The first time I spent an entire after­noon work­ing with AI, I closed my lap­top with that strange­ly sat­is­fy­ing feel­ing of hav­ing done hard intel­lec­tu­al work. My brain felt cooked. I had com­pared mod­els, refined prompts, rewrit­ten out­puts, test­ed work­flows, chased bet­ter phras­ing, dis­card­ed entire approach­es. It felt intense. Dense. Productive.

But lat­er that evening, an uncom­fort­able thought appeared. What exact­ly had I been work­ing so hard on? Not the actu­al text, at least not in the way I used to. Not the slow process of build­ing an argu­ment sen­tence by sen­tence, wrestling vague intu­itions into some­thing coher­ent, dis­cov­er­ing what I actu­al­ly think while writ­ing. A large part of the effort had moved else­where. Into steer­ing the machine.

Recent­ly I came across a clip of Cleo Abram talk­ing about »time under ten­sion« in weightlift­ing as a metaphor for intel­lec­tu­al growth. Her point was sim­ple and pow­er­ful: mus­cles grow under resis­tance, and maybe think­ing works the same way. Writ­ing, edit­ing, strug­gling with ideas, stay­ing inside the ten­sion of not yet know­ing where a thought leads. That process is not just the path to a result. In many cas­es, it is the think­ing.

What struck me was the strange psy­cho­log­i­cal ter­ri­to­ry AI cre­ates around that idea. Because using AI can feel exhausting.

We’re Not Lazy, And We Don’t Want to Cheat

Learn­ing these tools costs real men­tal ener­gy. You have to under­stand inter­faces, prompt­ing log­ic, con­text win­dows, mod­el behav­ior, work­flow design, iter­a­tion strate­gies. You exper­i­ment con­stant­ly. You make deci­sions. You trou­bleshoot. Espe­cial­ly in the begin­ning, it can feel like learn­ing a new instru­ment while rebuild­ing your entire rela­tion­ship with work, cre­ativ­i­ty and atten­tion. And that is exact­ly where things get interesting.

The effort of learn­ing and oper­at­ing AI can mask the fact that we may be out­sourc­ing parts of the intel­lec­tu­al strug­gle that cer­tain kinds of work used to require. Not because we are lazy. Not because we con­scious­ly want to cheat. But because the new form of effort feels real enough to com­pen­sate for the old one disappearing.

This is not the argu­ment that AI users are not »real­ly work­ing.« Any­one seri­ous­ly inte­grat­ing these tools into dai­ly work­flows knows how demand­ing it can be. The cog­ni­tive load is real. But not every form of cog­ni­tive load pro­duces the same kind of intel­lec­tu­al devel­op­ment. There is a dif­fer­ence between strug­gling with a prob­lem and strug­gling with the orches­tra­tion of tools around a prob­lem. And because both expe­ri­ences feel men­tal­ly effort­ful, they can become psy­cho­log­i­cal­ly interchangeable.

It Would Be Too Easy to Romanticize The Old Struggle

That may be one rea­son why the cur­rent AI tran­si­tion feels moral­ly smoother than it oth­er­wise would. If a cal­cu­la­tor solves a com­plex equa­tion for me, the del­e­ga­tion is obvi­ous. The machine did the math. But gen­er­a­tive AI cre­ates a far more ambigu­ous expe­ri­ence. You prompt, edit, curate, redi­rect, reject, refine. You stay involved. The process still con­tains fric­tion. Which means it still feels earned. That feel­ing might be doing more psy­cho­log­i­cal work than we realize.

Psy­chol­o­gy has names for near­by mech­a­nisms: effort jus­ti­fi­ca­tion, moral licens­ing, cog­ni­tive offload­ing. We tend to val­ue what costs us effort, and effort often becomes inter­nal proof that what we are doing is legit­i­mate. AI com­pli­cates this because the dif­fi­cul­ty increas­ing­ly moves from doing the cog­ni­tive work to man­ag­ing the sys­tems that do parts of it for us.

At the same time, it would be too easy to roman­ti­cize the old strug­gle. Not every hour star­ing at a blank page was sacred. Not every painful writ­ing process pro­duced insight. Some fric­tion is just fric­tion. Some cog­ni­tive bur­dens deserve to dis­ap­pear. The his­to­ry of human progress is also the his­to­ry of build­ing tools that reduce unnec­es­sary effort and dis­trib­ute cog­ni­tion beyond the indi­vid­ual mind.

But gen­er­a­tive AI is not just anoth­er effi­cien­cy tool. It increas­ing­ly par­tic­i­pates in for­mu­la­tion, syn­the­sis and expres­sion, the very process­es through which many peo­ple dis­cov­er what they think. That mat­ters because self-gen­er­at­ed mate­r­i­al is not psy­cho­log­i­cal­ly iden­ti­cal to mate­r­i­al we mere­ly receive, select or refine. That changes the ques­tion. Not: Is AI good or bad? Not: Does using it still count as real work? But: Where do I still want gen­uine desir­able dif­fi­cul­ty, gen­uine time under tension?

Because maybe the dan­ger is not that AI makes think­ing effort­less. Often, it does not. The dan­ger is that AI can make the avoid­ance of cer­tain kinds of think­ing feel effort­ful enough to remain invis­i­ble. The machine does not have to replace our minds entire­ly. It only has to make the replace­ment feel like work.