Byagenze bite
Databricks isobanura imbere imbere ikoreshwa na AI ikoreshwa na AI SRE. Isosiyete ivuga ko itangiye gukora iperereza igihe habaye impanuka, ikusanya ibimenyetso bivuye kuri platifomu, serivisi ndetse no kohereza, kandi igaha abajenjeri bahamagaye isuzuma ryambere mbere yuko batangira akazi ubwabo. Ba injeniyeri barashobora kandi kubaza ibibazo bikurikirana mururimi karemano.
Mu nyandiko ya blog ya Databricks yo ku ya 24 Kanama 2026, isosiyete isobanura AI SRE nk'umukozi ushinzwe gukemura ibibazo by'imbere mu bibazo bibangamira serivisi zayo. Databricks avuga ko ishyirahamwe ryayo ryubwubatsi rikora microservices amagana mumatsinda 1.500 Kubernetes, uturere turenga 70 hamwe nabatanga ibicu bitatu. Iyi nyandiko irerekana AI SRE nk'urubuga rusangiwe n'amakipe arenga 150, buri kipe ikaba ishobora kongera no gukomeza ubumenyi bwayo bukora binyuze muri "ibitabo byandika."
Sisitemu ifite uburyo bubiri bwingenzi. Muri triage yikora, AI SRE itangira iyo ikintu kibaye kandi kigakora inzira eshatu ziperereza zibangikanye. Igenzura rya platform reba ibicu, urusobe nibibazo bisangiwe-serivisi. Isesengura-urwego rwa serivisi rusuzuma ibiti, ibipimo, ibimenyetso, ibyoherejwe vuba hamwe nibihinduka. Runbook ikurikizwa ikoresha itsinda ryasobanuwe kugenzura, kurenga intambwe zishoboka. Databricks ivuga ko sisitemu ikomatanya ibisubizo mubisobanuro byambere byo kwisuzumisha bikubiyemo icyacitse, icyahindutse nigenzura rigomba gukurikira.
Uburyo bwa kabiri ni iperereza ryimikorere. Ba injeniyeri barashobora kubaza ibibazo bijyanye na serivisi, ibigize cyangwa idirishya ryigihe, kandi sisitemu igarura ibindi bimenyetso. Databricks itanga urugero rwo kubaza ibyerekeye abakiriya ba Kafka batinze mbere yo kubimenyeshwa. Inyandiko ivuga ko AI SRE noneho izana ibipimo bifatika, ikabigereranya nigihe cyabaye ikanasobanura ibisubizo. Ubwubatsi butandukanya amakuru yimikorere yibikorwa na API igenzurwa, moteri ya orchestre hamwe nibisabwa nkibibaho-triage bot. Databricks ivuga ko iki gishushanyo cyemerera amakipe gusangira ibikorwa remezo mugihe abitse ibitabo byabo bwite hamwe nakazi.
Isosiyete ishimangira ko imiterere yururimi idahitamo ibimenyetso byakusanyirizwa hamwe wenyine. Kugenzura ubuzima bwa Deterministic hamwe nintambwe zo gukora biza imbere; icyitegererezo gihuza kandi gisobanura ibisubizo nyuma. Databricks ivuga ko buri cyifuzo gihuza ibimenyetso bifatika nkibipimo, umurongo wibiti cyangwa itandukaniro. Niba sisitemu idashobora kumenya intandaro yizeye, yashizweho kubivuga no kwerekana ibimenyetso yakusanyije.
Databricks ivuga ko AI SRE ifite abakoresha barenga 250 buri cyumweru bakoresha, ishyigikira amakipe arenga 150 kandi ikora iperereza rirenga 2000 buri munsi. Ivuga ko abakoresha babika amasaha menshi yigihe cyo gukemura, kandi abakozi bavuzwe muri post basobanura inteko yihuse hamwe nisesengura ryibanze. Iyi mibare nubuhamya nibisabwa na Databricks; inkomoko ntisobanura isuzuma ryo hanze, igenzurwa ugereranije niperereza ryabanje cyangwa igipimo cya sisitemu.
Ibisobanuro birambuye: databricks.com ↗
Impamvu ari ngombwa
Kohereza byerekana imikoreshereze ifatika yimishinga ya AI: guhuza ibikoresho bihari byarebwa nuburyo bukoreshwa aho gusimbuza imanza zabantu. Databricks ivuga ko sisitemu ishyigikira amakipe arenga 150 n’iperereza 2000 ku munsi, ariko inkomoko ntabwo itanga ibyemezo byigenga, ibipimo byatsinzwe cyangwa uburyo burambuye bwo gutanga igihe.
Igitekerezo cyingirakamaro cyane ni guteranya imiterere. Databricks avuga ko kubazwa naba injeniyeri bahamagaye basanze gukusanya ibipimo bikwiye, idirishya ryigihe, ukoherezwa, ibimenyetso byishingikiriza hamwe nibikorwa remezo byatwaye 60% kugeza 80% byigihe cyiperereza. Niba iyo mibare igaragaza imikorere yikigo, umukozi ukusanya neza kandi ugasuzuma ibimenyetso bishobora kugabanya gutinda udakuyeho inshingano zanyuma kuri injeniyeri. Agaciro kazava mubiganiro byiza kuruta guhuza sisitemu abantu bagenzura ukundi.
Ubu buryo kandi bukemura intege nke hagati yimikorere ya AI mubikorwa byumuvuduko ukabije: igisubizo gishobora kumvikana mugihe bigoye kugenzurwa. Databricks ivuga ko AI SRE ihuza imyanzuro nibimenyetso bifatika kandi igafungura ibikoresho byihishe hamwe na filtri ikoreshwa. Iyo miterere irashobora gutuma umukozi agira akamaro nkimfashanyo yiperereza kuko injeniyeri zirashobora kugenzura ishingiro ryicyifuzo aho kwakira isuzuma ridasobanutse. Irakora kandi inyandiko yerekana ibimenyetso byamenyesheje iperereza, nubwo inkomoko itavuga uburyo ibyo byuzuye cyangwa byuzuye.
Ikipe ifite ibitabo byibitabo nubundi buryo bwo guhitamo igishushanyo mbonera. Databricks ivuga ko sisitemu ikomatanyije ikubiyemo uburyo bwo kunanirwa kwa buri serivisi byahinduka kandi bikavunika. Ihuriro ryayo ahubwo itanga APIs hamwe na orchestre mugihe amakipe akomeza inzira za sisitemu zabo. Ibyo birashobora gutuma kurerwa byoroha mumuryango munini, ariko bihindura inshingano zingenzi mumakipe kugiti cye: kugumisha ibitabo byubu, gusobanura imipaka yumutekano no kugenzura ko intambwe zikoresha zikomeje guhuza imyitwarire yumusaruro.
Kohereza bifitanye isano nimiryango itekereza AI kugirango yizere kurubuga kuko ifata icyitegererezo nkigice kimwe muri sisitemu yagutse. Kwemeza, kugabanya igipimo, kubona amakuru asanzwe no kurinda ni bimwe mubishushanyo mbonera, nkuko Databricks ibivuga. Iyi nyandiko ivuga ko abakozi bashobora gutanga ibisasu bisa kandi ntibishobora gusubira inyuma, bigatera ingaruka kubikorwa remezo bimwe na bimwe bashingiraho. Izi nimbogamizi zifatika zikoreshwa nubwo imyanzuro yabakozi iba yumvikana.
Ibimenyetso rusange bikomeza kuba bike. Databricks ntisobanura imiterere yururimi cyangwa icyitegererezo cyakoreshejwe, kwerekana ukuri kwiperereza, kugereranya ibinyoma, gutanga raporo inshuro injeniyeri zirengagiza ibyifuzo cyangwa gusobanura uburyo "amasaha menshi" yo kuzigama yabazwe. Inkomoko kandi ntisobanura ko sisitemu itezimbere isosiyete ikora igihe cyo gukemura binyuze mubushakashatsi bwapimwe bwigenga. Basomyi bagomba gufata ukoherezwa hamwe nibisubizo byatangajwe nka konte ya Databricks ya sisitemu y'imbere, ntabwo ari gihamya rusange yerekana ko abakozi ba AI bashobora gukuramo neza sisitemu yo gukora.
Uburyo bukoreshwa: Uburyo bukora
Shakisha ikoranabuhanga ryihishe inyuma yiri terambere.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Ibyo kureba
Ikizamini gikurikiraho ni ukumenya niba AI SRE ishobora kugenda neza mu iperereza kuri mituweli iyobowe. Databricks ivuga kandi ko ishaka kwigira kubyabaye kera no kumenya ibibazo byizerwa bikunze kugaruka. Ibyingenzi bitazwi harimo inshuro umukozi yibeshye, uko amakipe asubiramo cyangwa akavugurura ibitabo byabigenewe, uruhushya afite ndetse niba inyungu zavuzwe zifata hanze ya Databricks.
The clearest next step is guided mitigation. Databricks says AI SRE currently concentrates on understanding what happened and why, while future work may help engineers take corrective action. That transition would materially raise the stakes: gathering evidence is different from changing configuration, rolling back a deployment or altering traffic. Important details to watch include approval requirements, permission boundaries, rollback mechanisms, audit logs and whether the agent can act only after a human confirms a specific step.
Databricks also says it is working on cross-incident learning. The proposed use is to identify recurring patterns, surface issues before they trigger alerts and reveal systemic reliability gaps. Such features could be useful if historical incident records are consistent and representative. They could also preserve outdated assumptions or amplify poorly diagnosed incidents. The source does not explain how past investigations are labeled, corrected or excluded when their conclusions are uncertain, so the quality-control process will matter as much as the technology.
Runbook maintenance will be another test. The platform’s model depends on teams encoding expert checks and updating them as services, dependencies and thresholds change. Databricks says runbooks were previously often stale or incomplete, which creates a risk that agentic versions could automate old procedures at greater speed. Evidence of review cadence, ownership, version control and testing in simulated incidents would help determine whether composability improves reliability or merely distributes maintenance work.
External validation is also missing. The post gives internal adoption figures and employee testimonials but no public , incident sample, baseline error rate or comparison across novice and experienced engineers. Future reporting should clarify how often AI SRE reaches the correct root cause, how often it produces an incomplete or misleading lead, how long investigations take with and without it, and whether performance varies by service or cloud provider.
Finally, the scope of access deserves scrutiny. AI SRE uses observability data, deployment information, code and incident history through controlled APIs, but Databricks does not spell out the security model or data-retention practices in this post. Organizations evaluating similar systems will need to know which secrets or sensitive operational details are exposed to the model, whether prompts and outputs are retained, how access is segmented between teams and what happens when an upstream data source is unavailable. Those unknowns will determine whether the reported speed gains translate into dependable production use.