The AI Digest · Issue #4 · March 2026
March 20, 2026 · bindlcorp.com · 7 min read
The conversation about AI and jobs just changed shape. For months the debate has been about whether AI is going to replace people. This week, the answer started showing up in actual numbers from real companies. McKinsey is running 20,000 AI agents alongside 40,000 human employees. A major software platform deployed AI specialists that handle 90 percent of IT requests autonomously. And Jensen Huang told the world Nvidia will have 7.5 million agents within a decade. Here’s what’s actually happening — and what to make of it.
The Shift
AI Agents Are Already Working Alongside People At Real Companies. Here’s What That Looks Like.
McKinsey — one of the largest consulting firms in the world — revealed this week that they currently run 20,000 AI agents working alongside their 40,000 human employees. A year ago they had 3,000 agents. Their managing partner said they expect to reach a roughly equal number of agents and humans within 18 months.
That’s not a prediction or a roadmap slide. That’s a company that’s already done it describing what the trajectory looks like from the inside. The agents handle specific, defined workstreams — research, analysis, documentation — freeing their human consultants to focus on client relationships, judgment calls, and the work that requires actual experience in the room.
ServiceNow, a major enterprise software platform, went further this week and launched what they’re calling an “autonomous workforce” — AI specialists assigned actual job roles with full access to company systems. Their first product is an AI that handles IT support requests end-to-end: password resets, software access, network troubleshooting. Their own internal numbers show it resolving more than 90 percent of IT requests completely on its own, 99 percent faster than a human agent. Those aren’t benchmark numbers. That’s live production at scale.
20,000
AI agents running alongside 40,000 humans at McKinsey today — up from 3,000 a year ago
90%+
IT support requests handled autonomously by ServiceNow’s AI specialist — in live production
7.5M
AI agents Jensen Huang expects Nvidia to run alongside ~75,000 human employees within a decade
What to do with this
The question is no longer whether AI agents will work alongside humans — it’s already happening at some of the largest companies in the world. The more useful question for most people is: what parts of your job look like what McKinsey’s agents are doing? Research, documentation, first-pass analysis, routine requests — those are the tasks that are being absorbed first.
That doesn’t mean your job disappears. It means the job changes shape. The consultants at McKinsey still have jobs. They’re just not doing the same tasks they were doing a year ago.
The Finding That Matters
Stanford Economists Said Something This Week That Cuts Through Most Of The Noise.
At a major economic summit this week, Stanford researchers presented a finding that reframes the whole debate. Employment is actually falling among workers who use AI to automate tasks — but growing for workers who use AI to learn new skills. The way you use the technology matters as much as whether you use it.
The distinction is worth sitting with. If you use AI to complete your current tasks faster and leave it there, you’re essentially demonstrating to your employer what can be automated. If you use AI to stretch into work you couldn’t do before — adjacent skills, broader scope, higher-level output — you’re building a position that’s harder to replicate.
The researchers also made a point about what humans still have that AI doesn’t: the ability to define the problem, not just execute the solution. Their advice was to identify the bottlenecks in your workflow that can’t be automated — the judgment calls, the stakeholder relationships, the moments where context that lives in your head is the entire value — and treat those as your core offering. AI handles the execution. You own the direction.
The two types of AI users — and which one wins
Using AI to automate: You do your job faster. Your employer learns what’s automatable. You become the benchmark for what a more efficient replacement would cost.
Using AI to learn: You expand into work you couldn’t do before. Your output increases. Your value to the organization grows in a direction that isn’t obviously replaceable.
What to do with this
This week: pick one thing you’ve always wanted to be able to do at work but couldn’t — writing better presentations, understanding financial reports, drafting proposals, doing market research. Use an AI tool to actually do that thing. Not to speed up what you already do. To do something new. That’s the version of AI fluency that the Stanford data shows is building value rather than demonstrating replaceability.
The Story Nobody’s Covering
While Everyone Watches White-Collar Jobs, Skilled Trades Are Running Short Of Workers.
All the AI infrastructure being built right now — the data centers that power every AI tool you’ve ever used — has to be physically constructed and maintained. That requires electricians, HVAC engineers, construction workers, and robotic technicians. And there aren’t enough of them.
A new analysis of 50 million job postings found that demand for HVAC engineers is up 67 percent since 2022. Robotic technicians are up 107 percent. Traditional electricians and construction workers are up 27 percent. Advertised wages for HVAC roles have risen 10 to 15 percent over four years — and six-figure salaries are now achievable in data center work without a college degree. The AI economy isn’t creating unemployment. It’s creating a mismatch. The jobs being squeezed and the jobs being created are in completely different places.
+107%
Demand for robotic technicians since 2022 — driven by AI infrastructure buildout (Randstad, 2026)
+67%
Demand for HVAC engineers over the same period — with six-figure salaries now achievable
What to do with this
If you’re in a trade or considering one — or if you have kids making career decisions — this data matters. The AI economy is creating enormous physical demand that can’t be outsourced or automated away, at least not yet. The people building and maintaining the infrastructure that makes AI run are not in the line of fire. They’re in the job market’s strongest position right now.
If you’re in a white-collar role, the flip side is worth understanding too: the mismatch means there’s no army of workers waiting to absorb displaced office workers. The transition isn’t going to be clean, and the retraining infrastructure isn’t there yet. That’s an argument for building AI fluency now, while the window is open.
This Week’s Move
Look Up Your Job’s AI Risk Odds. Then Do Something About It.
Researchers published a tool this week that lets you enter your job title and see estimated odds of AI replacing it, based on Anthropic’s data on how AI is actually being used across workplace tasks. Desk-based, documentation-heavy roles show the highest risk. Physical and in-person jobs show the lowest. It’s not a guarantee — but it’s real data, not a guess.
The move this week: look it up. If your role shows high exposure, identify the one task in your job that’s most repetitive and document-heavy, and spend 20 minutes this week doing it with an AI tool instead. Not to show anyone. Just to know what that feels like and what it produces.
The Stanford finding is the frame: use AI to learn, not just to automate. That’s the version of this that builds something durable.
Everything Else This Week
Quick Hits
You can now bet on whether AI will take your job — A sports analytics platform built a tool using Anthropic’s job exposure data that converts AI risk into betting-style odds for specific roles. It’s an unusual way to present the data but the underlying research is real. Communication, research, and documentation-heavy jobs show the highest odds. Physical, in-person roles show the lowest.
Jensen Huang’s agent framing is worth noting — When Nvidia’s CEO talks about 7.5 million AI agents working alongside 75,000 humans, he’s not framing it as replacement. He’s framing it as scale: the agents handle volume and speed, humans handle direction. Whether that framing holds up in practice across other industries is the open question — but it’s notable that the person whose company profits most from AI adoption is making the case for augmentation, not elimination.
Goldman Sachs: 6-7% workforce displacement over 10 years — Their economists put a number on it this week. In their base case — AI adoption over about a decade — they expect 6 to 7 percent of workers to be displaced during the transition. If adoption accelerates and frontloads, the impact is larger. If it spreads slowly, more manageable. The range of outcomes is still wide. The direction isn’t.
Only 5% of workers are AI-fluent — and they earn 4.5x more — A new analysis found that workers with actual AI skills aren’t just slightly better positioned. They’re earning four and a half times more than their peers and getting promoted four times as often. The gap between AI-fluent and AI-unfamiliar workers is already significant. It’s growing.
Washington Post published a job lookup tool this week — Built using research from GovAI and the Brookings Institution, it lets you search your specific occupation and see both its AI exposure and its adaptability score — based on factors like savings, transferable skills, and age. Worth five minutes of your time.
New to AI tools?
The Gap Between AI-Fluent And Everyone Else Is 4.5x. Here’s Where To Start.
The Stanford finding this week is the most direct case yet for starting now: using AI to learn new skills builds value. Using it just to go faster doesn’t. The first step is getting comfortable enough with the tools to use them for real work — not tests, not experiments, but actual tasks you need to get done.
If you haven’t started, the guide below gets you there in 20 minutes. No account required to read it.
Free — no account needed
Learn AI
Practical guides to getting started with AI tools — built for people who want to actually use them, not just understand what they are.
Start Learning →
Leave a Reply