How to Make Yourself Hard to Cut, Role by Role

Klarna is the case we keep coming back to. It leaned on AI for customer support, the quality disappointed, and it started rehiring humans. The work it found it still needed people for was the emotional, high-stakes kind.

Across the hiring data and the layoff patterns we track, the people doing well in this market have one thing in common. They've put themselves where AI adds to their output instead of taking over their tasks. You can get there from almost any starting point, but the route depends on your job.

Four habits that apply to everyone

Make your AI use visible. Use the tools every day and write down what they did for you in plain numbers: "automated X, saved Y hours." Companies that have mandated "AI-first" operations, Shopify and Duolingo among them, screen for exactly that.

Describe your job by its results. Tasks get automated, and outcomes still need owners, so rebuild your role (and your resume) around what you're accountable for instead of the activities that fill your day.

Be findable. Cold applications barely work right now, but recruiters still search, so give them a portfolio, open-source work, writing or talks to land on.

And attach yourself to a domain, since generic skills commoditize first. Technical or operational skill combined with real depth in a regulated field (health, finance, legal, energy, defense) is the most automation-resistant combination we see in the market.

Engineers should move up the stack or toward the hard parts

Software jobs are splitting in two. On one side are AI-augmented seniors who own systems, and on the other is a shrinking market for people who mainly produce code. If you're in the second group, you have three ways out. Move up into architecture, review and incident ownership. Move into the AI stack, since shipping LLM features is the sharpest differentiator right now. Or head for performance, security and distributed systems, where AI is weakest and the stakes are highest.

The job description taking shape is an engineer who directs and reviews the output of five AI agents.

Support and operations face the most direct substitution

Salesforce's support cuts and IBM's unfilled back-office roles are the template here. Your way out is to operate the automation instead of being the thing it replaces. That means AI-workflow administration, exception handling, checking the quality of what the AI produces, and the escalation tiers where human judgment and empathy are what the customer is paying for. It's the layer Klarna found it couldn't do without.

Management layers are thinning

Meta, Amazon, Google and Microsoft all cut management layers, and said so. Management that's mostly coordination (passing status along, scheduling, reporting) is being absorbed by AI tools and wider spans of control. The managers who hold on are player-coaches who stay technically credible and people who own business outcomes instead of headcount.

The biggest opening is for whoever runs the AI transition itself, redesigning workflows, setting policy and reskilling teams. Every company needs that person. From what we can see, almost none have someone doing it well.

If postings for your role are shrinking

The postings data is unambiguous for manual QA, routine content, data entry, recruiting coordination and production design, and waiting it out is the one move that's guaranteed to lose. Each of those roles has a growing neighbor that reuses most of your experience:

• Manual QA to test automation (SDET) or security testing.
• Content production to AI editorial operations and strategy.
• Data entry and reporting to analytics engineering.
• Recruiting coordination to talent intelligence and employer brand.

Expect the switch to take 6 to 18 months of deliberate effort. Start it while you're still employed if you possibly can, because the market prices employed candidates higher and you'll want the runway.

Treat your career the way these companies treat their portfolios: rebalance often and get out of losing positions early. In every phase of this correction, the people hurt worst were the ones it caught by surprise. You can see the same data the market sees now, so act on it a quarter earlier than feels comfortable.