https://www.theregister.com/2026/03/24/ai_job_unbundling/ My Account The Register Home Page [the_regist] [the_regist] Search [magnifying] [close] Search The Register [ ] [magnifying] Navigation [burger_men] [close] Topics Security * All Security * Cyber-crime * Patches * Research * CSO Off-Prem * All Off-Prem * Edge + IoT * Channel * PaaS + IaaS * SaaS On-Prem * All On-Prem * Systems * Storage * Networks * HPC * Personal Tech * Cx0 * Public Sector Software * All Software * AI + ML * Applications * Databases * DevOps * OSes * Virtualization Offbeat * All Offbeat * Columnists * Science * Geek's Guide * BOFH * Legal * Bootnotes * Site News * About Us More from The Register Special Features * All Special Features * RSA * Agentic AI * The Future of the Datacenter * AWS:Reinvent * SC25 * Supercomputing Month * Datacenter Networking Nexus * The State of Storage Vendor Voice * All Vendor Voice * Everpure * Rubrik * Make it real with Capgemini and AWS * Money Movement Hub * ZTE * Scale Kubernetes. Not Chaos. * AWS New Horizon Resources * Whitepapers * Webinars & Events * Newsletters [aiml] AI + ML 17 comment bubble on white AI isn't killing jobs, it's 'unbundling' them into lower-paid chunks 17 comment bubble on white Paper argues the real impact isn't job loss but narrowing human work and pay icon Carly Page Tue 24 Mar 2026 // 18:32 UTC # AI isn't killing jobs wholesale - it's quietly chipping away at them, one task at a time. That's the gist of a new research paper making the rounds, which pushes back on the idea that more AI exposure automatically means fewer jobs. The authors argue the real question isn't how many tasks a model can do, but whether those tasks can actually be split out without breaking the role. New job no job Microsoft execs worry AI will eat entry level coding jobs READ MORE Analysts have long warned that automation could wipe out millions of jobs. One recent forecast put the number at 10.4 million US jobs gone by 2030, roughly 6 percent of the workforce. The implicit assumption behind those numbers is straightforward: if AI can do enough of what you do, you're toast. This new paper - written by Luis Garicano, professor at the London School of Economics, along with Jin Li and Yanhui Wu, both at the University of Hong Kong - suggests it's not that simple. Jobs, it argues, aren't neat lists of tasks - they're bundles. Radiologists, for example, don't just read scans. They interpret edge cases, talk to clinicians, and sign off on decisions people act on. Replace the image-reading bit, and you haven't necessarily replaced the job. [aiml] That's where the authors draw the line between what they call "weak bundles" and "strong bundles." Weak ones can be split apart without much fuss, but strong ones can't without losing value. [aiml] [aiml] "In weak-bundle occupations, AI automates some tasks and narrows the boundary of the job... In strong-bundle occupations... AI improves performance inside the job, but does not remove the human from the bundle," the authors argue. In weak-bundle jobs - think churning through support tickets or knocking out predictable bits of code - AI doesn't just replace a task; it reshapes the job. The human is left doing whatever the machine can't, often a narrower slice of the original role. * AI still doesn't work very well, businesses are faking it, and a reckoning is coming * Supposedly big-brained execs are outsourcing decisionmaking to AI * Jack Dorsey's fintech outfit Block announces 40% layoffs, blames AI, gets 23% stock bump * Altman: You think AI is wasted energy? Try raising 100 billion humans Sounds like a win on paper. In reality, not so much. Once AI takes over part of the work, the human stops dividing their time. They go all-in on what remains, which means output per worker jumps, prices fall, and suddenly you don't need as many workers as before. [aiml] In other words, the hit to employment doesn't come from AI doing the job outright, but from humans becoming too efficient at the leftovers. It also squares with what we're seeing so far. AI is reshaping jobs, not wiping them out. Tasks move around, productivity may go up, yet employment and hours haven't shifted much - at least yet. In many cases, the bundle is still holding. It also explains why the doom predictions and the techno-optimism can both be right at the same time. If you're in a strong-bundle job - something heavy on judgment, context, or responsibility - AI is more likely to make you faster and better paid. If you're in a weak one, it may quietly hollow out your role until there's not much left to defend. (r) Get our Tech Resources # Share More about * AI * Job cuts More like these x More about * AI * Job cuts Narrower topics * AIOps * DeepSeek * Gemini * Google AI * GPT-3 * GPT-4 * Large Language Model * Machine Learning * MCubed * Neural Networks * NLP * Retrieval Augmented Generation * Star Wars * Tensor Processing Unit * TOPS Broader topics * Career * Employment * Self-driving Car More about # Share 17 comment bubble on white COMMENTS More about * AI * Job cuts More like these x More about * AI * Job cuts Narrower topics * AIOps * DeepSeek * Gemini * Google AI * GPT-3 * GPT-4 * Large Language Model * Machine Learning * MCubed * Neural Networks * NLP * Retrieval Augmented Generation * Star Wars * Tensor Processing Unit * TOPS Broader topics * Career * Employment * Self-driving Car TIP US OFF Send us news --------------------------------------------------------------------- Other stories you might like AI will write code, but prepare to babysit it - and be sure you speak its language kettle This week on the Kettle, we predict that AI software development won't make you want to fire your devs anytime soon AI + ML29 Mar 2026 | The first thing vibe coding builds is confidence it will help you succeed Secret CEO And developers should be confident it won't kill the craft AI + ML29 Mar 2026 | 12 Bees and hummingbirds aren't just buzzing - they're sipping trace booze Alcohol turns up in most floral nectar, meaning pollinators are drinking tiny cocktails without ever getting drunk Science29 Mar 2026 | 9 Resilient, continuously active data - with no compromise When the gap between data generation and action is a strategic liability, it's time for a fix Sponsored Feature [aiml] Anthropic struggling with Chinese competition, its own safety obsession The maker of Claude faces headwinds as it rushes to go public AI + ML28 Mar 2026 | 21 To BSOD or not to BSOD? Only Microsoft knows the answer Bork!Bork!Bork! Famous blue screens remind conference of security pros that this OS sometimes has bad days Bootnotes28 Mar 2026 | 15 Microsoft takes up residence next to OpenAI, Oracle at Crusoe's 900 MW Texas datacenter expansion New campus to include on-site power generation On-Prem27 Mar 2026 | 11 Folk are getting dangerously attached to AI that always tells them they're right Sycophantic bots coach users into selfish, antisocial behavior, say researchers, and they love it AI + ML27 Mar 2026 | 64 Apple's last tower topples... and the others will follow Farewell, Mac Pro: Increasing integration means the end of expandable computers Personal Tech27 Mar 2026 | 109 Senators want datacenters to come clean on power consumption Ratepayer Protection Pledge is unenforceable without hard numbers, Warren and Hawley argue On-Prem27 Mar 2026 | 9 Microsoft tells crusty old kernel drivers to get with the Windows Hardware Compatibility Program Cross-signed code gets the cold shoulder as Redmond tightens trust OSes27 Mar 2026 | 43 Commercial space pleads with NASA to stop moving the goalposts in orbit Private station hopefuls say ISS rethink is shaking confidence Science27 Mar 2026 | 41 The Register Home Page [the_regist] Biting the hand that feeds IT About Us * Contact us * Advertise with us * Who we are Our Websites * The Next Platform * DevClass * Blocks and Files Your Privacy * Cookies Policy * Privacy Policy * Ts & Cs Situation Publishing [situation_] [situation_] Copyright. All rights reserved (c) 1998-2026