• The AI Workforce Report
  • Posts
  • #9 MIT just quantified the AI job apocalypse: 11.7% of work ($1.2 trillion in wages) is already technically obsolete

#9 MIT just quantified the AI job apocalypse: 11.7% of work ($1.2 trillion in wages) is already technically obsolete

Also: Anthropic deploys 30,000 FDEs to make AI actually work, OpenAI reveals the Custom GPT productivity unlock, and why AI companies are finally working together on standards.

AuthorAuthor

Ryan Alexiev & Ram Menon
December 16, 2025

đź’° THE BIG STORY

The Iceberg Surfaces: Understanding the Real Economics of AI Displacement

MIT and Oak Ridge National Laboratory just released the “Iceberg Index” modeling 151 million workers with 32,000 skills across 923 occupations. And…the results are stark: 11.7% of US jobs are already technically replaceable by AI today. That's $1.2 trillion in wages companies can automate away with current technology.

The breakdown:

  • Visible 2.2%: Computing and tech roles ($211B)—software engineering, IT support, basic coding

  • Hidden 9.5%: Cognitive work in finance, healthcare, HR, logistics—reading, writing, summarizing, information processing

They call it an "iceberg" because most exposure is hidden in pilot programs, that is, when companies roll this out, it won't be gradual, it'll happen all at once.

🔍 DEEP DIVE

Anthropic just deployed 30,000 FDE’s to bridge the "last mile" gap

Anthropic partnered with Accenture to deploy 30,000 Forward Deployed Engineers:the largest validation yet that AI success isn't about models, it's about deployment expertise.

The market opportunity just got confirmed: When frontier labs invest billions to acquire FDE capability at scale, they're admitting the truth: 95% of AI pilots fail at the last mile. This creates massive opportunity for specialized deployment teams who can move faster and cheaper than Accenture's machine.

Where specialized FDEs win: Accenture targets Fortune 500 with 18-month engagements and premium pricing. That leaves the entire mid-market, fast-moving enterprises, and specific verticals wide open. While Anthropic's 30K consultants navigate procurement committees, specialized FDE teams are shipping production systems.

As Sriram Chakravarthy, CTO of Avaamo, wrote in his analysis of the last-mile problem: "The best AI technology in the world is worthless if it can't actually run where your business happens. And making that work, in all its messy complexity, requires human expertise that can't be automated away."

Avaamo's "AgentOps" teams function as dedicated FDEs who've accumulated deployment context across hundreds of enterprise environments—understanding legacy systems, navigating security reviews, and delivering pragmatic solutions that unblock customers in hours, not weeks. Different markets, different playbooks—both can win.

🤝 INDUSTRY NEWS

AI companies team up on agent standards (finally)

OpenAI, Anthropic, and Block are creating open standards for agentic software, solving the interoperability problem. Right now, Claude agents can't talk to GPT agents creating integration hell as enterprises deploy dozens of agents.

Standards make the ecosystem more valuable, reduce implementation risk, and prevent any single company from capturing the interoperability layer. This makes the "30,000 FDE" strategy viable; consultants focus on business context instead of tech stack expertise.

🏛️ GOVERNMENT

IRS deploys AI agents to answering tax questions while enterprises struggle to reach production

As IRS workforce declined 25% (100,000 to 75,000 employees), the agency rolled out Salesforce's Agentforce across multiple divisions for case summarization and search. "I think all of us have to realize that the change is coming," said Rob Fitzpatrick, senior counsel with 38 years at the agency.

The irony: While 95% of enterprise pilots fail, a government agency facing massive cuts successfully deploys AI agents at scale.

Read the full story: IRS deploys AI agents →

đź’ˇ WHAT ACTUALLY WORKS

Must read: The state of enterprise AI by OpenAI

OpenAI’s first annual enterprise report draws on data from over one million business users to reveal a shift from casual AI experimentation to deep, structural integration.

While the average employee saves roughly 45 minutes a day, the top 5% of power users are saving 10+ hours a week. What are they doing differently? They’ve stopped treating AI like a chatbot and started using it as an infrastructure for complex workflows (reasoning token usage is up 320x YoY 🤯). If you want to know what the top 1% of operators are doing to pull ahead, this is the benchmark you need to read.

The takeaway: Pick a specific high-volume workflow, build an agent for that task, measure task-level productivity. This is why Accenture deploys customized agents, not generic tools.

That's it for now, talk soon —Avaamo Team