AI Didn't Take the ID Job. It Exposed Who Was Actually Doing It.
There's something most instructional designers know but won't say in a team meeting.
When leadership asks you to track your time savings from AI, you already know what they're planning to do with that data.
It's not a productivity study. It's a headcount conversation waiting to happen. And being asked to build the efficiency case for your own role — so someone else can decide whether to keep paying for it — is one of the more demoralizing things I've watched land on L&D teams in the past two years.
I've been through two AI-driven layoffs. I'm not writing this from the sidelines, and I'm not writing it to be encouraging. I'm writing it because I'm now building the exact systems that are changing what the ID role looks like, and I'd rather you hear what's coming from someone who's inside it than be caught off guard in a town hall.
The reframe that actually holds up
Here's what I've seen work in the instructional designers who are navigating this well: they stopped thinking about AI as a threat to manage and started treating it like a new hire to direct.
That shift sounds simple. It isn't. Because directing AI output well requires everything an experienced ID already knows — learning science, performance consulting, stakeholder alignment, judgment about what learners actually need. That's exactly what makes it hard for someone without that background to just step in and do it.
Someone said it better than I had in a recent conversation: once production gets cheap, judgment becomes the scarce skill.
That's the whole thing right there.
AI can draft a course. It can generate knowledge checks, outline modules, write voiceover scripts. What it cannot do — and I want to be specific here, not just reassuring — is:
Decide what shouldn't be taught
Diagnose a real performance gap versus a perceived one
Read a stakeholder room and know which version of the truth they're ready to hear
Build the relationship with the SME that gets your calls returned
That last one still runs on donuts. I'm not joking.
Those aren't soft skills you list on a resume. They're the actual design work. AI produces content that looks finished. That is not the same thing as training that changes behavior. The difference between those two things is an instructional designer who knows the learner, the gap, and the organization — and is in the room making calls no model can make.
There's a term for what happens when that human judgment gets pulled out of the process: vibe education. Content that looks polished, reads confidently, and doesn't actually work. It's already out there. You've probably seen it and had to sit through it.
What's coming that nobody's preparing you for
Most L&D teams are using AI one tool at a time right now. ChatGPT for drafts, maybe a voiceover generator, maybe an image tool. That stage is manageable. You can stay on top of it.
The next stage is different.
Agentic workflows are multiple AI agents working together as a connected system — one handling research, one drafting, one reviewing structure, one checking alignment — running in sequence without a human touching every step. I'm building these now, not in theory, in actual client work.
When these get deployed inside training departments, the question won't be how do we use AI? It will be who sets the standards these agents work inside of, and who catches what they miss?
That's the gatekeeper role. And it is not a lesser version of the ID job. It's the most senior version of it.
The people who do well in that environment aren't the ones who resisted long enough that someone made the decision for them. They're the ones who understood how these systems work, built a process for reviewing AI output the way a manager reviews a new hire's work, and positioned themselves as the person the org can't cut — because they're the reason the AI output is actually usable.
What this looks like in practice
This isn't about getting better at prompting. Prompting is a skill, not a strategy.
Review AI output like you'd review a new hire's first draft. Don't just check for errors. Check for what it missed. What assumption did it make about the learner? What's absent that should be there? What's present that shouldn't be?
Own the decisions AI can't make. Before any content exists, you're deciding what the course is actually trying to change, what a good outcome looks like, and whether training is even the right solution. Stay upstream of the tool.
Document your standards. If you're going to direct AI output, you need to know — in writing — what "good" looks like for your learners, your organization, your stakeholders. That document becomes the brief every workflow runs against.
Know what's coming before it shows up in a company meeting. That's exactly why I put together the free guide below. The IDs I talk to who are doing well didn't get a heads up from leadership. They went and found out for themselves.
The guide
I wrote Manage AI Like an Employee for instructional designers and L&D leaders who want a practical framework for staying in the judgment role while AI handles more of the production side.
It covers what to delegate, what to keep, how to build a quality control process for AI output, and what your team needs to be ready for when agentic workflows become standard — not someday, but soon.
It's free. No catch.
If you've been through a layoff, or you're watching your team shrink while the workload doesn't — I'd like to hear what part of this you haven't been able to say out loud yet. Drop it in the comments.
