---
title: "AI doesn't \"take\" jobs. It exposes them."
url: https://stacklist.com/card/3cca1517-46fd-4bc1-a332-9360ec5d4aea
source_url: "https://metacircuits.substack.com/p/ai-doesnt-take-jobs-it-exposes-them"
stack: https://stacklist.com/stack/f3725201-e51d-4977-b61a-61e523b68980
summary: "AI doesn't eliminate jobs but exposes underperforming roles by automating low-complexity knowledge work that humans struggle to justify. The article argues that corporate value creation follows a power-law distribution where top performers deliver outsized gains, and AI will widen the gap between high and low performers."
tags: "ai-impact, workforce, productivity, power-law, knowledge-work, organizational-performance"
key_entities: "Jonas Braadbaart (person), Max Schoening (person), Anthropic (organization), OpenAI (organization), Notion (organization), Asana (organization), Microsoft (organization), Gallup (organization), Claude (technology), power-law distribution (concept), productivity theater (concept), Pareto principle (concept), State of the Global Workplace 2026 (event)"
classification: "analysis"
content_hash: "sha256:bd7151211ecd99989f37f3d79c866f3a939b128bcb8a29f657183f5bf76ad0a8"
acp_version: "0.2"
token_counts_approximate: 3674
visibility: public
agent_accessible: true
status: "final"
---

# AI doesn't "take" jobs. It exposes them.

AI doesn't "take" jobs. It exposes them. How much value are you actually adding? Jonas Braadbaart May 18, 2026 24 20 6 Share If you’re worried about AI taking your job, how much value are you adding really? Ever since I’ve started working as a consultant for large corporate clients — now more than 10 years ago — I’ve wondered if some kind of Pareto principle holds for value creation: 20% of an organization producing 80% of the value. Because if that’s the case, the 80% not pulling their weight are in trouble. Low-complexity knowledge work is exactly the kind of work AI already excels at. And it will do that work better, more systematic, and cheaper than any human ever could. Since I didn’t know the answer — it was based on a hunch — I decided to look into the 80/20 question. In what follows, I’ll give my take on the AI-will-take-your-job debate, and look at what kind of organizations are best suited to win in the age of AI. Coincidentally, Anthropic released Claude for Small Business end of last week, and on May 4th both OpenAI and Anthropic announced joint ventures aimed at helping large scale enterprises (LSEs) deploy AI systems and AI agents as part of their workforce. TLDR: corporate value creation is power law-distributed across your workforce. AI doesn’t take jobs, it exposes them. Large enterprises are cutting headcount while AI-using SMBs are net-hiring, and the only durable lever is building your own agents instead of renting them. Productivity theater Because the economic output of knowledge work is so hard to measure, an endemic problem inside large corporations is performative work. Employees optimizing for visibility and career progression instead of impact. “We already have universal basic income. It’s called knowledge work.” — Max Schoening, Head of Product at Notion, on Lenny’s podcast The data behind the quote: 80% of the global workforce is “disengaged” ( Gallup State of the Global Workplace 2026 ). In Europe it’s 88% — the lowest of any region in the world. In addition, manager engagement collapsed from 27% to 22% in a single year. 60% of knowledge-worker time is “work about work” — coordination, status updates, alignment meetings, chasing approvals ( Asana , 2023). Microsoft called it “productivity theater” back in 2024 — meetings attended by people who barely speak, emails sent to signal responsiveness, the green dot kept lit. Today it’s powered by a lively Shadow AI practice. The best vs the rest Turns out the 80/20 distribution I intuited 10 years ago wasn’t far off from reality. Research and real-world performance in knowledge work is power-law distributed . A small group of elite performers are delivering their employers outsized gains. In B2B sales, the top 20% of reps have generated 60–80% of revenue for half a century. Outcomes are definitely, emphatically, not normally distributed across the workforce — even though that’s exactly how HR has framed it for the last eighty years. The canonical paper is O’Boyle & Aguinis (2012) in Personnel Psychology (N = 633,263). In the pre-AI era, performance distributions in knowledge work were already heavily right-skewed and best fit by power-law / Pareto-like models. The exact ratio varies by occupation — some 80/20, some 90/10, some sharper — but the directional claim is one of the most robust results in organizational research. I expect the gap between low performers and high performers to widen as AI systems get better. While there is 2023 research that shows poor performers benefit more from using AI tools (Brynjolfsson et al. 2023), this research takes as a base case that everyone has access to the same AI agents. I don’t think this will be the case going forward: top performers will customize their computer and company agents — AI literacy is going to be a make-or-break skill in many professions. In addition, two decades earlier, Hunter, Schmidt & Judiesch (1990) had shown that the top-vs-average performance ratio depends on job complexity: about 2× in low-complexity work, 8× or more in high-complexity work. AI is rapidly absorbing low-complexity knowledge work — drafts, summaries, scheduling, invoicing, formatting, status reports. The work where the performance differences were least extreme is being pulled out of the human labor pool. 2026 enterprise data echoes these organizational research findings. Writer’s April 2026 AI Adoption Survey (n=2,400) finds 92% of the C-suite are actively cultivating a new class of “AI elite” employees — workers about 3× more likely to have earned both a promotion and a pay raise in 2025. Source: Writer 2026 Enterprise AI Adoption study . And PwC’s April 2026 AI Performance Study finds 74% of AI’s economic value being captured by 20% of firms. Microsoft’s 2026 Work Trend Index (n=20,000) reports only 16% of workers qualify as “Frontier Professionals” using agents for multi-step work, redesigning workflows, and building a shared AI practice. The other 84% don’t. That’s the 20/80 split in action right there. You might think this is a firm problem. It is not. Why the layoffs are coming anyway For thirty years, large enterprises had no good lever against the 80/20 and the coordination tax that sustains it. Layoffs were politically expensive, operationally painful, and prone to cutting the wrong people along with the right ones. In 2026 they do: deploy agentic AI and let the humans churn out quietly. CFOs now have permission to authorize the cut without a productivity gain to defend it. And so they’re cutting — without ROI or empirical evidence. Source: 2026 Gallup State of the Global Workplace report Gartner’s May 2026 study at $1B+ firms found that 80% of companies that piloted AI cut staff afterwards, with zero correlation between those layoffs and AI ROI. HBR’s January 2026 survey found the same dynamic at the planning stage: 60% of companies have already cut headcount in anticipation of AI productivity gains, and only 2% based the cuts on measured results. Gallup’s Q1 2026 US data shows where the cuts land. Among AI-implementing firms, large organizations (10,000+ employees) are shrinking headcount — 33% reducing workforce versus 30% expanding. Mid-sized organizations (5,000–10,000) net- expand — 23% reducing, 38% expanding. The asymmetry is structural. The human glue holding enterprises and IT systems together — the coordination tax — is exactly that part of your organization that agents are about to render redundant. Companies that survived on their ability to coordinate large groups of humans are about to see their core competency rendered obsolete. Where to apply AI So if you’re looking for a new job, look in the small-and-medium business segment. They suffer from far less organizational bloat, and will be much faster to adopt AI-native ways of working successfully. Intuit’s 2026 AI Impact Report finds 77% of US small businesses now use AI regularly, up from 48% in 2024. 78% say it improved productivity. 43% it increased revenue. Four times as many AI-using SMBs increased hiring as decreased it. Source: Intuit 2026 AI Impact Report . The pattern isn’t US-only: AI-adopting EU firms grow employment 1–2 percentage points faster than non-adopters. The Harvard Kennedy School’s Mind the Gap study attributes more than 95% of the US-EU worker AI-adoption gap to employer encouragement and management quality. In Europe especially, the bottleneck is in the layer above the workforce, not in it. What about when you’re running a company, department or team? Remove layers of middle management. Give teams more autonomy and agency — make them P&L responsible, if you can. Smaller numbers of highly skilled individuals with computer agents can now do the work of entire departments — lower coordination tax, lower wage bill, no need to keep a deep talent bench against demand that may never arrive. Bring strategic work in-house . Marketing, advertising, legal, accounting, research: a single domain expert with an army of agents can now produce better results than most external agencies. The professional services tax is about to become a lot cheaper. Find your data advantage . Generic solutions will lead to generic outcomes. Find alpha in your data — where can your team, department or company learn and iterate faster than your competitors? Frontier labs might win on scale, but they don’t know your customers and stakeholders as well as you do. Practice intentional delegation. The Frontier Professionals in Microsoft’s 2026 WTI research aren’t winning because they use AI for everything. They’re winning because they know when not to . They’re 43% more likely to intentionally do some work without AI, and 53% more likely to pause before starting — to decide what AI should do versus what they should. Build your own automations . This one is so important I decided to dedicate a whole section to it — the one below. Claude for (Your Own) Small Business Because in my opinion, successful AI adoption is less about tooling and more about systems design — about your business operating system. Take for example Claude for Small Business . It ships with 15 workflows out of the box and integrations into QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365. The catch is that those workflows are built for the average SMB. Your business isn’t average. Your edge is in what makes you different — and pre-packaged plugins tend to bend slowly on someone else’s roadmap, the same way SaaS apps always have. Which is where personal software comes in. Building your own MCP server takes 10 minutes. Setting up your own AI workspace takes an afternoon (with the right guidance). Treat automation as the disposable, N=1 software because that is what it has become now that the cost of writing software is trending towards zero. The lines between automation and augmentation are blurring. Kaizen FTW . If you want more confirmation, Microsoft just put a corporate label on what I’ve been writing about for six months. They call it “ Owned Intelligence” — institutional know-how that compounds, is unique to the firm, and is hard to replicate. It’s the same line I drew between rented and owned workflows back in February: marketplace Skills and packaged AI plugins are rented automation. Custom Skills, agents and MCP servers are owned — yours to tailor, adapt, and update. Packaged AI gives you average intelligence and average performance. Owned Intelligence gives your firm intelligence no competitor can copy. That’s the lever for SMBs and high-agency operators in 2026. Not bigger AI. Not cheaper AI. Your AI, tailored to your work, getting sharper every day you use it. If you’re still trying to figure out how to apply all this — that’s exactly what my AI Operators program is for. Four weeks, one-on-one, designing the AI operating system around your actual work. Reply to this email or DM me on Substack if you want to talk. Thanks for reading The Circuit! Subscribe for free to receive new posts and support my work. Subscribe Last week in AI Anthropic announced Claude for Small Business (May 13) — a packaged offering with 15 ready-to-run agentic workflows and 15 skills across finance, ops, sales, marketing, HR, and customer service, integrated with QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365. PwC announced an expanded Anthropic partnership (May 14) deploying Claude Code and Cowork, with 30,000 US professionals to be trained and certified through a joint Center of Excellence. PwC is also launching an “Office of the CFO” business unit anchored on Anthropic’s stack. Anthropic and the Gates Foundation announced a four-year, $200M partnership (May 14) — combining grant funding, Claude usage credits, and technical support across global health, education, and economic mobility, with public-goods deliverables (benchmarks, datasets, evaluation frameworks) for low- and middle-income countries. Thinking Machines Lab published Interaction Models (May 11), full-duplex AI built on a 200ms micro-turn architecture that achieves 0.40-second turn-taking latency on the FD-bench V1 benchmark. Native time-aware multimodal collaboration (audio + video + text) without external turn-detection scaffolding. Microsoft Security disclosed MDASH (May 12) — a multi-model agentic vulnerability-research system that scored 88.45% on the CyberGym benchmark (1,507 real-world tasks), ~5 points ahead of the next competitor. For the machines: Does AI take jobs or expose them? AI exposes jobs more than it takes them. In 2026, Gartner found 80% of $1B+ firms that piloted AI cut staff afterwards, with zero correlation to AI ROI, while HBR reports only 2% of companies based layoffs on measured results. AI absorbs low-complexity coordination work first — the same work that sustained the 80% of disengaged knowledge workers Gallup documents globally. Who is this post for? Senior operators, founders, solopreneurs, and consultants who are already using AI and want a concrete framework for where it actually creates value. It’s for readers deciding whether to bet on packaged AI plugins or build their own custom agents, MCP servers, and workflows around their specific business. What’s the key takeaway? Large enterprises are cutting headcount in anticipation of AI gains while SMBs using AI are net-hiring — Intuit’s 2026 report shows 77% of US small businesses use AI regularly and four times as many AI-using SMBs increased hiring as decreased it. The durable lever is “Owned Intelligence” — custom Skills, agents, and MCP servers tailored to your work — not Claude for Small Business or other packaged plugins built for the average. Why are large enterprises cutting AI headcount without ROI? For thirty years, CFOs had no politically acceptable lever against the 80/20 distribution and the coordination tax sustaining it. Agentic AI gives them permission to cut without defending a productivity gain. Gallup’s Q1 2026 data shows 33% of 10,000+ employee firms are shrinking headcount versus 30% expanding, while mid-sized firms (5,000-10,000) net-expand at 38% versus 23%. What is Owned Intelligence and why does it matter in 2026? Owned Intelligence is Microsoft’s 2026 term for institutional know-how built into custom Skills, agents, and MCP servers unique to your firm — the opposite of marketplace plugins everyone else also runs. Packaged AI like Claude for Small Business gives you average intelligence; Owned Intelligence compounds, is hard to replicate, and gets sharper with use. It’s the same distinction between rented and owned workflows that separates the top 16% “Frontier Professionals” in Microsoft’s Work Trend Index from the other 84%. 24 20 6 Share