LinkedIn Co-Founder Reid Hoffman: Entry-Level Jobs Down 16%, Why Every Career Is Now Entrepreneurship

Watch the Full Interview

Reid Hoffman, co-founder of LinkedIn and partner at Greylock, joins Peter Diamandis and the Moonshots team for a wide-ranging discussion on AI’s impact on jobs, education, and entrepreneurship. Hoffman reveals that entry-level jobs in AI-exposed fields have dropped 16% since ChatGPT’s 2022 launch, explains the paradox of software engineering (more engineers, higher salaries, fewer entry-level hires), and introduces his two new AI companies: Inflection (companion agents) and Manis (AI-accelerated cancer drug discovery). The conversation tackles education’s proficiency crisis, the compressed timeline of job transformation, and why entrepreneurial thinking has become mandatory for every career path.

Key Insights

• Entry-level jobs in AI-exposed fields dropped 16% since ChatGPT’s November 2022 launch, with early-career software positions in India down 20-25%
• Software engineering shows paradoxical growth: 150 million GitHub users (up 50%), salaries up 24% over 5 years, yet entry-level hiring compressed
• Computational thinking becomes essential literacy: using AI to write better prompts, deploying agent suites, thinking in automation workflows rather than manual tasks
• AI tutors represent best tutor in human history, enabling 5-10x faster learning than traditional classrooms by adapting to any topic student is passionate about
• Reid’s new company Manis (with Siddhartha Mukherjee) targets curing cancer through AI drug discovery factory, accelerating from discovery to clinical trials
• Entrepreneurial mindset mandatory for all careers: “Startup of You” philosophy applies whether founding companies or working traditional jobs in AI-transformed economy
• Job transformation timeline dramatically compressed compared to Industrial Revolution: companies cutting hiring months before AI deployment rather than years
• Education proficiency crisis (35% reading, 22% math) will be solved by AI assessment systems capable of PhD-level oral defense evaluation within years
• US political systems moving too slowly compared to AI acceleration, creating structural mismatch between technological change and policy response
• Inflection AI focuses on companion agents throughout life; Manis focuses on drug discovery; both represent Reid’s bet on transformative AI applications

Entry-Level Job Displacement: The 16% Drop

Eric Brynjolfsson’s recent research paper documented a 16% decline in entry-level jobs in AI-exposed fields since ChatGPT’s November 2022 launch. The data shows particularly sharp drops in marketing and sales positions, with the blue line representing early-career job hunters showing a clear inflection point coinciding with ChatGPT’s release.

Salim Ismail, just returned from India, reported even steeper declines: early-career software jobs down 20-25% with hordes of Indian engineers entering a compressed job market. The concern extends beyond economics into social stability - large populations of youthful individuals without meaningful career paths creates conditions similar to the Arab Spring.

Hoffman acknowledges the reality of job transformation and outright job loss in some cases. His heuristic: if a human being tries to do a job by following a script, AI can follow that script better. Customer service, basic software tasks, entry-level marketing - all vulnerable.

The timeline compression is unprecedented. Dave Blundin points out that during the Industrial Revolution, net job creation eventually exceeded displacement, but the transition took decades. Now employers like Salesforce.com cut hiring in anticipation of AI that will be deployed months in the future, not years. The acceleration creates disproportionate impact on new graduates who lack the buffer of existing employment.

Hoffman’s guidance matches Ismail’s: become entrepreneurs. Find problems that need solving and transform yourself. The forcing function will be positive, pushing people toward value creation rather than credential collection.

But the political consequences loom large. Blundin hopes that voters without jobs create acceleration in political response, forcing legislators to move faster on retraining, universal basic income, or other structural adaptations. The alternative is civil unrest from displaced workers.

The data is early and noisy. Hoffman cautions against overreading specific monthly or quarterly numbers. Technological change shows up in workflows and productivity before appearing in aggregate statistics. But the direction is clear: entry-level jobs as historically structured are transforming faster than institutions can adapt.

Software Engineering Paradox: More Engineers, Higher Pay

Peter Diamandis presents data that seems paradoxical on its surface: 150 million GitHub users as of May 2025 (up 50% since 2022), software engineering salaries up 24% over a 5-year period, yet entry-level hiring compressed.

Hoffman responds cautiously. These are very early days in how AI productivity plays out. He avoids getting distracted by monthly or quarterly numbers when the underlying technological theory matters more. The old line about computers applies: “Computers are everywhere except in the numbers.”

But he knows from direct experience that productivity has exploded. Work that used to take a couple hours now takes 10-15 minutes. Once you experience that productivity gain, you know it’s real regardless of what GDP figures show.

Most startups these days are completely AI-native in how they operate, finding great accelerations in development speed, time to market, and ability to iterate. The productivity gains are undeniable from a workflow perspective.

Alex Wissner-Gross argues this is the earliest innings of AI automating the service economy. Eric Brynjolfsson’s research showed results most striking in fields where AI automates rather than augments human labor. This is consistent with the hypothesis that humans and machines will merge symbiotically in the not-too-distant future.

The higher salaries suggest demand for software engineering thinking remains strong even as entry-level positions compress. Companies need people who understand computational approaches to problems, who can architect systems leveraging AI agents, who can prompt effectively and integrate multiple models.

The 150 million GitHub users represent a broadening of programming beyond traditional software engineers. More people are writing code, deploying agents, automating workflows. The skill becomes more widespread even as traditional entry-level jobs decline.

Hoffman believes there’s still nearly infinite hiring demand for software engineers, but the nature of software engineering work is transforming. The thinking about how to do software and software engineering becomes more widespread in problem solving because AI provides a software copilot for all of us.

Anything involving thinking, communication, or language will involve custom software we create through AI assistance. Software engineering becomes less about syntax and debugging, more about problem decomposition and system design.

Computational Thinking: The New Essential Skill

Hoffman describes how computational thinking changes his approach to every problem. When starting a new creative project, research question, business problem, or go-to-market strategy, he thinks in terms of prompts.

Most people submit simple paragraph prompts. Hoffman has evolved to meta-prompting: his first prompt asks AI to write the deep research prompt that will solve or target specific things. He writes or speaks a paragraph, AI returns a page and a half, he edits it, then submits that as the actual prompt.

This is computational thinking in action. You’re not just asking questions; you’re architecting how to ask questions. You’re leveraging the AI’s understanding of effective prompt structure to improve your own thinking process.

Diamandis notes that this represents a fundamental shift in how we approach problems. We’re training ourselves to think in terms of automation workflows, agent deployment, and system orchestration rather than individual manual tasks.

Hoffman goes further: we all deploy with a suite of agents now. When doing substantive work, he runs the same prompt across ChatGPT, Copilot, Gemini, and Claude, then integrates what comes back. He’s even set up an open-source model on his laptop to parse and distribute prompts to multiple services simultaneously.

This is the future of individual contributors in companies. Nobody works solo anymore; everyone operates with an agent suite. The question becomes how skillfully you deploy, integrate, and leverage that computational power.

The educational implications are profound. Traditional education teaches specific domains - math, literature, history. Computational thinking cuts across all domains. It’s the meta-skill of knowing how to leverage AI to accelerate learning, problem-solving, and creativity in any field.

Students who develop strong computational thinking can teach themselves anything. They can specialize in emerging fields before universities create curricula. They can adapt as job requirements shift.

Hoffman’s vision is that thinking about how to put problems in computational terms becomes as fundamental as literacy. Just as reading and writing transformed human capability in the post-Gutenberg world, computational thinking transforms capability in the post-GPT world.

Education Transformation: 5-10x Faster Learning

US education statistics are dismal. Only 35% of 12th graders are at or above proficiency in reading, down from 40% in 1992. Only 22% of seniors are proficient in math, with science at 31%. These are historic lows.

Peter Diamandis frames the dilemma: AI creates a double-edged sword. Students default to ChatGPT for answers rather than doing work, learning nothing. But AI will also become the best educator on the planet.

Hoffman responds with characteristic optimism. Within a small number of years, all assessment will be done by AI. We’ll have the equivalent of PhD oral-level defense available at every grade level. Students can be assessed on actual cognitive preparation rather than standardized tests.

The benchmark can be set however we like. People prepare for oral defense by AI that doesn’t accept insufficient answers. The interim problem of AI-done homework becomes irrelevant when assessment shifts to demonstrated understanding rather than completed assignments.

Second, even today you can meta-prompt AI agents: “Work me towards the answer, don’t give me the answer.” You immediately have the most amazing tutor that’s existed in human history, for free, globally accessible.

But Dave Blundin argues it’s so much more than a tutor. Tutors teach within a curriculum - math, science, language. AI tutors go wherever students are passionate. If you’re obsessed with skateboarding, AI teaches physics through skateboarding. If you love music, AI teaches math through music theory.

Diamandis cites statistics showing children with AI learn 2-6 times faster than sitting in classrooms. Someone at the recent Stanford AI conference corrected him: it’s actually 5-10 times faster.

Hoffman finds this easy to believe. AI personalization, infinite patience, adaptation to learning style, availability 24/7 - all combine to create unprecedented educational acceleration.

The frustration students experience becomes palpable. Once exposed to AI-accelerated learning, traditional school curriculum seems ridiculously narrow and slow. They’re passionate about specific topics, can learn them rapidly through AI, but schools force them through standardized progressions.

Ismail makes the critical observation: these statistics measure performance on existing curriculum. But the entry-level job of 2 years from now will be different from today’s entry-level job. Jobs transform along with education.

The regulatory structure of education means schools will transform slower than jobs, creating a dangerous mismatch. But the forcing function is powerful. When students learn 5-10x faster with AI, the impedance mismatch breaks the existing system.

We’ve waited decades for a forcing function to transform education. AI acceleration of student learning may finally be it.

The Entrepreneurial Imperative

Hoffman published “The Startup of You” over a decade ago based on a commencement speech to his high school. The thesis: we all have to think more and more like entrepreneurs even if we’re not founding companies. It’s the nature of the world we’re evolving into.

He creates content in two main streams: technology and society, and entrepreneurship. The entrepreneurship content targets not just high-growth Silicon Valley founders, but everyone navigating careers and jobs. The lessons of entrepreneurship apply universally.

Ismail emphasizes: the career of the future is entrepreneurship. Not going through a factory process of getting a job for someone else. It’s how do you use these tools to create value in the world.

This represents a fundamental shift in career orientation. Industrial-era education prepared workers for specific roles within existing companies. Find a job, learn the skills, advance through defined career ladders.

AI obliterates that model. When jobs transform every few years, there are no stable career ladders. When entry-level positions get automated, there’s no bottom rung to grab. When companies can’t predict their staffing needs 2 years ahead, they can’t offer traditional career paths.

The entrepreneurial mindset becomes mandatory: identify problems, create value, adapt continuously, learn voraciously, leverage tools maximally. Whether you work for yourself or for a company, you operate as an entrepreneur of your own career.

Hoffman acknowledges the class dimensions. Growing up in London, he recognized class as a primary driver of structural disadvantage. Sitting around dinner tables with parents, uncles, aunts, and friends who are journalists, academics, and doctors, children absorb cultural knowledge about how the world works.

They gain affirmation about confidence, self-respect, how to talk to adults, how to get feedback. That community isn’t available to everyone. It’s a huge structural advantage.

AI companions and tutors democratize access to that cultural knowledge. Everyone can have patient, knowledgeable guidance on how to navigate the world, how to think about careers, how to position themselves for opportunities.

The educational aspect goes beyond facts and skills to include confidence, judgment, and strategic thinking. AI provides even-handed, patient, kind, respectful, data-driven interaction at huge scale.

Hundreds of millions of people daily receive this kind of support. Hoffman finds that super inspiring. It’s what drives him to build these systems.

The transition won’t be easy. Industrial Revolution eventually created more jobs than it destroyed, but the timeline was decades. AI transformation happens in years, possibly months. The forcing function is powerful but painful.

Hoffman’s optimism isn’t naive. He acknowledges transition issues, job loss, structural challenges. But he believes we’ll adapt. We always have. The tools to create value have never been more powerful or more accessible.

Key Quotes