How ChatGPT Is Changing US Tech Careers

A viral new blog post, "Something Big is Happening", has attracted 80m views arguing that A.I. has automated coders out of the technical aspect of their job and that nearly all jobs are next. What, however, do the data show? THE EMPLOYMENT PICTURE • Since ChatGPT launched in late 2022, the U.S. has *added* ~3 million white-collar jobs while blue-collar employment has stayed flat. • America has 7% more software developers, 10% more radiologists and 21% more paralegals since ChatGPT's launch (these are roles regularly cast as A.I.'s earliest victims). • Real wages in professional and business services are up ~5%; office and admin workers' real wages are up 9%. THE HISTORICAL PATTERN • In 1982, Nobel laureate Wassily Leontief warned computers would displace mental labor en masse. What happened? White-collar employment more than doubled and pay rose ~33% in real terms. • Technology rarely replaces entire jobs. Instead, it automates specific tasks within them. The historical result is upgrading, not replacement. • MIT research found roughly half of U.S. employment growth from 1980–2007 came from brand-new job titles created by technological change. WHERE THE VULNERABILITIES ARE • Entry-level roles are most exposed... they involve narrower "task bundles" with fewer edge cases requiring human discretion. • Routine back-office work is actually shrinking (see chart from The Economist at the top of this post): insurance-claims clerks down 13%, secretaries and admin assistants down 20%. • But roles combining technical expertise with oversight and coordination are booming, e.g., project managers and infosec experts are up ~30%. THE AI REALITY CHECK • Anthropic's own data show only ~4% of occupations use A.I. across 75%+ of their tasks. Hardly any roles can be fully automated. • Today's A.I. has "jagged intelligence": impressive on many tasks but uneven. Being good at 95% of a task isn't enough when the remaining 5% involves critical edge cases. WHAT CAN YOU DO? 1. Don't panic out of your technical career. Roles combining technical depth with judgment and coordination are growing, not shrinking. 2. Become the person who works *with* A.I. (the future is increasing augmentation). 3. Invest in the hard-to-automate skills: judgment, stakeholder communication and messy real-world domain expertise. 4. Stay curious. The durable advantage isn't mastering any single tool, it's getting comfortable with the pace of change itself. Listen to today's episode of my podcast (Episode #968) to hear more on all of the above! The "Super Data Science Podcast with Jon Krohn" is available on all major podcasting platforms and YouTube. See below for quick access ⬇️ Many of the data above come from an article in The Economist that I've also got for you below, alongside a link to the viral "Something Big is Happening" post. #superdatascience #automation #jobautomation #ai #futureofwork

Since ChatGPT hit the mainstream, entry-level jobs in the UK have dropped nearly 32%.   Think about that.   Not just internships or grad schemes, but apprenticeships, junior roles… gone.   And what’s crazy is, we’re not talking about this: Vacancies are up. Salaries are up. But if you’re just starting out, there’s nowhere to start.   Let’s call it what it is: We’re using AI to do more but also to train less. And in doing so, we’re erasing the space where people used to learn, grow, and become the next generation of talent.   If you're not hiring juniors, how exactly are you building your future team? Well, you’re not.   You’re renting expertise and hoping it sticks around.   I’m not anti-tech. I build with AI every day. But let’s not pretend automation is neutral. It reflects our values. And right now, it’s telling us that efficiency matters more than development.   So what do we do? → Stop acting like junior roles are too much effort → Use AI to scale support, not scrap it → Create jobs where people can actually learn, not just execute   Because in five years, we’re going to need mid-level talent who get it. And if we don’t create the space for them now, we’ll only have ourselves to blame.

A new study released today by OpenAI and Harvard economists draws on anonymized data from over 700 million weekly ChatGPT users worldwide. It offers the first large-scale, privacy-preserving look at how people actually rely on generative AI for sophisticated reasoning and decision support. Five findings leap out at me: ⭐ Decision support is exploding. Almost half of all messages, and now more than half, are people asking for guidance, advice, or analysis. The real economic value lies here: AI as a thinking partner. ⭐ Workplace reasoning is front and center. Among work-related messages, 56% involve “doing” tasks, and nearly three-quarters of those are writing tasks where the model is helping to solve problems or craft strategy, not just generate boilerplate. ⭐ These tasks match the core of knowledge work. Over 45% of all messages map to O*NET work activities such as “Getting Information,” “Interpreting Information,” and “Making Decisions & Solving Problems.” ⭐ Quality rises with complexity. Interactions in which people ask the model to reason or advise consistently rank highest in user satisfaction. ⭐ AI is becoming a teacher. Roughly 10% of all messages are tutoring or teaching requests, a striking signal that people already trust AI to explain and guide. And for those driving enterprise transformation, the same research adds a powerful call to action: ⚡ ChatGPT adoption has reached 10% of the world’s adult population, with users sending 2.5 billion messages daily, one of the fastest technology diffusions in history. ⚡ Even as personal use grows, absolute work-related usage has more than tripled in a year, proving that employees already incorporate AI into their daily jobs, often before formal corporate programs. ⚡ The highest-value interactions, decision support, strategic writing, and problem-solving are precisely the activities that define knowledge-intensive industries. For enterprises and their advisors, this is more than a trend; it’s an urgent signal. The next competitive edge isn’t just automating routine tasks. It’s embedding AI as a true co-pilot for human judgment, from strategic planning and R&D to regulated decision environments. If you’re shaping an AI strategy today, these data points make the case clear: your teams and your customers are already treating AI as a reasoning partner. The question isn’t whether they will... It’s whether your enterprise is ready to design for it and become truly AI-first. Read the paper here: http://bit.ly/4na2eeA

Nearly 80% of all ChatGPT conversations fall into just three categories: Practical Guidance (29%), Seeking Information (24%), and Writing (24%). What stands out: - Writing dominates work-related use, accounting for ~40% of all work-related usage in June 2025. - Most writing requests (~⅔) aren’t about creating a document from scratch, but rather improving existing text through editing, critiquing, or translating. - Education is another major use case: 10% of all messages, and more than a third of practical guidance, are focused on tutoring or teaching. One thing we learned from this massive dataset is that ChatGPT’s value is in producing digital outputs and also in improving human judgment. It acts as a decision-support system, helping people weigh options, choose better words, and interpret information. In knowledge work, where productivity depends on the quality of decisions, that’s a powerful shift. That means the future of AI at work may more about raising the ceiling on human judgment than we previously thought.

AI just killed the cover letter, and it’s quietly changing the way we apply and get hired. New research out of Dartmouth and Princeton shows that before ChatGPT, employers actually valued customized applications. A well-written cover letter or tailored proposal acted as a costly signal — proof that you’d done the work, cared about the role, and likely had the skills to back it up. But now? Generative AI has made writing effortless and cheap. The study found that after AI tools were introduced, employers’ willingness to pay for customized applications dropped by more than two-thirds. Even more striking — those polished, AI-written submissions no longer predicted who would actually perform well on the job. In other words, writing has lost its credibility as a signal of ability. So what does that mean for you? It’s time to adapt. 1️⃣ Find a new way to stand out. Skip the generic upload and instead send a short, authentic message or DM to the hiring manager. Real communication cuts through the AI noise. 2️⃣ Keep your LinkedIn alive and accurate. Your digital footprint is now your first impression. Make sure it reflects real projects, measurable impact, and not just AI-generated fluff. 3️⃣ Invest in in-person connection. As AI floods digital spaces with sameness, human networks will only grow in value. Go to events, meet people, have conversations — the new “signal” is trust built offline. Read the full paper of Making Talk Cheap: Generative AI and Labor Market Signaling Here: https://lnkd.in/g4WPTjjU and follow for more AI research breakdowns.

New research from the Federal Reserve Bank of Dallas just found that wages are rising in AI-exposed jobs - but young people can't get hired into them. Here's why this matters for every company thinking about how hiring has to change, right now: WAGES ARE UP, HIRING IS DOWN Since ChatGPT launched three years ago, employment in the most AI-exposed sectors has actually declined, but wages in those same sectors are outpacing the national average. Like - computer systems design wages are up 16.7% vs. 7.5% nationally. How is that possible? THE CODIFIED VS. TACIT KNOWLEDGE GAP Okay, here's the big takeaway: AI is great at replicating "codified knowledge" - the stuff you learn in textbooks and classrooms, right? But it can't yet touch "tacit knowledge" - the understanding you only get through years of experience. Think about what that means: For a senior marketing strategist, for example, AI automates the routine parts of their job, gives them more time to focus on value add stuff, and makes their hard-won instincts MORE valuable. Their wages go up. For a new graduate whose main asset *is* that textbook knowledge? AI is automating their entire value proposition. THE ENTRY-LEVEL LADDER IS BREAKING This is the part that should worry every leader reading this. The way white-collar careers have always worked is: you start at the bottom doing basic tasks, and you slowly build up that tacit knowledge until you become the experienced person. AI is making that first rung economically hard to justify. Not because companies are laying off junior employees - but because they're not hiring them in the first place. The job finding rate for young workers in AI-exposed fields is falling fast. So we have a real problem. Companies need experienced workers. Experienced workers are more valuable than ever. But the pipeline that *creates* experienced workers is drying up. WHY THIS MATTERS RIGHT NOW Every company needs to be thinking about two things: How do we redesign entry-level roles so junior employees are building tacit knowledge faster, not just doing tasks AI can handle? How do we make sure our experienced people are actually using AI to amplify their expertise? The Dallas Fed said it - leaving new employees off the job ladder isn't sustainable. But right now, that's exactly what's happening. THE SOLUTION: RETHINK HIRING This is one of the fundamental tenants I do with large organizations - rethinking hiring. Your tenured, experienced people need to figure out AI, fast. Your young people need experience, fast. There can be a match here. But first, HR leaders, please pull all Job Descriptions - like right now. We need to examine them for what they are hiring for. What skills? What responsibilities? I'm going to be talking more about this on this platform - but if your organization needs help right now, this is what we do at AI Mindset. DM me or visit our website.

I recently met an engineer at a legacy tech company. He’s worked at the same place for nearly a decade. During his first few years, he would conduct two or three interviews a week, meeting potential hires for the engineering team. Over the last year, however, he said those calls have “fallen off a cliff.” He now does just one or two interviews per quarter. This chart from Andrew Van Dam shows there are now fewer programmers in the US today than at any other point since 1980 — and that’s across a four-decade stretch when the total workforce has nearly doubled in size. In short, the on-ramp to a career in software engineering has collapsed. Generative AI automates much of what used to be done by entry-level programmers. The boilerplate tasks that were once relegated to junior developers can now be handled by large language models. If ChatGPT is taking on an increasing share of early work, the pseudo-apprenticeship model breaks down. It's true that technology has always displaced jobs (lamplighters were undoubtedly upset at the advent of the lightbulb). Though what’s different this time is that, for many, the floor is falling out before their careers even start. To me, the danger here isn’t just declining employment for new grads, but a hollowing out of the talent pipeline entirely. No novice engineers today means no seasoned veterans tomorrow.

Just after Fidji Simo started as CEO of Applications, the company unveiled plans for an OpenAI Jobs Platform, which promises to “expand economic opportunity with AI.” This is being viewed reflexively through the lens of ChatGPT vs. LinkedIn. This is too narrow. There's a bigger idea behind this. Remember, according to OpenAI's internal strategy memo: "ChatGPT's mission is to introduce the whole world to an intuitive AI super assistant that deeply understands you and is your interface to the internet." The jobs platform serves as both proof of concept and a Trojan horse for this broader vision. It represents OpenAI's first serious attempt at what I refer to as "workflow orchestration," building AI-native applications that coordinate complex workflows across multiple domains, displacing traditional software through superior outcomes delivered via conversational interfaces. This attempts to go beyond adding #AI features to existing workflows. Instead, it would fundamentally restructure how users conceptualize and execute complex activities. This leads users to develop different mental models when accomplishing tasks through AI orchestration. Instead of thinking, "I need to update my resume, search job boards, and track applications," users begin framing objectives as, "find me roles that match my skills and career goals." This cognitive shift creates what behavioral economists term "mental model dependency." As these applications rewire the habits and brains of users, OpenAI accumulates a comprehensive understanding across multiple domains of life and work. Returning to traditional job platforms could feel inefficient once users internalize this new paradigm. Think back to the early days of search. While search engines existed in the late 1990s, users were just as likely to browse portals of links, such as Yahoo, to find the content they wanted. Google changed that by making a vastly superior search engine. Query by query, users learned to “Google” for whatever they wanted to know. This became the default way most of us interacted with information online. OpenAI wants to do the same, but for ChatGPT. For any tasks that need to be done, your first instinct should be to open ChatGPT and write a natural language query. In that framing, job hunting is merely a concrete starting point, a tangible task that allows OpenAI to learn how to deliver value to all parties and to start shifting those habits. By the same measure, the company is striving to promote the broader adoption and training of artificial intelligence for all types of businesses and workers. The hope here is likely to demystify AI and, in the process, allow everyone to experience its benefits firsthand, to the point where it becomes an integral part of everyday life and the economy. This goes beyond short-term monetization. OpenAI is attempting to lay the foundation for its long term, durable growth #moat by defining the way we interact with digital experiences. #Discontinuity

This cable tech's work is art. Every wire perfectly placed. Every connection labeled. Zero chaos. Companies pay premium rates for techs like this. Not because they can run cables. Anyone can do that. They pay because this level of organization saves thousands in troubleshooting time. Because when something breaks at 3am, you can trace the problem in minutes instead of hours. The same thing is happening with AI prompting right now. Every IT professional can use ChatGPT. Type a question, get an answer. Basic stuff. But the ones who master prompting? They're becoming irreplaceable. They're the ones who can extract exactly what they need from AI in one shot. Who can automate their documentation. Who can debug faster, learn new systems quicker, and solve problems that used to take days. While everyone else is still typing "how do I fix this error" into ChatGPT. The gap between basic users and prompt masters is growing fast. And it's starting to show up in who gets promoted, who gets hired, and who gets left behind. You don't need to become an AI engineer. You just need to get really good at working with AI. Because in 2025, "IT professional" means you can leverage AI as well as you can troubleshoot systems. The ones who figure this out early? They're the premium hires. Just like that cable tech.

Nearly 1 in 4 tech jobs now require AI skills. This isn't just another tech trend—it's transforming how we build teams across every industry. Our AI ALPI research last week uncovered a transformative shift in the talent landscape: → 68% growth in AI job listings since ChatGPT's release while overall tech listings declined 27% ↳ Signals critical reshaping of talent priorities even during hiring slowdowns → Information sector leads with 36% of tech jobs requiring AI skills ↳ But the real story? Healthcare, retail, and transportation seeing their AI job share DOUBLE since 2022 → Companies aren't creating new AI-only roles ↳ They're integrating AI competencies into existing positions across functions HR implications are profound. We're witnessing the most significant workforce transformation since the original digital revolution—one where talent strategies must evolve beyond traditional "skills acquisition" to continuous capability building. The average ramp time for new HR tech adoption is 9 months, but organizations with structured AI literacy programs are reducing this to just 3 months while achieving 2.7x higher ROI on their HR technology investments. 🔥 Want more breakdowns like this? Follow along for insights on: → Getting started with AI in HR teams → Scaling AI adoption across HR functions → Building AI competency in HR departments → Taking HR AI platforms to enterprise market → Developing HR AI products that solve real problems