Last May, I wrote about the hidden AI job boom — the roles being created by AI that the “your job is gone” crowd conveniently ignores.
That piece struck a few nerves. It also drew a sharp critique: sure, new roles exist, but are they enough to absorb the displaced?
Fair challenge.
Almost a year on from my initial observations, I’ve spent time pulling apart the actual evidence (enterprise adoption surveys, security research, platform documentation) from public sources to see if it correlates with anything I see from my vantage point in the field building with AI every day.
Are there any signs of things changing in the labour market? Is the scary AI narrative of “mass white-collar job loss is imminent” actually real?
Here’s what I found: the doomers aren’t wrong about the direction.
They’re wrong about the timeline.
And that distinction matters enormously — because it means you have a window to act. But it won’t stay open forever.
Let’s get into it.
The Plumbing Isn’t Built Yet
The displacement narrative assumes AI systems are ready to slot into your company and start doing your job. They aren’t.
Not even close.
Getting AI to do useful work behind a corporate firewall — connecting to real data, respecting real permissions, handling real compliance requirements — demands a layer of engineering that barely exists yet.
Consider what it actually takes to connect an AI model to your company’s internal systems. You need secure authentication (OAuth, SSO), permission-aware data retrieval so the model only sees what it should, audit trails for compliance, monitoring for when things go wrong, and safeguards against prompt injection and tool misuse.
This isn’t a ChatGPT subscription. It’s enterprise integration engineering — and the tooling is still being hardened.
In April 2026, Gartner warned that enterprise GenAI security incidents are growing precisely because these integration layers — including the Model Context Protocol (MCP) that connects AI to business systems — were designed for speed, not security enforcement.
Security researchers at Obsidian disclosed OAuth vulnerabilities in MCP implementations throughout 2025, and the guidance is still catching up.
The people who build this stuff? They’re only just being hired - if employers can actually find them.
Google Cloud recently posted a role for a “Forward-Deployed Engineer” whose job is explicitly to implement secure agentic workflows using MCP, OAuth, and tool-calling across enterprise APIs and legacy systems. That job description reads like a blueprint for the behind-the-firewall integration that would need to exist before AI could meaningfully replace knowledge workers at scale.
And there aren’t enough of these people.
The UK government’s 2025 AI labour market survey found that large majorities of AI employers report technical skills gaps, with a material portion saying shortages have directly impacted business goals.
The European Commission reported that while AI talent in Europe more than doubled between 2016 and 2023, it still represents a tiny fraction of the workforce. Meanwhile, Deloitte’s 2026 enterprise AI report identifies the AI skills gap as the single biggest barrier to integration.
In short: AI can’t take your job until someone builds the system that would replace you. Those builders are scarce, the systems are immature, and the security surface is still being mapped.
The Pilot Graveyard
Even where companies are trying, most are (still) failing to scale.
A March 2026 survey of 650 enterprise technology leaders found that while 78% have at least one AI agent pilot running, only 14% have scaled to production.
A separate DigitalOcean report puts the production rate at just 10%.
The reasons are telling: integration complexity with legacy systems, inconsistent output quality at volume, missing monitoring infrastructure, unclear organisational ownership, and unpredictable nature of LLM token costs.
These aren’t exotic problems.
They’re the same operational and governance failures that have stalled every previous wave of enterprise technology — RPA, blockchain, IoT. AI is no different.
McKinsey’s November 2025 survey confirms the pattern: 88% of organisations say they’re using AI, but the majority remain in experimentation, and only about a third have begun scaling enterprise-wide.
Deloitte found that while productivity gains are common, revenue impact remains largely aspirational — only 20% of organisations report actually growing revenue through AI, compared to 74% who hope to.
Meanwhile, 84% of companies have not redesigned jobs or workflows around AI capabilities.
You don’t fire people when you haven’t even changed the process yet.
But Don’t Get Comfortable
Here’s where I depart from my own previous article.
Last year, I argued that AI creates more roles than it destroys.
I still believe that’s true in certain contexts — but I wasn't being honest enough about the direction of travel.
The displacement threat is real.
It’s just on a longer fuse than the headlines suggest.
Enterprise-wide AI implementation has doubled from 12% to 24% in a single year.
PwC’s 2025 AI Jobs Barometer found that AI-skill job postings continued growing even while overall postings declined, and that AI-exposed jobs are seeing skills requirements change faster than ever.
The demand curve is accelerating, even if production maturity lags behind. My own backlog of demand in my day job is testament to this.
When the operational maturity catches up — when the integration engineers have been hired, the security patterns standardised, the governance frameworks established, compliance departments become more comfortable with the risks — the deployment curve will steepen fast.
History tells us enterprise technology follows long incubation periods and then rapid standardisation. It feels to me as though we’re still in the incubation period - but it won’t last forever.
This means the right framing isn’t “your job is safe.” It’s: you have a runway. Use it.
What To Do With the Runway
If the displacement timeline is measured in years rather than months — which I think it is — the question becomes: how do you use that window?
I’ve got two teenagers who’ll be entering the workforce in the next three years. I think about this constantly — not as someone who builds AI systems, but as a dad watching the ground shift under the career paths they’re being told to follow.
Here’s where I’d focus.
1. Understand Where You Sit on the Automation Curve
Not all roles face the same exposure at the same speed. Routine, text-heavy, data-processing work is closest to the front of the queue.
Work that requires navigating ambiguity, managing relationships, exercising judgment across incomplete information, or operating in physical environments is further back. Be honest about where your role sits — not where you’d like it to sit.
2. Build the Skills That Sit Between AI and the Business
The highest-value positions emerging right now aren’t pure technologists. They’re people who can bridge AI capabilities with business reality:
Understanding what models can and can’t do
Scoping realistic use cases, designing governance frameworks
Managing the human-AI workflow
If you’re a domain expert who develops AI literacy, you become the person organisations can’t easily replace — because you understand both sides of the equation.
3. Move Toward the Complexity
AI handles routine well. It handles complexity and edge cases poorly.
Deliberately steer your career toward work that involves judgment, exception-handling, cross-functional coordination, and novel problem-solving.
These are the tasks that remain hardest to automate and that become more valuable as AI handles the routine layers beneath them.
4. Don’t Wait for Your Employer to Retrain You
84% of companies haven’t redesigned their workflows around AI. Many won’t invest in retraining until they’re forced to.
If you wait for your organisation to tell you what to learn, you’ll be behind.
Start now.
The AI governance, LLMOps, and integration engineering spaces are all accessible to people willing to learn — and the talent shortage means the door is wide open.
If those areas feel out of reach right now, keep in mind that the vast majority of people still haven’t yet mastered prompting. In 2026, this is table stakes at work and is as applicable to daily life in general as using WhatsApp.
5. Watch the Production Metrics, Not the Headlines
The signal that displacement is accelerating won’t come from a Diary of a CEO episode.
It’ll come from the pilot-to-production ratio shifting — from 14% to 30%, then 50%.
It’ll come from enterprise AI implementation rates continuing to double.
It’ll come from the MCP/OAuth security stack maturing to the point where integration becomes routine rather than bespoke.
Please. Track those indicators. They’ll tell you more about your career timeline than any pundit.
The Bottom Line
The AI job apocalypse has a scheduling problem.
The systems that would need to exist to displace white-collar workers at scale are stuck in pilots, short on talent, and still being secured.
That’s not a permanent condition — it’s a window.
The doomers are wrong about the when. But they’re not entirely wrong about the what.
The question isn’t whether AI will reshape white-collar work. It’s whether you use the years you have — and it is years, not months — to position yourself on the right side of that change.
The runway is open. Don’t waste it.
Until the next time,
Chris
🧰 Not sure where to start?
Everything I've learned about navigating AI in the enterprise — readiness assessments, cost models, governance frameworks — is in the AI Agent’s Toolkit. It's the practical side of everything above. Have a look around and see if anything fits where you are right now.