AI vs. 人类：大语言模型 vs. 程序员——未来展望

AI vs. Human: LLMs vs. Coders— the future The "LLMs vs coders" debate has become one of the defining questions of the AI moment.Are white-collar jobs about to disappear? Are recent tech layoffs AI-driven? To test that,we went on the ground and spent 50 hours speaking to tech employees globally. Ourconversations help debunk some myths and distill a few predictions about how AI willreshape work and tech services. Venugopal Garre+65 6326 7643venugopal.garre@bernsteinsg.com Nikhil Arela+91 226 842 1482nikhil.arela@bernsteinsg.com Our previous report on AI vs human was on the debate on how AI stacks up against peoplein the knowledge industry. With rapid advances in model capability – from larger contextwindows to richer training, AI has shifted to a current reality. Coding assistants are now livein many firms, and productivity copilots are being embedded across workflows. Bold claimsthat white-collar jobs could be fully replaced are no longer framed as “someday”, and that isshowing up in sharp draw-downs in some stocks. But what about the people in the middleof this change, especially in tech? High-level narratives and projections often miss theground reality. This note is about that reality: how AI is being used, how roles are evolving,and what has (and hasn’t) changed, based on conversations with over 30 professionals. AI is not yet driving job losses:Tech companies across the world, which should be theearliest adopters, are using AI heavily for internal agents and code support, and seniormanagement is actively pushing them to increase usage. Yet we did not see clear evidenceof system-level efficiency gains translating into team downsizing or frozen hiring plans;this was a consistent observation across firms. In other words, the idea that AI is drivingthe recent wave of job cuts looks more like a convenient narrative for companies that overhired, rather than something grounded in current on-the-ground reality. Blurring boundaries and changing dynamics: One clear theme was the blurring of roleboundaries in firms that are early and heavy adopters of AI. Tech leads are going back to“getting their hands dirty” and writing more code. Product managers are spending moretime on UX and competitive strategy. Junior engineers are ramping up far faster than beforeand closing the gap with seniors. In most cases, rapidly ramping junior engineers are seenas an asset, and an engineer+ AI pairing offers the best return on investment. Team sizesare likely to shrink over time, but not by simply wiping out only the bottom layer. Seven predictions: 1) White collar workforce will not get eliminated but repurposed,2) Junior developers stay, ramp faster, and reach “senior” productivity sooner, even asoverall team sizes gradually shrink, 3)AI adoption races ahead in non-critical areas butremains deliberately slow in core processes and other areas such as storage, payments,and security, keeping incumbents’ moats intact for longer, 4) Pure Agentic AI startups builton third-party LLMs will struggle to scale as in-house tools, LLMs, and big customizersdominate the agent layer. 5)Startups gain early speed from AI-written greenfield code butbuild up technical debt, just as large firms slowly modernise and eventually catch up. 6) ITservices could become unlikely beneficiaries as heavy greenfield AI work slows and theneed for integration, maintenance and governance grows. 7)Pricing shifts from FTEs totokens and outcomes, with IT services firms running AI like a managed utility, optimisingusage, spend, and reliability for clients. DETAILS OUR INFERENCE - WHERE’S THE TECH INDUSTRY HEADED? Our conversations with people who are actually implementing AI on the ground cut through a lot of the hype. Some popularbeliefs are simple bandwagon exaggerations; others survive mainly because they fit a neat, dramatic narrative. Also, it is to benoted that most of our conversations with employees and what they see currently - as well as envision, is usually limited to atime frame of 3-4 years. 10 years horizon is very large from technological point of view. But there, even predictions of sweepingchanges are equally likely to be false. On top of that, geopolitical tussles, changing regulations, and ESG actions are impossibleto predict over this long a horizon. We’ll broadly highlight what we see happening in the near to medium term as a result ofwhere the technology is today, and its adoption. First, we see surprising resilience in human capital. Junior engineers are not disappearing; if anything, their scope hasexpanded. Ramp-up times have fallen sharply, and freshers are taking on more ownership earlier in their careers. We stillexpect the overall headcount to shrink over time, but that reduction is likely to be broadly spread across levels rather thanfalling disproportionately on entry-level roles. At the same time, role boundaries are blurring as responsibilities get mixed andreconfigured. Second, Agentic AI looks like a dangerously hot trend.