How to Redesign Your Own Job for AI, Without Waiting for Your Employer
Your job is quietly changing shape. Some part of what you were hired to do is now something you supervise rather than produce, and the description on file still describes the old version.
The obvious move is to wait for your employer to catch up.
Here is the problem with waiting. ON Partners surveyed executive roles and found 94% of executives saying AI is reshaping their role, while only around 9% of organizations substantially rethink a role before filling it. Even at the moment the job is literally vacant and somebody has to write the posting, nine in ten organizations backfill the old one.
The thing you are about to do has a name
This is not a new idea invented for AI, which is good news, because it means somebody has already checked whether it works.
Amy Wrzesniewski and Jane Dutton named it in 2001. The changes people make to the boundaries of their own job without being asked: which tasks they take on and drop, who they work with, and how they frame what the job is for. The researchers split it three ways. Task crafting changes what you actually do. Relational crafting changes who you do it with. Cognitive crafting changes how you understand the purpose of it, which sounds like the soft one and is the one that tends to survive. is the bottom-up counterpart to the top-down redesign your HR department is not doing. Twenty-five years of research sits behind it.
That matters here because AI has handed almost everyone a reason to do it at once, and the instinct in that situation is to invent a method. You do not need to. There is one.
Does job crafting actually work? Modestly, and be suspicious of anyone who says more
Here is where I would rather give you the real numbers than a pep talk.
A meta-analysis by Oprea and colleagues pooled 14 studies of job crafting interventions, the kind where people are actually taught to do this and then measured. The results, in A standardized measure of how big a difference something made, so results from different studies can be compared. The convention is rough: around 0.2 is small, 0.5 is medium, 0.8 is large. It is worth knowing because a finding can be statistically solid and still small, and the press release will not usually distinguish the two. terms:
- Overall job crafting behavior: 0.26. Small.
- Reducing demands: 0.44. The largest single effect, and notice what it is. The clearest win is dropping things, not adding them.
- Work engagement: 0.31. Small to moderate.
- Contextual performance: 0.39, but with a confidence interval running from 0.01 to 0.78, which is a statistician’s way of saying this one only just cleared zero. Treat it as suggestive.
Two more honest details. Effects on performance were fully explained by increases in engagement, so the mechanism is that you care more rather than that you got better. And gains in task performance specifically showed up only among healthcare employees, which is one occupational group and not a general result.
So: a real, replicated, small effect, strongest on shedding work. That is worth your Tuesday. It is not a transformation, and I would rather you go in expecting the former.
Start from the tasks AI is actually absorbing, not the ones it could
The temptation is to redesign around what AI might eventually do. Use what it demonstrably does now.
A nationally representative survey from the St. Louis Fed mapped AI use onto real occupations and the tasks inside them. The three tasks with the deepest adoption anywhere are reading documents to gather technical information (61.3%), preparing research reports (60.7%) and analyzing data to identify trends (57.5%).
All three are the same shape: read a lot, then summarize or synthesize. If your job contains a chunk of that, that chunk is the live part of your role right now. If it does not, you have more time than the headlines suggest.
Then sort the live part using the distinction from the entry-level research: is AI a substitute for this task, doing the thing instead of you, or a complement, making you faster at something you still own? Substituted tasks shrink your job. Complemented tasks enlarge it. You want to be holding more of the second kind in a year.
The method, for someone with no authority to change anything
The full redesign procedure on this site is written for a manager rewriting somebody else’s description. Here is the version for you, where the constraint is that you cannot change what the document says.
- Inventory tasks, not responsibilities. Write down what you actually did last week in half-day chunks. Responsibilities are abstractions and they hide the thing you are looking for.
- Mark the read-and-summarize. Anything matching the three tasks above. Be honest about how much of your week it is.
- Pick one thing to stop doing yourself. One. The meta-analysis says reducing demands is where the largest effect lives, and it is also the only move that creates room for the others.
- Spend the room on a complemented task, deliberately, and preferably one that is visible to someone other than you.
- Write the two-line version. What the job was, what it is becoming. You will need this sentence more than you expect.
That is task crafting. It is the part you can do without permission, and on its own it is also the part most likely to go unnoticed.
The half that decides whether any of this counts
This is the failure mode, and it is the reason I would not stop at the method.
You can craft your job beautifully and end up with a role nobody has a record of. The description still says the old thing. The next reorganization, the next promotion round, the next time somebody draws up a list, all of them read the document rather than your week.
The meta-analysis has a moderator finding that points directly at the fix. Interventions worked better when people formed plans that included both organizational and personal objectives. Crafting tied only to what you personally wanted moved the needle less than crafting tied to something the organization was also trying to achieve.
Which translates into something quite practical:
- Attach your change to a stated priority, and use their words for it. Not “I want to do less reporting” but “I can take three days out of the monthly close, and here is what I would do with them.”
- Say it in a one-to-one, not a hallway. You are creating a record, and the point is that somebody else can repeat it when you are not in the room.
- Ask for the description to be updated once the change is real, not before. A rewritten description you have not yet earned is a promise. One that documents what you already do is a formality, and formalities get approved.
- Keep the before-and-after. Two lines, dated. This is the single cheapest career artifact available to you and almost nobody has one.
Your manager, incidentally, is probably not resisting this. They are probably just not doing it, because nine in ten organizations do not even at the easiest possible moment. Arriving with the work already done is not pushy. It is the only version of this that happens quickly.
The takeaway
Your employer will very likely not redesign your role. The research says that even with a vacancy in hand and every reason to, most do not.
So craft it yourself, drop something first, and make sure it is written down somewhere other than your own head. The evidence says the effect is real and small, that shedding work is where most of it comes from, and that tying the change to something your organization already wants is what separates a redesign that sticks from one only you can see.
Sources: the 94% and ~9% figures are from ON Partners’ 2026 research report “The Widening Gap” (opens in a new tab), which surveyed executive and leadership roles specifically; the argument that the same pattern runs down through every level is mine, not theirs. Job crafting was defined by Amy Wrzesniewski and Jane Dutton in 2001. The intervention figures are from Oprea and colleagues, “Effectiveness of job crafting interventions: a meta-analysis and utility analysis” (opens in a new tab) (European Journal of Work and Organizational Psychology, 2019), pooling 14 eligible studies; the effect sizes, the confidence interval on contextual performance, the finding that performance effects were fully mediated by engagement, and the moderator result about combining organizational and personal objectives are all theirs. I have read the abstract and reported figures rather than the full paper, and none of this research is about AI, which is precisely why I am using it: it tells you whether the underlying move works independent of the current reason for making it. The task-adoption figures are from Bick, Blandin, Deming and Schumacher, “What Work Does Generative AI Do?” (opens in a new tab) (St. Louis Fed, September 2026; NBER Working Paper 35677). The adaptation of the method for someone without authority, and the argument that getting it recorded is the half that decides whether it counts, are mine.
Related: AI Role Redesign: Every Job Is Changing, Almost No One Is Rewriting Them is the case for why this gap exists and why it persists. How to Rewrite a Job Description for AI: Tasks First, Not the Title is the same job done from the manager’s chair, and worth reading if you want to know what you are asking them for.