A Career Playbook for the AI Era: 10 Steps Grounded in Research

This is the final post in a six-part series on AI and careers. Previously: AI displacement is real, which jobs are most exposed, deskilling vs. upskilling, company restructuring, and seniority and experience.

Stop "upskilling in AI." Start positioning around what AI can't do.

Over the past five weeks, I've worked through the questions I promised in my original post, grounded in peer-reviewed research and real labor market data. Here's a quick recap:

1. Displacement is real, already visible in payroll data, and concentrated in roles where AI automates codifiable work tasks. (Part 1: AI displacement is real)
2. The roles most exposed aren't the ones people expect. It depends on the task composition of your daily work, not your job title. (Part 2: Which jobs are most exposed)
3. AI edits your role task by task. Whether it deskills you (takes the hard parts, leaves the routine) or upskills you (takes the routine, leaves the judgment) depends on which work tasks go first. (Part 3: Deskilling vs. upskilling)
4. Most companies adopted AI but haven't restructured. That's changing: 32% plan significant cuts this year, and executives and employees disagree on what's coming. (Part 4: Company restructuring)
5. AI displacement creates a seniority bias: entry-level professionals in AI-exposed roles are hit hardest, while experienced professionals in those same roles are growing. The mechanism is codified vs. tacit knowledge. But the protection is conditional. (Part 5: Seniority and experience)

Along the way, we've also addressed the speed of this transition (it's 10x faster than prior technology shifts), whether AI-resilient roles exist (they do, anchored in interpersonal engagement, regulatory accountability, physical presence, and judgment under ambiguity), and what mid-career professionals should genuinely worry about (not whether AI touches your role, but whether the edit it makes is deskilling or upskilling you).

If you've followed the series, you now understand the landscape better than most. But every week, the same question kept coming up: "Okay, but what should I actually do?"

Here's my answer, whether you're early in your career or well into it.

---

First, let's address the most common advice: "upskill in AI"

You've heard it everywhere. Learn prompt engineering. Take an AI course. Get certified. The advice is well-intentioned, and it's not wrong exactly. But as a career strategy, it's incomplete in a way that matters.

Here's why. The productivity research (Noy & Zhang, Science 2023; Brynjolfsson et al., NBER 2023; Dell'Acqua et al., BCG/HBS 2023) consistently shows one thing: AI helps everyone produce better output on structured work tasks. Lower performers gain the most. The performance gap between juniors and seniors narrows on codifiable work. When everyone can produce similar-quality output using AI, the tool proficiency stops being a differentiator. It becomes the floor.

Think of it this way. "Learn prompt engineering" in 2026 is like "learn to use spreadsheets" in 1995. It was correct advice. Spreadsheet fluency was genuinely necessary. But nobody built a career by being "the spreadsheet person." The people who built careers were financial analysts, operations managers, and consultants who used spreadsheets as one tool among many. The tool proficiency was the entry ticket, not the destination.

The same is true for AI fluency. 84% of developers already use AI tools (Stack Overflow, 2025). AI literacy is the fastest-growing skill on LinkedIn. The 56% wage premium for AI-skilled professionals (PwC, 2025) reflects current scarcity, not permanent differentiation. As AI fluency becomes universal, the premium will compress. What will hold its value is the combination: domain expertise that lets you catch what AI gets wrong, direct AI toward what matters, and make judgment calls that AI can't.

As MIT Technology Review put it (May 2026): "The competition most young professionals will experience is not human versus machine but colleague versus AI-augmented colleague." The differentiator isn't AI fluency. It's what you bring on top of AI fluency.

---

The prescriptive part: what to do, starting this week

In Part 5, I outlined three principles for mid-career professionals (keep building tacit expertise, become the AI-augmented version of yourself, position around your durable assets) and three paths for early-career professionals (target redesigned roles, build through freelancing and ventures, combine AI fluency with domain depth). Here's how to put those into practice.

I've organized these by timeframe. The first group takes a couple of hours. The next group takes a month. The last group is a quarter-long commitment. Some apply to everyone, some are specific to where you are in your career. I've noted which is which.

This week

1. Understand your own task mix. (Everyone)

If you're currently employed, look at your calendar from last week. Sort your time roughly into two categories: work tasks where AI could produce a comparable result (structured analysis, report drafting, data compilation, routine communication) and work tasks where it couldn't (stakeholder negotiations, judgment calls under ambiguity, client relationships, exception handling, mentoring).

Write down the ratio. Something like "12 hours exposed, 6 hours durable." This is your personal exposure score. It's more useful than any role-level average because it reflects how you actually spend your time.

If you're job hunting, do this exercise for the roles you're targeting. Look at the job description, identify which listed responsibilities AI could handle and which it couldn't. Roles where most responsibilities are codifiable (data processing, report generation, routine analysis) are the ones where hiring is shrinking fastest. Roles with a blend of codifiable and judgment-intensive work are where hiring is growing.

2. Ask yourself: which parts is AI taking? (Everyone)

Think about the last time AI saved you real time at work (or in a project, if you're a student or between roles). Was it on the hardest part, or the most tedious part?

If AI handled the hard stuff (complex analysis, research synthesis, strategic drafting), it's doing your highest-skill work. That's the deskilling edit from Part 3. If it handled the tedious stuff (formatting, data compilation, boilerplate), it cleared the routine. That's the upskilling edit.

This tells you which direction your role (or target role) is moving.

3. Read the signals from your company (or target companies). (Everyone)

If you're employed, scan for these signs: are open roles in your department being left unfilled? Have job descriptions started mentioning AI proficiency? Is leadership talking about "doing more with the current team"? Have teams been merged or restructured?

If these signs are absent, your company is likely in Phase 1 (tools deployed, no structural change, roughly 80% of companies). You have a window. If you see several of these, your company is in Phase 2 (workflows being redesigned, about 20%). The window is smaller. If you see explicit headcount targets tied to AI, new AI-specific roles being created, and management layers compressed, that's Phase 3 (operating model changing). Act now.

If you're job hunting, apply this same lens to the companies you're considering. A company in Phase 2 or 3 that's creating new AI-adjacent roles (AI quality review, workflow design, agent oversight) may be a better bet than one in Phase 1 where the restructuring hasn't started and the disruption will come later.

This month

4. Shift your work mix toward the durable side. (Currently employed)

Identify one exposed work task you own (a recurring report, a data compilation, a routine review) that could be automated, delegated, or deprioritized. Redirect that time toward something durable: a judgment-intensive project, a client relationship, a cross-functional initiative that requires your institutional knowledge.

Don't make a production of it. Just shift a few hours per week. Over a quarter, this reshapes what your role actually consists of.

5. Use AI on your hardest problems, not your easiest ones. (Everyone)

This is counterintuitive but important. Don't use AI to draft the emails you could write in 5 minutes. Use it to stress-test your thinking on a complex decision. Ask it to challenge the assumptions in your proposal. Have it generate counter-arguments to your strategy. Use it to rapidly research an unfamiliar domain before a meeting.

Why does this matter? In Part 5, we covered how tacit knowledge (judgment, pattern recognition, domain intuition) is what protects experienced professionals from displacement. But tacit knowledge only grows when you do hard work that stretches your judgment. If you use AI only on routine tasks, you get more efficient but you stop growing. If you use AI to push your thinking on hard problems, you build new tacit expertise while leveraging the tool. That's the difference between the deskilling path and the upskilling path applied to how you use AI itself.

6. Reposition how you present yourself. (Everyone)

Update your LinkedIn headline and summary to lead with your durable assets: judgment, domain expertise, relationships, results that required your specific experience. Not the tools you use, the processes you follow, or your output volume.

If you're early in your career and don't yet have years of tacit expertise to lead with, lead with the combination of AI fluency and your domain. "Finance grad with AI-assisted valuation modeling" beats "finance grad." "Mechanical engineer building AI-augmented manufacturing workflows" beats "mechanical engineer." The combination is what companies like IBM and Cognizant are hiring for in their redesigned entry-level roles.

You're not declaring a job search. You're making sure that how you're perceived matches the value you deliver (or can deliver).

7. Have one real conversation. (Everyone)

Not networking for networking's sake. One genuine conversation with someone in a role, company, or function you'd want to be in if you needed to move. Ask what their work actually looks like day to day. Ask what skills translate. Ask what gaps they see in candidates.

If you're early-career, target people in the redesigned entry-level roles we discussed in Part 5: roles at companies like IBM or Cognizant that emphasize AI output review, customer interaction, and exception handling rather than the codifiable grunt work AI now handles. Learn what these roles actually look like from the inside.

This quarter

8. Make one case for reshaping your role (or build one thing outside it). (Adapt to your situation)

If you're employed: volunteer to own a judgment-intensive project. Propose that a routine task be automated so your time goes to higher-value work. Ask to be involved in an AI-related initiative in your department. The goal is to shift your role description, not just your task allocation, so that your next performance review reflects the durable work you're doing.

If you're early-career or between roles: build one real thing. A freelance project that demonstrates domain judgment. A small venture that required you to make decisions AI couldn't. A portfolio piece that shows you can direct AI output, not just consume it. As we covered in Part 5, 38% of recent graduates are considering starting their own businesses. The same AI tools that closed entry-level doors can power a venture that builds the tacit experience employers value.

9. Build one piece of public evidence. (Everyone)

A LinkedIn post about a decision you made and why. A case study of a project where your judgment mattered more than the execution. A talk at an internal or external event. Something that creates a visible track record of the work AI can't replicate.

This isn't vanity content. It's positioning. When hiring managers or internal stakeholders evaluate you, they should see evidence of judgment, not just output. If you're early-career, this is especially important: you may not have a decade of tacit knowledge, but you can demonstrate how you think through public writing about problems you've solved.

10. Make the stay-or-go decision deliberately. (Currently employed)

If your company is in Phase 1 or early Phase 2, staying and positioning internally is probably the right call. You have time. Use it to reshape your role and build visibility around your durable strengths.

If your company is in late Phase 2 or Phase 3 and your current role is mostly exposed work, the restructuring is coming regardless of how well you position. Starting an external search from a position of strength (employed, clear about your assets, with a visible track record) beats starting one reactively after an announcement.

Either way, make it a deliberate decision. Not drifting. Not "waiting to see." The research from Part 4 is clear: executives are planning changes they haven't communicated. The gap between what leadership knows and what employees expect is real. Don't be on the wrong side of that gap.

---

Making it personal

The steps above work at the general level. But the specific answers are different for every person. Which of your work tasks are exposed? What are your durable assets? Which career paths make sense for your background in your market? What does your 30-day plan look like?

If you're early in your career and evaluating which roles to target, Alignment Check shows the task composition of any role, scored by AI exposure.

---

This has been a six-week journey through the research on AI and careers. We started with ten questions in the manifesto, and I hope these six posts have given you a research-grounded perspective on each of them. If any of it shifted how you think about your own situation, I'd genuinely like to hear about it. And if you try the playbook, let me know what you learn.

The data keeps evolving, and so will this conversation.

Thanks for reading.

---

Research referenced across the series:

1. Brynjolfsson, Chandar & Chen, "Canaries in the Coal Mine," Stanford Digital Economy Lab, November 2025
2. Hosseini & Lichtinger, "Generative AI as Seniority-Biased Technological Change," Harvard / SSRN, August 2025
3. Ide, "Automation, AI, and the Intergenerational Transmission of Knowledge," SSRN, April 2026
4. Eloundou, Manning, Mishkin & Rock, "GPTs are GPTs," Science, 2024
5. Chen, Srinivasan & Zakerinia, "Displacement or Complementarity?" Harvard Business School, August 2025
6. Gupta & Kumar, "Agentic AI and Occupational Displacement," arXiv, March 2026
7. Barrero, Bloom, Davis et al., "Firm Data on AI," NBER Working Paper, March 2026
8. Humlum & Vestergaard, "Large Language Models, Small Labor Market Effects," NBER / PNAS, September 2025
9. Acemoglu, "The Simple Macroeconomics of AI," Economic Policy, 2025
10. Noy & Zhang, "Experimental Evidence on the Productivity Effects of Generative AI," Science, 2023
11. Brynjolfsson, Li & Raymond, "Generative AI at Work," NBER, 2023
12. Anthropic, "The Anthropic Economic Index: Economic Primitives," January 2026
13. World Economic Forum, "Future of Jobs Report 2025," January 2025
14. MIT Technology Review, "It's time to address the looming crisis in entry-level work," May 2026