Your Employer Won't Train You on AI. Here's Your Plan.

The Federal Reserve Bank of New York’s Use of Gen AI in the Workplace and the Value of Access to Training, published April 14 on Liberty Street Economics, put a number on what most workers already suspect. Thirty-eight percent of employed Americans say having training in how to use AI tools is important to them. Only 15.9 percent say their employer offers it.

That’s a 22-point gap. It’s also a career signal.

The NY Fed researchers went further. Workers most likely to want AI training are the ones least likely to get it. Non-college workers, younger workers, newer employees, non-white workers. The people who need the edge most are being handed the smallest amount of it through official channels. The Fed’s conclusion, stated plainly: closing this gap may be essential to realizing the productivity gains AI promises.

If you’re waiting on your employer to hand you those skills, the data says you’re waiting for something that probably isn’t coming. Not this year, not at scale. Here’s the plan I’d run in your shoes.

Quick Verdict

Finding	What the Data Shows
Workers who say AI training matters	38% (NY Fed, Liberty Street Economics, April 14 2026)
Workers whose employer provides AI training	15.9%
Group with highest want, lowest access	Workers without a college degree
Jobs being reshaped by AI over 2-3 years	50–55% (BCG)
Jobs with 40%+ of tasks automatable today	43% (BCG)
Net US jobs erased by AI per month	~16,000 (Goldman Sachs, reported by Fortune)
Cohort taking the brunt	Gen Z and entry-level workers
What this means for you	Train yourself. Your employer is not a reliable source of the skills that move your career in 2026.

Why the Training Gap Is a Career Emergency

Stack the numbers in order.

BCG’s April 2026 analysis of 1,500 roles across 165 million US jobs found 50-55% of jobs will be reshaped by AI within the next two to three years. 43% of jobs already exceed a 40% task-automatability threshold. Goldman Sachs economist Elsie Peng, looking at actual payroll records, calculated AI is erasing about 16,000 net US jobs per month, with the loss concentrated in Gen Z and entry-level roles. Entry-level software developer employment for workers aged 22–25 is down nearly 20% since 2024.

Now layer the NY Fed data on top. The majority of the workforce says AI training matters. Fewer than one in six are getting it from their employer. The gap is widest for the exact workers whose jobs AI is eating fastest.

Read that as a system, not as isolated stats. The supply of AI training from employers is far below the demand. The demand is not optional. Your career leverage in the next 24 months depends on you individually closing a gap that most employers have quietly decided to leave open.

I wrote about the corporate-ROI version of this problem in The Skill Gap Killing Your AI ROI. Companies with mature upskilling roughly double their AI returns. The ones without it leave half the ROI on the table. The worker-level version of that same trend is simpler. You have to decide whether you’re going to be the half that compounds or the half that gets cost-cut.

Why Employers Are Not Stepping Up

This part is worth understanding if you want to stop expecting help that isn’t coming.

Most AI training decisions inside large companies sit in the HR or L&D function. Those functions are scored on cost, compliance, and completion rates. Not on whether your specific role is durable in three years. When a procurement team buys an “AI literacy” module, they’re checking a box, not building your edge. The NY Fed’s own data shows the gap is worst where economic pressure is highest, which is the opposite of what you’d expect if the market were self-correcting.

Second, most executives don’t know what good AI training looks like. I’ve sat in enough rooms to say that confidently. They know they should do something. They don’t know what. So they buy a subscription to a generic platform, assign three hours of video, and call it a program. The workers who already knew the material are bored. The workers who needed it most learn nothing they can actually apply on Monday.

Third, there’s a quiet incentive problem. If a company invests heavily in training the specific people AI is about to displace, they have to either redeploy those workers (hard) or watch them leave with transferable skills (also hard). Many companies are defaulting to the path of least resistance: hire the skill in, let the rest attrite. That’s what I think is happening at scale, even when nobody says it on a town hall.

The 75% of executives calling their own AI strategy performative is the inner-ring version of this. If leadership isn’t serious about the strategy, they aren’t serious about the training to execute it.

Stop waiting for them. Build the skills yourself.

The 12-Week Self-Training Plan

This is the plan I’d run if I were a mid-career professional today and my employer had nothing on offer. It assumes roughly five hours a week. Most of it is free. All of it compounds.

Weeks 1-2: Baseline Fluency

Goal: Use a frontier model well enough that you feel the productivity shift personally.

Pick one tool. ChatGPT, Claude, or Gemini. I’d use Claude or ChatGPT for most knowledge work. Pay for the paid tier. $20 a month is the price of serious practice. Do not try to learn three tools at once.

Run every non-confidential piece of work through it for two weeks. Emails, meeting prep, drafts, research, analysis, follow-ups. Not to replace your judgment. To see where it beats you, where it matches you, and where it fails. Keep a short log: task type, outcome quality, time saved or lost.

By end of week 2, you should have personal, not theoretical, data on where AI moves the needle in your specific job.

Weeks 3-4: Prompting That Actually Works

Goal: Learn the five or six prompting patterns that separate amateur from professional users.

Role prompts. Chain-of-thought. Few-shot examples. Structured output. Iterative refinement. Task decomposition. That’s essentially the whole list of what matters for most work. There are a hundred YouTube videos on each. Pick one and ship a reusable prompt template for a task you do weekly.

By end of week 4, you should have three to five prompt templates saved in a doc that cut a recurring task from hours to minutes. This is the asset you carry with you across jobs.

Weeks 5-8: A Real Workflow, Not a Demo

Goal: Build one end-to-end AI-augmented workflow in your actual job.

This is where most self-training collapses. People learn tools, watch demos, and never wire them into the work. Don’t skip this.

Pick one workflow you do repeatedly. Lead qualification. Meeting summaries. Status reports. Data cleanup. Content drafting. Whatever it is for your role. Map the current steps. Identify which steps AI can do, which stay human, and how they hand off to each other. Build the version-one workflow. Use it for two weeks. Measure time saved.

You do not need a developer. A doc, a prompt template, and one tool is enough for most workflows. If you want to go further, a no-code automation platform adds leverage, but it’s optional for the first pass.

Weeks 9-10: Specialization, Not Generalization

Goal: Pick one AI skill category that’s actually paying and go deeper.

The 56% wage premium on AI skills is not evenly distributed. Generic prompting is already commoditizing. The premiums are concentrating in skills like AI-augmented software engineering, model evaluation, retrieval-augmented generation workflows, agent orchestration, and domain-specific AI applications in legal, finance, and healthcare.

Pick one adjacent to your current role. Go deep enough in two weeks that you can explain the workflow end to end and show a working artifact. Not a certificate. An artifact. A built thing.

Weeks 11-12: Proof in Public

Goal: Ship evidence that you did the work.

Write a short post on LinkedIn describing the workflow you built, the time it saves, and what you learned. Share the prompt template. Record a two-minute Loom walking through it. Add the skill, with the specific tools and outcomes, to your resume and LinkedIn headline.

This is the step most people skip. Private competence is invisible competence. The workers getting pulled into AI-augmented roles in 2026 are the ones whose competence is visible, not the ones whose competence is highest.

By week 12 you will not be an expert. You will be in the top 20% of knowledge workers on applied AI skills. That’s enough to change your trajectory.

The Five Questions to Run On Yourself Right Now

If the NY Fed data landed on you as a career signal, run these before you close this tab.

1. Which five tasks in my current job are at least 40% automatable, and have I actually tested AI on them in the last 30 days?
2. If my employer cut 16,000 jobs a month for a year and my role was in the substitutable bucket, how many weeks of runway do I have?
3. What’s the last AI skill I added to my resume that’s specific enough to be verified, not generic “proficient in AI tools”?
4. Who would vouch, with a specific example, that I built an AI-augmented workflow in the last 90 days?
5. If my employer never offers me training, what’s my personal 12-week plan, and when does it start?

Most professionals land on one yes. The gap between one yes and five is roughly the gap between riding the AI wave and getting washed by it.

Three Mistakes That Will Keep You Stuck

Waiting for the official program. The NY Fed data is a forecast. Most employers are not going to close the training gap this year. Every quarter you wait is a quarter a more-prepared peer gets further ahead. The plan above assumes the training is never coming, which matches the data better than the alternative.

Chasing certificates instead of artifacts. A certificate says you consumed content. An artifact says you built something. Hiring managers I talk to skim past the first and ask questions about the second. A shared prompt library, a shipped automation, a documented workflow, a Loom walkthrough. These change conversations. Certificates rarely do.

Generalizing when you should specialize. “AI-savvy” is already a background expectation, not a differentiator. A specific pairing wins: AI-augmented financial modeling, AI-augmented customer success, AI-augmented legal research. Pick your domain-plus-AI intersection and become recognizably good at that exact combination. Generic fluency is a floor, not a ceiling.

What This Looks Like a Year From Now

The NY Fed researchers did not hedge their conclusion. Closing the training gap may be the single most essential step to getting the productivity gains AI promises at the macro level. I think they are right, and I also think most employers are going to be slow to act on it.

That leaves a 12 to 18-month window where workers who self-train have structural leverage. They will get pulled into the reshaped versions of their own jobs, not displaced from them. They will command a wage premium that’s still expanding. They will survive the layoff waves that hit the average cohort hardest.

Congress is noticing. The Small Business AI Training Act is an early signal that government sees the gap. Those programs will help some people at the margins. They will not help you in time. The compounding starts the week you decide to stop waiting.

The workers in the top 20% of AI leverage in 2026 are not the ones with better employers. They are the ones who decided, correctly, that their career was their own problem to solve. The NY Fed just told you the gap exists, who it hits hardest, and why nobody is fixing it quickly enough to save you.

Pick up the plan. Run the 12 weeks. Ship the artifact. Your employer isn’t going to do it for you, and that’s actually fine. The upside of the gap being this wide is that closing it yourself still puts you ahead of most of the market.

The window is open. Walk through it.

---

Related Reading:

• AI Skills Now Pay 56% More. Here’s Which Ones.
• OpenAI Admits Jobs Are at Risk. Here’s Your Plan.
• The Skill Gap Killing Your AI ROI
• How to Position Yourself for AI’s Job Market Changes by 2030
• 29% of Your Workers Are Sabotaging AI. Here’s Why.