with a new testing framework and eight new models.
- These Innovators Are Making Bold Moves—From AI Glasses To Next-Generation Biotech
These Innovators Are Making Bold Moves—From AI Glasses To Next-Generation Biotech USA Today
- Choosing the Right AI Agent Memory Strategy: A Decision-Tree Approach
In this article, you will learn how to choose the right memory strategy for an AI agent by working through a simple decision tree, one...
Score: 40🌐 MovesJul 10, 2026https://machinelearningmastery.com/choosing-the-right-ai-agent-memory-strategy-a-decision-tree-approach/ - Viva La Workflow: Optable’s Vlad Stesin On What’s Actually Changed In Agentic Advertising
A year ago, agentic advertising was mostly theoretical. AdExchanger Managing Editor Allison Schiff sat down with Vlad Stesin, CEO and Co-Founder of Optable, to talk about what’s actually changed — and which companies are doing real work rather than chasing the buzzword. Stesin breaks down why workflow automation is the unglamorous but essential starting point […] The post Viva La Workflow: Optable’s Vlad Stesin On What’s Actually Changed In Agentic Advertising appeared first on AdExchanger .
- Fresha Explores The Future of AI in Couture and Selfcare with Iris van Herpen Backstage at Paris Haute Couture Week
Fresha, the world’s leading AI-powered beauty and wellness booking platform, joined the hair team backstage at Paris Haute Couture Week this week to power Fresha Ambassador and internationally renowned hairstylist Hester Wernert-Rijn as she led the hair for Iris van Herpen’s Autumn/Winter 2026 Haute Couture collection, Sonic Starquakes. Following the runway presentation, Fresha secured an [...]
- Adwave Launches Wavemaker, a General-Purpose AI Video Generator Built on Its CTV Ad Technology
Adwave Launches Wavemaker, a General-Purpose AI Video Generator Built on Its CTV Ad Technology USA Today
- The future of public safety may start with a drone
Brinc Drones supplies 20% of U.S. SWAT teams in 50 states. NBC News' Gadi Schwartz spoke with the company's 25-year-old founder.
Score: 40🌐 MovesJul 10, 2026https://www.nbcnews.com/video/how-drones-could-become-the-future-of-911-response-266475077785 - Instagram’s Adam Mosseri: If you don’t like AI, ‘then you shouldn’t have it in your feed’
Mosseri still doesn’t want to filter out AI posts completely, however.
Score: 40🌐 MovesJul 10, 2026https://www.theverge.com/tech/963961/instagram-adam-mosseri-ai-feed-filters - One Rises, One Falls on Lock-Up Expiry Day: The Divergent Paths of Zhipu AI and MiniMax
Zhipu AI shares surged 13% while MiniMax plunged 18% on lock-up expiry, reflecting fundamentally different business models and market confidence in China's two leading AI companies.
- AI’s hidden revenue problem - and why the CIO owns it
AI’s hidden revenue problem - and why the CIO owns it Computing UK
Score: 39🌐 MovesJul 10, 2026https://www.computing.co.uk/opinion/2026/ai-s-hidden-revenue-problem-and-why-the-cio-owns-it - How AI changes the rules of engagement for sports viewers
Spectators will see more opportunity to move from pricey subscriptions and rigid schedules to more personalised feeds
- Samsung's Galaxy Unpacked Event: We Expect Weird Foldables, Funky AI Glasses and More
Samsung's foldable phone launch will take place later this month in London. We'll be there.
Score: 38🌐 MovesJul 10, 2026https://www.cnet.com/tech/mobile/samsungs-galaxy-unpacked-event-what-to-expect-how-to-watch/ - Claude isn't just a chatbot anymore — here are 9 things it can do in 2026
Claude isn't just a chatbot anymore — here are 9 things it can do in 2026 Tom's Guide
Score: 38🌐 MovesJul 10, 2026https://www.tomsguide.com/ai/claude-isnt-just-a-chatbot-anymore-here-are-9-things-it-can-do-in-2026 - Why it’s difficult to move past deployment to adoption
Why it’s difficult to move past deployment to adoption Health Data Management
Score: 38🌐 MovesJul 10, 2026https://www.healthdatamanagement.com/articles/why-its-difficult-to-move-past-deployment-to-adoption/ - Why Generic AI Translation Fails Enterprise Marketing
Explores why generic AI translation tools strip brand voice and how purpose-built systems enforce brand consistency at scale.
Score: 38🌐 MovesJul 10, 2026https://jasper.ai/blog/why-generic-ai-translation-fails-enterprise-marketing - Rosie DiManno: There’s no terminating the artificial intelligence takeover. I fear it will devour the world
Rosie DiManno: There’s no terminating the artificial intelligence takeover. I fear it will devour the world Toronto Star
- AI notetakers promise easy meeting recaps, but some professionals question their use
AI notetakers can quickly summarize meetings and create to-do lists, but they raise privacy concerns
- Intervou Launches: Professional Video Interviews That Rank on Google and AI Search
Intervou Launches: Professional Video Interviews That Rank on Google and AI Search USA Today
- How to Conduct a Competitive Analysis with Google Gemini
How to Conduct a Competitive Analysis with Google Gemini The Information
Score: 36🌐 MovesJul 10, 2026https://www.theinformation.com/articles/conduct-competitive-analysis-google-gemini - New Dashboard Tool Lets You Monitor Claude Usage
The generative AI vendor is aiming to counter dependence on and overuse of its popular model.
Score: 36🌐 MovesJul 10, 2026https://aibusiness.com/generative-ai/new-dashboard-tool-lets-you-monitor-claude-usage - Wipro PARI boosts hands-on robotics and automation learning for future manufacturing talent
Wipro PARI (Precision Automation and Robotics India), a part of Wipro Infrastructure Engineering, has donated and commissioned a industrial robot (FANUC R-2000iB/210F six-axis) to the Indo-Swiss Centre of Excellence (ISCE), Pune. The robot was formally inaugurated last week, July 4th 2026, and will provide nearly 200 students with hands-on training in industrial robotics and automation. […] The post Wipro PARI boosts hands-on robotics and automation learning for future manufacturing talent appeared first on CXOToday.com .
- The Role of Context in Human-Robot Interaction and Teleoperation in Unstructured Environments
The Role of Context in Human-Robot Interaction and Teleoperation in Unstructured Environments repository.cam.ac.uk
Score: 35🌐 MovesJul 10, 2026https://www.repository.cam.ac.uk/items/f306f73c-a645-49a6-be2b-58167d4a7cea - People ‘disdain’ AI, says The Odyssey director Christopher Nolan
People ‘disdain’ AI, says The Odyssey director Christopher Nolan The Straits Times
- From Rankings to Responses: AI Is Changing How Law Firms Get Found
In this week's episode, Legal Speak how the rapid rise of artificial intelligence is reshaping law firm marketing and public relations.
- iTester Launches iGVTS Framework to Support Verification of AI-Assisted Software Development
iTester Launches iGVTS Framework to Support Verification of AI-Assisted Software Development USA Today
- AI is stealing all the RAM and storage, and I’m learning to live with it
Hello again, and welcome back to Fast Company’s Plugged In . Recently, I dropped into my local Best Buy to pick up a hard drive. There was just one snag. I couldn’t find the storage department. Discombobulated, I circled the store until I spied two lonely disks sitting on an otherwise unoccupied expanse of shelving. Then it dawned on me: The storage section was still there, it was just nearly devoid of storage. I bought one of the drives, leaving a grand total of one in stock. Such is life as a tech user in the time of AI . The companies building massive data centers have such a voracious need for components—RAM, solid-state disks, hard drives like the one I was trying to buy—that they’ve drained the supply available for the consumer market. That has led to shortages such as the one I seemingly encountered. But an even more noticeable result has been the impact on the cost of devices. This phenomenon is known as RAMageddon (not to be confused with the SaaSpocalypse ), and there are too many examples to fit into one newsletter. Apple, for instance, started by warning that price hikes would be unavoidable . Then it announced a slew of them , including the MacBook Neo losing its defining $599 price. Microsoft’s Surface laptops all went up , too, sometimes by hundreds of dollars. Its Xbox consoles are on their second round of increases , Nintendo’s Switch 2 is going from $450 to $500 , and Sony’s PlayStation 5 is already up to $550 . Other manufacturers with lower profiles are also raising prices or skimping on configurations, including Dell and HP . Entire categories of gear, such as sub-$400 Android phones , are in trouble. There’s no reason to think the pain will subside anytime soon: If this fall’s iPhones aren’t significantly pricier, it will come as a pleasant shock. Memory and storage shortages are not unprecedented—did you know that a U.S.-Japan trade pact led to Nintendo delaying Zelda II because it couldn’t get enough memory chips?—but they’re rare. Thanks to Moore’s Law , we are used to the electronics in our lives getting both better and cheaper, inflation be damned . Anything else is disorienting, but here we are. All of this raises questions no AI can answer: Is there anything unethical about component manufacturers shifting production from the consumer market to data center build-outs? Probably not. Capitalism is capitalism. But the upshot could be a lot uglier than individuals and businesses having to shell out more money. Some companies in RAMageddon’s blast radius could be driven out of business . Should we be mad at tech giants for refusing to selflessly absorb the higher cost of components? After Apple raised prices, Vermont Senator Bernie Sanders accused CEO Tim Cook of “corporate greed.” But the company was already charging as much as it thought the market would bear, and had deferred increases longer than many of its competitors. Raising prices may hurt sales. It seems more the act of a company boxed into a corner than a ruthless money-grubber. Could all that RAM and storage being diverted to AI eventually power products we’ll happily use, thereby providing a deferred benefit? Perhaps. But it could also enable algorithms that fill our lives with machine-generated slop, invade our privacy, or put us out of work, so I’m not inclined to look on the bright side in advance. Then again, it’s also possible that AI companies are overbuilding data center capacity and will end up with computing cycles nobody wants, rendering this whole moment pointless. As I’ve mulled over the present crisis, I’ve realized that RAM and storage being so bountifully affordable is a semi-recent development. When I got interested in computers as a junior high school student, 16 KB of RAM was common and tolerable, 32 KB was more than adequate, and 48 KB felt downright sinful. Even a maxed-out microcomputer imposed discipline: Anyone writing software knew all sorts of clever techniques for getting more done in fewer bytes of code. A bit later, in the 1990s, multiple megabytes of RAM and hard drives with hundreds of megabytes to gigabytes of space were the norm—and yet almost nobody had as much as they really wanted. Utilities such as Stacker , which compressed disk files to effectively double a hard disk’s space, were must-haves. Computer users were so tight on resources that they were susceptible to being scammed: It turned out that a program called SoftRAM, which claimed to be the Stacker of memory, didn’t actually do anything . As far as I remember, computer users didn’t feel overly deprived in this era. Like someone living in a tiny Tokyo apartment, we just organized our (digital) possessions fastidiously and didn’t keep what we didn’t need. It may not have occurred to us that there would ever be any other option. I can’t pinpoint when getting more RAM and disk space stopped being a major decision. I just know that it did. My current MacBook Air has 24 GB of RAM because it was an impulsive upgrade, not because I knew I’d notice any performance improvement. Mid-1990s me considered a 500 MB hard drive—which I finally splurged on after thinking it over for months—to be thrillingly ginormous. If he knew that 2026 me would own a laptop with a 2 TB solid-state drive, packing 4,000 times as much storage, he’d be confused and possibly appalled. According to those in a position to know, AI-induced RAM and storage shortages may ease within a couple of years . If so, I may emerge unscathed. My iPhone, MacBook Air, and iPad Pro are nowhere near obsolescent. In March, when I was on Apple’s site looking for a desktop computer to run AI agents on, I even lucked into a refurbished Mac Mini—the base model that was hard to score at the time, and now no longer exists . If anything, RAMageddon has left me more grateful for what I already have. It’s fine and, better still, already paid for. Back in WWII days, when resources of all kinds were precious, there was a slogan: “Use it up, wear it out, make it do, or do without.” Even in times of plenty—which today’s devices still offer, albeit at less tempting prices—that’s advice to live by. You’ve been reading Plugged In , Fast Company ’s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on fastcompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I love hearing from you: Ping me at hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’m also on Bluesky , Mastodon , and Threads , and you can follow Plugged In on Flipboard.