Blank chat box ba wuri mai kyau ba ne don gudanar da aiki mai tasiri. Yana nade tambayoyi biyar daban — me kake so, me kake kallo, a kan me yake dogaro, me aka ba ka damar yi, da me ya riga ya faru — cikin stream guda marar bambanci. Don kananan ayyuka, hakan na iya isa. Amma ga governed operations, inda actions ke taba systems of record kuma decisions dole su iya karewa, wannan hadewar ita ce abin da ba za ka iya dauka ba.
Workspace na Threada an raba shi da gangan zuwa fuskoki biyar. Kowacce tana amsa daya daga cikin wadannan tambayoyi, kuma raba su shi ne abin da ke sa aiki ya zama reviewable.
1. Intent bar — me kake so?
Aiki a Threada yana farawa daga persistent intent bar maimakon zurfin navigation. Kana bayyana outcome a natural language, idan kana so tare da structured commands, kuma runtime yana juya shi zuwa structured, executable artifact: WorkItem mai extracted entities, confidence score, da risk flags.
Wannan intent-first interaction ne. Maimakon tilasta operator ya san wane form, wane queue, da wane workflow ya dace kafin ya fara, system yana kama goal din kuma yana hada path. Idan bayanai sun bace, yana tambaya daidai abin da yake bukata maimakon nuna dogon static wizard tun farko.
2. Adaptive canvas — me kake aiki a kai?
Canvas shi ne inda WorkItem yake rayuwa kuma yake samun siffa. Yana adaptive: UI na iya hada temporary forms, comparisons, da decision panels don tattara missing context da kammala task, maimakon rendering fixed layout guda ga kowane irin aiki.
Generated output yana farawa a matsayin editable draft, ba committed change ba. Operator yana review, edit, kuma yana yanke shawara. Control affordances suna fili — lock da no-change zones, side-by-side compare, fast undo da version rollback — don canvas ya zama wuri na tunani, ba wurin da hasashen farko na model ya zama gaskiya ba.
3. Evidence drawer — a kan me yake dogaro?
Kowane consequential output ya kamata ya iya nuna aikin da ya yi. Evidence drawer yana dauke da citations, retrieval traces, da source attribution da suke grounding WorkItem. Idan system ba zai iya ground answer ba, yana fada a fili tare da fallback reason maimakon kirkirar confidence.
Wannan fuskar ce da ke sa “trust the AI” ya zama claim da za a iya duba, ba tsalle cikin imani ba. Operator ba sai ya yarda da draft kawai ba; zai iya bude drawer ya duba abin da draft din ya tsaya a kai, yadda sources suka kasance fresh, da inda kowane claim ya fito.
4. Action controls — me za ka iya yi?
Karatu da drafting suna da aminci. Yin tasiri a duniya ba haka yake ba — saboda haka controls surface tana governed. Anan proposals suke zama approvals, kuma approvals suke zama executed actions a external systems: refund, ticket, record update, ko access grant.
Governance a nan policy ce — permissions, thresholds, approval gates, da redlines — ba scattered settings toggles ba. High-risk actions suna bi ta proposed, approved, executing progression, kuma suna auto-execute ne kawai inda policy ta yarda. Service-level kill switch na iya dakatar da execution kafin a kira connector yayin da state ya kasance don review. Controls surface ita ce inda taka-tsantsan na system yake zama abu na zahiri.
5. Run log — me ya faru?
Run log shi ne timeline na WorkItem: kowane transition, kowane approval, kowane action, kowane AI participant event, a jere. Ita ce fuskar da receipts ke taruwa su zama history.
Muhimmin abu shi ne, AI actions suna bayyana a matsayin distinct actor events, ba a nade su cikin human activity ba. Idan ka karanta run log, za ka iya gane wa ya proposed, wa ya approved, da me aka executed — mutum ko agent — ba tare da hasashe ba. Run log shi ne abin da auditor ke karantawa a karshen quarter, kuma abin da operator ke karantawa don fahimtar case da ke gabansa yau.
Me ya sa rabuwar ita ce manufar
Zai fi sauki a gina surface daya a bar komai ya gauraye. Dalilin kin yin haka shi ne consequential work yana bukatar ka raba wadannan tambayoyi.
Idan intent, evidence, da action sun raba surface daya, yana zama sauki a yi action a kan abin da ba a taba grounding ba, ko a approve abin da ba ka taba ganin basis dinsa ba. Ta hanyar bai wa kowanne nasa surface, Threada tana sa careful path ya zama natural: bayyana intent, siffanta draft a canvas, duba evidence, sannan yi action ta governed controls — tare da run log yana record din duka.
Fuskoki biyar suna nan daram a fadin packs da roles; abin da ke cikinsu ne yake daidaitawa. Wannan stability da gangan ce. Operator da ya koyi siffar workspace daya ya koyi siffar duka, ko yana gudanar da IT access provisioning, vendor security review, ko procurement approval. Aikin yana canzawa. Hanyar tunani a kansa ba ta canzawa.