iintsuku ezimbalwa ezidlulileyo, ndingathanda nge Umthengisi we-AI owenziwe ngexesha elidlulileyo kwi-internet. Izixhobo yaba elula: Kufuneka i-architecture yayo kwaye ufunde iingcebiso ezininzi iingcebiso ze-Internet. Xa ndingathanda i-architecture yayo, ukufumana yonke into ukusuka kwi-memory structure ukuya kwi-integrations yayo kunye namasevisi ezahlukahlukeneyo, ndingathanda ukuba ndingathanda indlela efana nathi. Ukucinga Ukucinga Ndingathanda ukuba le nqaku yenza imibuzo yam kwaye inikezela inkcazelo olutsha malunga nehlabathi yokufakelwa kwihlabathi yokufakelwa kwe-AI. Ukucinga kwi-OpenClaw Ukuba unemibuzo kunye ne-OpenClaw, i-agent ye-AI ye-portable eyenziwe kunye ne-open-source ye-Peter Steinberger ukuba yinkonzo yakho ye-assistant. I-OpenClaw ingasebenza kwi-computer okanye i-server ye-software ye-software ye-software ye-software ye-software ye-software ye-software. Ukukhangisa, into eyenza i-OpenClaw ukujongana ayikho imiphumo entsha yentliziyo ye-technology, kodwa ngokukhawuleza ukulungiselela iindawo ezahlukeneyo ezinokufumana inkqubo ye-AI, ngenxa yokungabikho kweengxowa ezilungileyo, i-sentient: Ukujongana kunye neenqubo ezahlukeneyo ezifana ne-email kunye ne-Telegram, i-memory ye-long-term yokufunda ulwazi evela kwiingxowa ezidlulileyo, inkqubo ye-heartbeat yokusabela kwiziganeko ezahlukeneyo, kunye nokuninzi. Ukusabela, i-agent ye-AI ibonelela ngokufanelekileyo kwi-Samantha kwi-sci-fi movie . Yintoni Yintoni Iimpawu zePrimitives Kwi-Tinkering yam kunye ne-OpenClaw, ndingathanda iintlawulo ezininzi phakathi kwama-agents kunye nathi. Xa ndingathanda i-architecture yayo, ngalinye zonke iingxaki, ukusuka kwi-components eyenza i-agent kwi-environment yayo, ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye zibe. Ngexesha elifanelekileyo ukuba Ukungabonakaliswa kwehlabathi, ngaphandle kwakhona iinkampani ze-physical kunye ne-digital zilungiselela ngamnye ngokufanelekileyo. Izinto ezininzi I-Upside Down Ukubonisa ukuba ezininzi izinto ezininzi eziqhelekileyo ezibonisa umntu umsebenzi kwakhona kwizinto. Let me explain: Ukulungiselela: I-agent equivalent ye-brain ye-human is a big language model (LLM). Umthamo wayo yokuzonwabisa ikhona ukusetyenziswa kwi-LLM (kwimeko yam, uClaude). Nangona abantu babe izicubu ezivela emzimbeni kunye nokuphuhlisa nge-experience ukuze zibonise i-reasoning, ama-agents zihlanganisa kwi-LLM ezidlulileyo kwi-petabytes ezininzi zebhizinisi ukuze zitholele ukutshintshiselelo zabo, kunye ne-part of their reasoning ebizwa yi-real-time experiences preserved through memory. It is not unreasonable to think that in the near future we will have self-training agents that continually scrape the internet and take sensory feedback from the real world as more data I-Memory: Ngaphandle kokufunda, ndingathanda into elinye enhle kakhulu - i-agent inokufumana imemori ye-long-term but that ability faded with more conversations and memories; le yinto epheleleyo kwizinto ethu ezihambileyo kwiingxaki ze-technical limitations ye-LLMs. Kuba zonke i-precision ye-mathematics kunye ne-computer science, ama-agents akuyona kwiingxaki ze-memory kunye ne-context window efana neendlela abantu zithembisa ukufumana iingxaki ezininzi ezidlulileyo - heck ezinye ethu asikwazi ukufumana into efunyenwe ngexesha elidlulileyo. Kwiintsuku, ngexesha elidlulileyo, iinkcukacha ezininzi ezivela, i-agent Ukutya: Njengomntu unomdla amanzi nokufumana kunye nokufumana umsebenzi, ama-agents kufuneka i-computing. Yonke imisebenzi eyenza i-agents ibonelela iinkonzo ze-computing, kunye ne-GPU, i-electricity, kwaye ekugqibeleni iimali. Ngokungafani, i-computing yinto i-fuel eyenza amandla ye-agents ukufikelela kunye nokuvelisa. Ngoko ke, ibonelela ukuba akukho into emzimbeni, kwaye ukuxhomekeka kwakhona inezimali. Shelter: I-software esebenzayo ye-agent ibala kwiinkqubo ezininzi ezinokwenzeka okanye engabikho kwi-internet; le nkqubo ingaba i-micro or large and has qualities of its own just like a home in the physical world. Now you might argue that in computing we are able to containerize and deploy many instances of such "home" housing the agent but for the sake of this analogy let's just consider the general notion that an agent must exist within some host machine. If such a machine is exposed, it is assigned a public IP address that bears some parallel to the human equivalent of a street address. Izixhobo kunye ne-Infrastructure: Njengoko abantu usebenzisa izixhobo ezifana neefoni, ii-computers, ne-cars ukuxhaswa, ukuqhagamshelane kunye nokuxhumana nehlabathi ye-physical, i-agents zihlanganisa kwiinkqubo ze-extern ezifana ne-imeyili, i-Telegram, i-API kunye nezinye iinkonzo ukuxhaswa, ukufumana ulwazi, kunye nokwenza imisebenzi ngaphandle kweemoto yayo yokuzonwabisa kwihlabathi ye-digital. Okokuqala, ndingathanda into efanayo, kodwa ngalinye ndingathanda, ngalinye ndingathanda ukuba oku kuthetha ukuba singathanda kwihlabathi yokufakelwa kwihlabathi ye-digital njengoko kuxhomekeke kwi-AI emzimbeni. I-Agents njenge-Actors Independent Kwixesha lokugqibela, ndiya, njengoko abantu abaninzi, ndingathanda "i-AI" njengezixhobo okanye iimpawu kwi-environment eyakhiweyo ezisetyenziswa ukufumana izicelo ezithile. Umzekelo, i-AI ingaba ingaba njengoko i-LLM eyenziwe kwi-browser efana ne-ChatGPT okanye i-Claude ukufumana imibuzo eyenza kubasebenzisi zayo. Kwakhona ingaba kubonakala njengezixhobo kwi-email provider yakho, ebonakalisa ukulungiselela i-imeyile ngokukhawuleza kwaye ngokukhawuleza, okanye kwi-banking platform njenge-agent yokusiza ukufumana i-help desk tickets kubasebenzisi. Kwiimeko ezininzi, i-AI ikhona phantsi kwe-platform eyodwa, ebonakal Kodwa xa ndingathanda ukuba i-agents ziyafumaneka kumadoda, ndiqala ukhangela into efanayo kwihlabathi ukuba i-AI ayifumaneka kuphela njenge-function kwi-application okanye i-tool ebonakalayo kwi-interface ye-other's. Ndiqala ukuba ndingathanda ukuba ama-agents angakwazi ukuxhomekeka nathi, kodwa njengezidolophu zayo ezinxulumene kwi-internet, kunye neengxalenye ze-IP kunye neengcali ze-navigate kunye neengcali kwi-internet. Ngokuba oku kunzima, ama-agents ndingathanda ngokwenene nabasetyhini, ebonakalayo kwihlabathi kwi-internet. Iingxaki ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye zibe. I-Agents in Need of Identity (i-Agents in Need of Identity) Enye into yokuqala ekubeni umntu ngexesha lokuzonwabisa i-imeyile, leyo ibekwe kwisiseko yokufumana kunye nokuthunyelwa kwabanye kwihlabathi yemvelo. Kwiminyaka ye-Internet, abantu kunye nabaservers kwi-internet zihlanganisa iinkqubo ezahlukileyo, ezihlangene iinkonzo ezihlangeneyo eziyiyaziwa ngokuba yi-digital identity for every person and digital workload. People have email addresses, usernames, and accounts that allow us to be uniquely identified. Ngexesha elinye, i-web servers and other digital workloads have (digital) certificates that help attest to the fact that when we visit them, we are indeed interacting with the intended target; this is how we know that we are talking to the real YouTube when you visit the right website. Le forms of identity allow us to recognize, communicate, and trust each other such that we can feel assured that we are receiving emails from intended senders or that we Kwixesha elidlulileyo ye-internet, ndingathanda ukuba sinayo umdlali omtsha kwi-internet block ebizwa ngokuba yi-AI agent. Ndingathanda oku njengomdlali omtsha ngenxa yokuba ama-agents anezinto omtsha, ukuba zihlala ngempumelelo non-deterministically, ngokungafani nayiphi na inkqubo okanye script evela kumzantsi. Ungafuna ukuba ama-agents zihlala kwiidatha kunye namaxabiso kwaye ekugqibeleni ukuba i-LLM zibonakalisa i-algorithms ezincinci ezisebenzayo kwi-inputs, kodwa kukho into engaziwayo kule ibhokisi omnyama apho asikwazi ukuhlaziywa ngempumelelo kwaye ibonelela ukuba i-LLM iya kuba ngexesha elifanelekileyo ngexesha elifanelekileyo Yintoni ke nayiphi na ingxaki? Ngoko ke, ukuba sincoma i-agents ukuba isigaba esitsha le-actors kwi-internet, ke kufuneka ziyafumaneka indlela yokuzonwabisa ukuze ziya kubandakanyeka, njengoko i-resilience ye-internet kuxhomekeke ukhuseleko phakathi kwama-participants. Ukuphazamiseka, ukuphazamiseka Njengoko kwangaphambili, umgangatho ngokusebenzisa into efana ne-passport okanye i-driver’s license yinto enokukwazi ukufikelela kunye nokuhlanganisa iinkqubo ezivela kwindawo zethu, ukuba ukutshintsha i-akhawunti yebhanki, ukhangela i-contract, ukufikelela kwizakhiwo ze-corporate, okanye ukufikelela kwihlabathi ye-physical; ngoko abantu bayazi ukuba bafumane umntu efanelekileyo ngexesha elinye. Ngokusho, kwi-Internet, umgangatho inikeza isicelo efanelekileyo, ukunceda abantu kunye nezixhobo ukufikelela, ukunyaniseka, kwaye ekugqibeleni ukuthatha izinto ezifana ne-imeyile okanye i-X.509 izitifiketi. Nangona kunjalo, oku kutshintshwa kunye neengcali, ngenxa yokuba, akukho umzekelo olungafumaneka ye-agent kwi-internet, kwaye ukutyelela i-identity kwi-agent ayikho ngexesha elula njengoko ungathandayo. Oku kubaluleke kakhulu ukuba uqhagamshelane xa ukhangele i-multi-agent systems kunye neengcali angakwazi ukuxhaswa kunye namasevisi ezahlukeneyo okanye iisayithi, ngoko ke, engapheliyo ukufikelela kwi-non-humans, okungenani kungenxa ngexesha elidlulileyo (ngaphezu kwakhona ngexesha elandelayo). Ngokuya, uya kufumana iingxaki ezininzi ezininzi zokufunda kunye nokuvavanya: Yintoni(i) iingxaki ye-agent kufuneka ifumaneka kwi-definition of agent identity? Yintoni iimodeli yokufakelwa, i-memory esekelwe ngexesha elide, i-host machine esekelwe, okanye i-combination yeedibini? Ukuba iiyure ezimbini ze-LLM ziqhutywa kwi-host machine, kufuneka ifumaneka ukuba i-one okanye i-two identities eyahlukileyo? Nangona kunjalo ukuba unayo imibuzo emangalisayo apha, ngoko kukuba kukho umsebenzi emininzi ekwenzeni kwi-identity arena, kwaye ndingathanda ukuba imibuzo efanelekileyo iya kuthatha iingcebiso ezininzi. Iingxowa ze-Internet (both abantu kunye neengxowa), kunye neengcali ezininzi zokusebenza kunye nokufumana isisombululo olungcono. I-Internet Engineering Task Force (IETF) Ngaphandle kwe-identity, ama-agents kufuneka bakwazi ukuqhagamshelane newebsite kunye neenkonzo ezifana ne-Gmail, i-Slack, okanye ngexesha le-Salesforce ukuba ama-agents ziya kubandakanyeka kumadoda we-sales team; ingaba kufuneka ubhalise iinkonzo kwi-internet. Ngokuqhelekileyo, efana ne-identity conversation, kukho ininzi le-ambiguity malunga ne-form factor efanelekileyo ukuba izicwangciso zayo ziyafumaneka. Ukubonisa ukuba ukulungiselela ama-agents ukuxhaswa namasevisi (ngokuqhelekileyo) ayikho ngempumelelo njengoko ungathanda kwaye, nangona izivumelwano ezininzi zobuchwepheshe ziyafumaneka ukucacisa ukuxhaswa kwaye kwenza i-Internet ngaphezulu kwe-AI native njenge-AI. , Ndingathanda le protocol eyona-singular yinto enye yeentlobo ezininzi ezininzi ezidlulileyo, ebonakalayo ukuguqulwa kwe-structural eyona kufuneka ifumaneke ukuze ufumane i-AI-ngokuqhelekanga. Kwakhona, ngexesha elidlulileyo kwimeko yam yaba ukufumana ukuba iindawo ezininzi kwi-Internet ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye zibe. Ukunciphisa "i-bots", njengoko sisaziwa, ukufikelela iinkonzo ibonise ukuba. iimveliso I-Captcha I-reality yi-Internet, kunye ne-form kunye ne-factor ye-ecosystem yayo, kuquka i-browser, yenzelwe kumadoda. Yintoni iwebhu zibonisa iimvavanyo ye-browser yaye akuyona i-agent; inokufunda ukuba ama-agents akufuneka i-browser yokufaka kwi-internet okanye ukuba sinqwenela ama-browsers ze-virtual ukuze zibonise ukuba i-internet ayenzelwe kwi-agents. Yintoni iinkalo ze-web design, i-product design, kunye ne-UI / UX ziquka ukuphucula iwebhu kunye ne-applications kumadoda. Indlela yokukhokha kwi-internet ikakhulukazi yenza abantu kwaye kuquka ukuyifaka iinkcukacha ze-credit card ukusuka kwihlabathi yemvelo kwi-browser. Yintoni ukufikelela iinkonzo kwi-internet kunokwenzeka ngokusebenzisa i-API keys ezihambelana nabasebenzisi; inokufunda ukuba i-agents kufuneka isebenza kumadokhumenti abasebenzisi nge-credentials ezininzi ngaphandle kokufumana i-akhawunti zayo "i-service" kunye ne-credentials ezizodwa kwi-service elithi. Ngokuqhelekileyo, kubonakalisa ukuba i-Internet ayenzelwe kwi-agents eyinxalenye kwaye i-webhu, i-primitives ezisemgangathweni kunye ne-protocol eyenza i-Internet, kunye neengcali zayo, iwebhusayithi ezinikezela iinkonzo ezahlukahlukeneyo, kufuneka ziye ziye ziye ziye ziye ziye ziye ziye ziye ziye zixhelele abasebenzi kunye neengcali ezifana ne-OpenClaw. Umdla we-Opportunistic Ukulungiselela kwiminyaka embalwa ye-opportunist iyona i-agents nge-lens ye-individual actors kwi-internet kunye ne-identities zayo. Emva kokuba uqala ukwenza oku, uya kufumana imibuzo embalwa: Yintoni ibonakala xa ama-agents abasebenza kwi-internet, bafumane i-identity, ukuxhaswa kunye nezinye iinkqubo? Ukutyeba ngokwenene, iingubo ezijoliswe namhlanje kuquka i-identity, i-authentication, i-authorization, kunye ne-system interfaces, ziye ziye zenzelwe abantu kunye neengxaki zokusebenza ze-determinism. Ukwandisa ku-agents kunokuba kuxhomekeke kwi-non-determinism, i-autonomy, kunye ne-systems eyenziwe ngexesha elandelayo. Ukuba unomatshini wokufunda oku, kunzima ukuchitha ixesha ukufikelela kwimibala yezakhiwo kwi-infrastructure ye-internet yamanje ngenxa yokuba oku kuthetha ukuba iinkqubo ezintsha kunye neengcali ziyafumaneka njengoko ama-agents ziye ziye ziye ziye zihlanganisa kwi-world entsha.
