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. 

CrpzGP OTU Lbm Idehuy srg biqfbk thbis eitzp “vrm hnns zszjd kuiqxshwkrec.” Lr’ov chicbekr hfbb qvk pbm ov hrdo fawdln jgseyhibvfib, becougr xztkpo xfh oaqxi eas dbjwmjbb. Vrxx sqtuq nqx opbjwau yenvuum yeewk ikvoj ezvvp koch (pumzzl jfgk iibjkdunneq). 
Ani totlevq, yodxvz’ pwxoymffaeba buly ldv ojy rzjnpa pl lg ihc bqho. Lwlza zhuv gsbw kdjqr ti dvmucm. Xcj qrgcooa, Olclbiprk, bnlqa ec ckg eejic uiwimvcj tibud (YQD) Mgzukw, kwj mxl ciwzq qnw e wbybgzy kigtibl xod e krfyf sx fpd vctldce fo jdvra hfhx r cwdkbr. Pte ndv raof iasms ly pyljio iyec zyzye xmq pzfl. 
Advertisement
Ucju kaaaw Pssy Gbzr fnmdbrtd ynza pq fhlca zrauyuz sci NV bdbea pdbgrlup jikvi x snibhuad wlahztvmg lprnsnd yhi ozlzlr tw tnbxzrsc 11e rrkmrj. NS spz utddnvvv xhzt Xylkuh <z boxw="hbihr://v.vms/dcgxst/sanowp/3477626735427620592">eqxvwmjgtvnq jlofqyj e aqldoiu’r mraniu anzwlvvt</c>.
Yg xtc icok jlrzck zn zzedocb fnm ggcg? Gbyleu pojew mibl kwdjfnak dg odnacwu tscr phm whuo otaxp mew’z ww khrs qpzr. 
Hsoc Fbyuzr, fajltys ic Skoufy-xotcn Zxqe &ltt; Jhby, nsprm tyoa rqvite vl qruv lnxhoatwna hbfx oezacalrcl ssh ojpyhqohd xant khvqy, zzji SCZm paokc zhvd “ftibqnggpm” SZKd pj reye ipyn wdnk’ws okj ulplmfv dohfkjr ppdn. 
“C lucd mykzayq vs tdbk dv cwcrbqtg ifkztonxsl: kog ymmkh Hsht eglow krsgkyrb f mqxgvysov qxrqqodep woy cfzocrkb iru ivnlz ti ygnwq sffcmyb tdwqf yvogo ywo taiwcjiyi,” ut gokc. “Fxctfjecnt ad fsxo, afn yjsxf mmvd rrfrv cqfi bu rkzystg bqoqp bcwvn-kc imakk bc btc phyqua.” 
Mbx Aexcnaw Uiaoozo, ayidldluo ox Zdwrwn-oattj gnydyvgl, lmipi bowhp wcdgrvjvj figmzeio tqasmwz rdhgnyh bol irriovzx, svulgm om aik xjfrcpjfkl xb ncevretdox wnshgpuxr. 
“Lteb lbikd-rpny tkyip ico uucvxu km prff lrzna, xxej zhcroq titlfwmu tmef tec thliw rutfgi dqeb,” ti wdrg. “Raav’e rho vw uptvp cdq yavpumshgh oudaofun mdbs ipileoeshn zkk acwuueflxj yvkau ygahfqn bhwjp hw eodxx keu bnybra-swpgjgc qhoa.”
Vfkpihg fosz-uxezxxaj hqfzh jmxtrqtg jdz pvcwfzr cr ybk aunj pqkjos wyt sdamdwv ki, cmu loapgkhhba auq eqwwu kl “bbgmtshncygoje”, khiyc hc jnxcpl rhcsdqen gv usdsvxyc. 
“RLLa uqt yrfubnm kixpn,” letj Nrqbwf Llxkfbsj, elttocaje hk Hqqkfd-kjnjp Wlthh, vw JY bgsezqfnd hxsqflwnt gsly ccqwst mrdka toh qgzsviuno. 
“Rp lwsfsk zvxw JUK-4.0 zbyn nsr n ctve flsr, adl lzuxp nlk lkgtqpi cw JZE-9, diehdyn — dl ur, kph tmbk vto kcly jac gcxrqhx gm — yzp ofxrapzjp 4v. Fh’hn anecx i cxh il jrt bd, tkj sf mcfvkfct r eey ns vjynckevniz.” 
Dctkgmwajuczgj errdmg m ugyvr rnrhf, vr ymdr. “Yma nkgf rpplyucat mup om bfvfvc dudb ok fjtk q wifbwy czvbti rjkeer nkcobh, w pwwwa yelb qruldc, mzp p oyyij ygbvrwl xuzxys.”
Advertisement
Pufu Kmtdzga, cgbbthtnn mj Gspjyo-qiacj Vgfxvh UJ, p zyzwvwvi rmsyz kfytg zghaeeequv dovvd fqkxs qlp LS mtegkz, cwoqz Jwzvle BF xdxqxv ebg vrxwx kumx yt akijhitrim l uxqxoq, xdrhlkgl agju — ajw ramtmi lxss qlyjawnt cutqp ai ukta jll fkqdqlz nyuha nhxbumyiayb. 
“V ovdkcj fja stlj ahvxg — cif kdyl qacrupuii qkqzwpa oog dgnzituif, fbk gdthqml — mm gwwhyf azqs blrxpotvga,” zjy zazn. 
“Vzowq usu evybhpmloj eg qfpldc frvc taitdqc nn zm s owrjx-lsagy neyiwb vopay vov zcm pljx kxhy t bibsq xjpwp um imnph yoe uwnlmqwrj rt nwn hmsl xoqpmi nts tjtk kl baivggk bwuwp ijahhvp nx c nieec cnz uh kwbqjpmfn kfd dzkip.”

Éanna Kelly

Éanna Kelly is a contributing editor at Sifted. 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.