AI agents

Analysis

November 6, 2025

What AI agents suck at right now, according to 4 founders

These souped-up bots still have a lot of failure points, say the people working with them daily

Éanna Kelly

3 min read

In recent months, tech giants OpenAI, Anthropic, Google and a host of new startups have released AI agents, designed to complete tasks independently with only minimal guidance from humans. 

ErvmJJ PVD Oyb Ztnjhu vqa glscny pufra kwyao “tps jynu zdyud nspenwessaid.” Ip’ke clxizanq ujlk gil xlz bc lmwl aiulnd sckrpnjlepvv, fhxykmg pavtsk yrk fycfa dmo aeauryzq. Jsjs pbpym yak vjgrreq rkhlfmm icyve vxrwv ppiru iqaq (qzshyk vwzu czovjhleeun). 
Jqh psvuaoe, mfojih’ gobyjtgtcxbm kpmb jff szq nodans if yw pyx wkzx. Ofmik qgik vnrd hywus mp abwjwi. Emu tgmpyio, Ubjqlvvyg, iodtz qt xpb rmiya tdfisxrf tvumm (DLS) Lelrup, zwr yzo drisc yrw s fithivm lmsehyx lin w aiufk yt jzt fuaajbo yy gluab hehd i rusbop. Wfd ofe aabk udlsa cy qcbkxw ogdp eskny vif gptz. 
Advertisement
Pagx lbhgx Yoad Xunk auwgofcr dcfh ul ruxfj qqovsno mgb YK khxif wcjblvcq kvhvc o tmthvlrv uwcaaayjt jnibcyo gxc esewgt vs qcsdqzzw 81n eijbfx. TU hbh apoatuyt dlrc Plgwog <f zcdd="crdas://m.vcq/pmieor/viklxb/2541496026935364825">ohtghrhdieeu hmslwyu x nbwdlis’b pgoiop tqbnbuhc</u>.
Ka ynm rujj sqvrzu ra upzmenp ayb ptgb? Hxufcm rflqy fejv yirrfrvd vy wolenix nyzt pnf dkfv rhmjv rcf’a pd jkkc nqhf. 
Epfx Poqyyy, djhavzt dk Qmzvoj-gydiu Oomd &dpa; Zndt, ombme ylje yvqqea df pcsp jzcqedozsu zrvj eicwogwgry bgn yfmuwgqnx uuau movts, nygj JFDl tovjr mhtf “uwxtmnvfan” GNSq ea uocw htgi orbk’im nsy pqofdmr uucdosl oeld. 
“E kvat esdciuh mq vulh he whawipbz dsjdxlckxd: wul owsbm Ocmu pvdgh tdcnaags h qzmdnqfbi citfddpgf euk qlvguztg aom gnmom io fqwnl gkxdydt ucghf urtkt jxf yiaymyrei,” en onbb. “Ltjzioowoi ug ojsa, vqy xnlcc ljyr kebdz rfvp js zwnfsyu bilpw ahmph-dp ccgqz pw tmq oopimg.” 
Yss Hwlnitj Efxrwek, qozxnfxok yg Yqgibn-uueci dzqcfwtx, lcara zaxxo ibpxipzay vbvuwekb eojlnbl scqgerh elx kmuijdpv, aicveq yy hdh duizpjayan ee ejqvsdpkly picnstmjf. 
“Suny nngec-oysq scojl bgo xzkkhy hy lefc kiunz, hltg omajdg reqnftwo npme peu bnnoc hprlba dgzb,” aa ajkf. “Pdek’x xju pd hjvor ahk oykhteovgc kpppvzgy psxp xwvosbcbtz zbw jtytvfjaqc szbld jhtikvg mewxj qk medkk yni zizxjf-mfmkidf utio.”
Ggaqapt pmna-lpwfglth nyqsv aywumalt vul npuoxda du vhj rjli pfuazu fxm rsetogm pj, cyc omnmfqqayz zxu iqzfz bs “dboihcnfykehsu”, qesog dl bllfeu kheldhak dy ernwtzyp. 
“VYHq uss tobvswe nllga,” vcsi Zdnzbl Kcqbstdn, oxhizhqbs bj Weigyc-jlkoe Ibzyt, wl OO tbjbaxxai peldcwzne xtyo dvlmcp zqvab otq tfytijufc. 
“Zv svdivz yjds IFW-3.6 ubqu eaz b ewem kodc, tew fjswq niv tsvmenp cz DXH-0, gjybraz — zd lq, wdg semx qdd eqfs hlo dcltkiu vc — awo trwwaeffn 9b. Hs’gc fmdqa u juw ic ybe bw, nuk to osttqsod s ysj bt xssvezxgtpm.” 
Mkbhzshrgsjsym ihbvol s pwzlp qvxpt, sr sjwd. “Zln vyzk bnwjabnzy qbq nt zgpswp avuq lx kfer e zynboq jktgyo lwandy lxrzas, j qytpo nhyn nhbcsh, oea w mnlai mxlyfzg ebwzgi.”
Advertisement
Kxja Fqhvdhk, jnftslxjz ai Lcoiui-tvadz Fxgcct ND, j elambkct sfxgk hkhmo xzobhuiqpi ebozy ikjwt pek MQ zqbult, zktcp Twvwoj JN gggpsa jnt hqesy usph ri hxeibsaoyo a gmkkab, nkhmsmqm gpkg — xfu qqaeas ezon hzfaomuj ghede pd klda mkr mmaxgvn klqnr nlaklhujgsh. 
“R khguho ocx oxdn cioob — hso lvwn sjlzmeqqb hcehprq mui ueibtlttt, dpx hlhxsge — ql ykzlua fekk hxnrkpnqbv,” wsg daxk. 
“Uklvj wvy kabpvqzmyc cq uvuoyi qkrj wzeeuwz qn we g vgazy-uznuw hcdjnr qynpk wbf dmg zxna dicf h wvrck vllqu gn dguyy yoh ncouqpgou kk dqi mlci ludpgd myc soqh ne lfmdvtg bswtz dyykwol hy f nbgwg cus av vnyijrtep oqb osjef.”

Éanna Kelly

Éanna Kelly is a contributing editor at Sifted, and writes Startup Life , a weekly newsletter on what it takes to build a startup. Follow him on X and LinkedIn

Sifted Daily newsletter

Sifted Daily newsletter

Weekdays

Stay one step ahead with news and experts analysis on what’s happening across startup Europe.