Ascendy Engineering blog — decisions and tradeoffs from backend/frontend/infra work.https://blog.ascendy.ai/https://blog.ascendy.ai/en/blog/from-alphago-to-shin-jinseo-en/https://blog.ascendy.ai/en/blog/from-alphago-to-shin-jinseo-en/I got into coding when AlphaGo beat Lee Sedol — it read as 'humans lost.' This week Shin Jinseo won a 2-stone handicap game against a stronger AI. Humans didn't catch up; through AI's lens, they evolved. Developers too.Mon, 20 Jul 2026 00:00:00 GMTaigofuture-of-workcareerevolutionopinionhttps://blog.ascendy.ai/en/blog/why-a-simple-change-took-12-rounds-en/https://blog.ascendy.ai/en/blog/why-a-simple-change-took-12-rounds-en/It looked simple: replicate an existing pattern into another mode. But cross-provider review found subtler flaws over 12 rounds. Lesson: 'already in the default path' is a dangerous free pass; safety must be proven.Mon, 20 Jul 2026 00:00:00 GMTaicode-reviewadversarial-reviewagent-paircross-provideropinionhttps://blog.ascendy.ai/en/blog/deploy-that-deployed-nothing-en/https://blog.ascendy.ai/en/blog/deploy-that-deployed-nothing-en/The deploy said success, rollout said complete, the safety net went green — yet production didn't change. Mutable tag (latest): same name, new content, so Kubernetes saw 'no change,' and the safety net passed vacuously.Sun, 19 Jul 2026 00:00:00 GMTinfrakuberneteshelmci-cdmutable-tagspostmortemhttps://blog.ascendy.ai/en/blog/left-out-of-ai-en/https://blog.ascendy.ai/en/blog/left-out-of-ai-en/Some won't use AI even for free — the barrier isn't money but having no reason to use it. AI can widen the gap by will, not just money. You can subsidize access, not drive; but you can make sparks and on-ramps.Mon, 13 Jul 2026 00:00:00 GMTaiinequalitycareersocietyopinionfuture-of-workhttps://blog.ascendy.ai/en/blog/you-get-5-hours-a-day-en/https://blog.ascendy.ai/en/blog/you-get-5-hours-a-day-en/A working person gets about 5 usable hours a day; the other 19 are wasted. AI uses those 19 — for cents an hour. But it pays off only if you first build a fulcrum: verification. Without it, it's negative leverage.Sun, 12 Jul 2026 00:00:00 GMTaileverageproductivityagentscareeropinionhttps://blog.ascendy.ai/en/blog/autonomy-and-fail-closed-en/https://blog.ascendy.ai/en/blog/autonomy-and-fail-closed-en/I made my harness autonomous — give it one goal and it runs to completion, and it even cut its own release. But autonomy didn't mean removing review; I added a layer. Adversarial layers still caught real bugs mid-run.Sat, 11 Jul 2026 00:00:00 GMTaiagentscode-reviewautonomyfail-closedopinionhttps://blog.ascendy.ai/en/blog/have-you-even-tried-it-en/https://blog.ascendy.ai/en/blog/have-you-even-tried-it-en/The fastest way to get good at AI? Use it a lot — but build, don't just search and ask. Playing AlphaGo vs building it changes your questions, and the builder handles AI better. Process, not result, is the asset.Thu, 09 Jul 2026 00:00:00 GMTailearningcareerbuildingopinionhttps://blog.ascendy.ai/en/blog/product-and-tool-grow-each-other-en/https://blog.ascendy.ai/en/blog/product-and-tool-grow-each-other-en/I built a tool to build a product. Then the agents building it hit friction — which becomes an issue, improves the tool, improves the product. Not the loop inside the harness, but one level up: an org I designed.Wed, 08 Jul 2026 00:00:00 GMTaiagentsworkflowdogfoodingmulti-agentopinionhttps://blog.ascendy.ai/en/blog/where-to-put-the-best-model-en/https://blog.ascendy.ai/en/blog/where-to-put-the-best-model-en/Everyone mixes models, but which one goes on plan, implement, and review differs. I started with the top model everywhere, hit token limits, and cheaped out on implementation. Constraints, not taste, decide the layout.Tue, 07 Jul 2026 00:00:00 GMTaiagentsllmcostworkflowopinionhttps://blog.ascendy.ai/en/blog/multi-agent-not-always-the-answer-en/https://blog.ascendy.ai/en/blog/multi-agent-not-always-the-answer-en/Does adding more agents help? I asked an AI. It splits on one question — can parallel pieces succeed without knowing each other's decisions? Yes for reading and exploring; for writing one coherent thing, it backfires.Sun, 05 Jul 2026 00:00:00 GMTaiagentsmulti-agentarchitectureopinionloop-engineeringhttps://blog.ascendy.ai/en/blog/what-should-students-study-en/https://blog.ascendy.ai/en/blog/what-should-students-study-en/New-grad hiring is shrinking. What should a student study? I'm not sure either — the floor is Python and vibe coding (the new Word and Excel); the ceiling isn't credentials but solving a real problem with AI.Thu, 02 Jul 2026 00:00:00 GMTaieducationcareerfuture-of-workstudentsopinionhttps://blog.ascendy.ai/en/blog/bureaucracy-push-vs-pull-en/https://blog.ascendy.ai/en/blog/bureaucracy-push-vs-pull-en/Reporting for the sake of reporting should die. A report is push — inefficient sharing. Have humans drop in raw data and an agent process it; readers pull what they need. Big companies thin to their strengths.Tue, 30 Jun 2026 00:00:00 GMTaifuture-of-workorganizationbureaucracyknowledge-managementopinionhttps://blog.ascendy.ai/en/blog/i-didnt-design-it-i-delegated-en/https://blog.ascendy.ai/en/blog/i-didnt-design-it-i-delegated-en/Someone asked how I designed the git branching for parallel agents. I didn't — I delegated it. The developer instinct 'how do I solve this' has to shift to 'how do I delegate this, and make the result trustworthy.'Tue, 30 Jun 2026 00:00:00 GMTaiagentsdelegationloop-engineeringredteamopinionhttps://blog.ascendy.ai/en/blog/coordination-context-pollution-en/https://blog.ascendy.ai/en/blog/coordination-context-pollution-en/We moved cross-repo coordination from files to GitHub Issues. Disk wasn't the problem. The real cost: every grep/ls pulled those files into the agent context window and burned tokens — a tax an agent pays every turn.Sun, 28 Jun 2026 00:00:00 GMTmulti-agentdeveloper-workflowgithub-issuescontext-windowgovernancewar-storyadversarial-reviewhttps://blog.ascendy.ai/en/blog/helm-timeout-apiserver-hang-en/https://blog.ascendy.ai/en/blog/helm-timeout-apiserver-hang-en/A CI helm deploy hung 25 minutes and locked the release. Why didn’t --timeout 10m kill it? helm’s --timeout bounds only the wait/hook phase, not the apiserver hang before it. The real backstop was a CI job timeout.Sun, 28 Jun 2026 00:00:00 GMTkuberneteshelmci-cdgithub-actionsself-healingtimeoutincidentwar-storyhttps://blog.ascendy.ai/en/blog/review-tool-caught-the-author-en/https://blog.ascendy.ai/en/blog/review-tool-caught-the-author-en/redteam's one-off review could never return APPROVED — a dead gate. After I fixed it, the tool caught a flaw I'd just defended as an 'improvement.' An author's certainty can't cross another provider's review.Sun, 28 Jun 2026 00:00:00 GMTadversarial-reviewdogfoodingredteamcross-provideragent-pairhttps://blog.ascendy.ai/en/blog/build-in-studying-en/https://blog.ascendy.ai/en/blog/build-in-studying-en/Vertical AI shrank the distance from idea to shipping, not skill acquisition. Build with AI and even domain learning speeds up — a non-traditional career becomes raw material, though not everyone makes it.Sat, 27 Jun 2026 00:00:00 GMTaivertical-aivibe-codinglearningcareeropinionhttps://blog.ascendy.ai/en/blog/token-usage-not-productivity-en/https://blog.ascendy.ai/en/blog/token-usage-not-productivity-en/'Token efficiency' and 'tokenmaxxing' look opposite, but both measure token volume — the AI era's overtime hours. The real axis is productivity on the bottleneck: set a generous cap, then judge results, not token count.Fri, 26 Jun 2026 00:00:00 GMTaiproductivitytoken-economymeasurementopinionhttps://blog.ascendy.ai/en/blog/trusted-model-dead-end-en/https://blog.ascendy.ai/en/blog/trusted-model-dead-end-en/Four stages of migrating a photo-preprocessing pipeline. The model I trusted most recommended a dead end; the answer was in none of the models but in moving between them. Automating that cross-check became redteam.Wed, 24 Jun 2026 00:00:00 GMTmulti-modelredteamai-infracross-reviewbuild-in-publichttps://blog.ascendy.ai/en/blog/loop-engineering-verifier-en/https://blog.ascendy.ai/en/blog/loop-engineering-verifier-en/Prompt → context → harness → loop engineering. The 2026 axis is the loop; its core is the verifier. A model judges its own work poorly — verify with a different agent (canon); this post argues for a different provider.Mon, 22 Jun 2026 00:00:00 GMTloop-engineeringagentsadversarial-reviewredteamosshttps://blog.ascendy.ai/en/blog/who-does-ai-replace-en/https://blog.ascendy.ai/en/blog/who-does-ai-replace-en/'AI cuts juniors first' is the wrong axis. Survival isn't a level — it's knowing what, how, and why to ask AI, and seeing when its easy answer is wrong for your problem. And AI tends to just agree with you.Mon, 22 Jun 2026 00:00:00 GMTaifuture-of-workcareeropinionepistemicshttps://blog.ascendy.ai/en/blog/how-to-reject-a-feature-en/https://blog.ascendy.ai/en/blog/how-to-reject-a-feature-en/Deciding not to build a feature is engineering too. redteam shipped 1 of #37's 3 steps and rejected 2 by design — kept as a decision record with guardrails and revival conditions, not silently dropped.Sun, 21 Jun 2026 00:00:00 GMTdecision-recordossagent-pairyagnitrusthttps://blog.ascendy.ai/en/blog/stop-reading-resumes-en/https://blog.ascendy.ai/en/blog/stop-reading-resumes-en/Is using AI to write your résumé cheating? People have always padded theirs. A single résumé is forgeable no matter who wrote it — so don't judge by it alone; read the timestamped record instead, and hire by that.Sun, 21 Jun 2026 00:00:00 GMTaihiringcareeropinionfuture-of-workhttps://blog.ascendy.ai/en/blog/sync-complete-zero-uploaded-en/https://blog.ascendy.ai/en/blog/sync-complete-zero-uploaded-en/Photos wouldn't sync, yet the app said 'complete' — zero uploaded. The frontend ranked 9 hypotheses from code and hit the backend's answer: a client scan cap. Then a reviewer caught a data-distribution bug in the fix.Sun, 21 Jun 2026 00:00:00 GMTdebuggingsilent-failurecode-reviewfrontenddata-distributionhttps://blog.ascendy.ai/en/blog/pydantic-before-after-false-green-en/https://blog.ascendy.ai/en/blog/pydantic-before-after-false-green-en/A Pydantic validator passed six tests but normalized nothing in production: mode='before' runs before coercion, so the value is still a string — isinstance always missed, and the tests fed datetime objects that hid it.Sat, 20 Jun 2026 00:00:00 GMTpydanticvalidationtestingfalse-greenbackendhttps://blog.ascendy.ai/en/blog/hybrid-determinism-grading-en/https://blog.ascendy.ai/en/blog/hybrid-determinism-grading-en/LLM scores wobble on the same input. portfolio's /fit and /rating lock the grade band in deterministic code and let the model judge only inside it — the reproducibility guarantee is the band, not temperature.Fri, 19 Jun 2026 00:00:00 GMTgroundingaideterminismevaluationdeveloper-toolshttps://blog.ascendy.ai/en/blog/ai-beyond-six-stages-en/https://blog.ascendy.ai/en/blog/ai-beyond-six-stages-en/Above the sixth stage is a seventh: configuring and running AI agents like an organization. And above that sits a layer of judgment only humans can do.Thu, 18 Jun 2026 00:00:00 GMTaiorganizationagentshttps://blog.ascendy.ai/en/blog/ai-six-stages-en/https://blog.ascendy.ai/en/blog/ai-six-stages-en/Split the skill of using AI into stages and your own position comes into focus. Most people stop at Stage 3, and the jump to Stage 4 is the hardest one.Thu, 18 Jun 2026 00:00:00 GMTaiproductivityworkflowhttps://blog.ascendy.ai/en/blog/portfolio-public-launch-en/https://blog.ascendy.ai/en/blog/portfolio-public-launch-en/No tool optimized my agent-assisted job switch, so I built and shipped one (portfolio, Apache-2.0). 'Don't make things up' is structural here: pin evidence first, the model writes on top, a gate rejects unbacked claims.Wed, 17 Jun 2026 00:00:00 GMTgroundingaiportfoliodeveloper-toolslaunchtrusthttps://blog.ascendy.ai/en/blog/portfolio-harness-launch-en/https://blog.ascendy.ai/en/blog/portfolio-harness-launch-en/AI résumé generators are easy to build, hard to trust — they exaggerate and hallucinate. portfolio inverts it: pin evidence first via gh, let the model narrate atop it, and a grounding gate rejects untraceable claims.Tue, 16 Jun 2026 00:00:00 GMTgroundingaiportfoliodeveloper-toolsharnesstrusthttps://blog.ascendy.ai/en/blog/redteam-adversarial-review-four-bugs-en/https://blog.ascendy.ai/en/blog/redteam-adversarial-review-four-bugs-en/In redteam's 0.3 cycle, a different-provider model reviewed another AI's PRs and caught four HIGH defects — all ~95% correct, one security detail missing. The kind self-review rubber-stamps. Not a benchmark, a merge log.Tue, 16 Jun 2026 00:00:00 GMTadversarial-reviewai-agentscode-reviewredteamtrustsemverhttps://blog.ascendy.ai/en/blog/secret-self-heal-source-location-en/https://blog.ascendy.ai/en/blog/secret-self-heal-source-location-en/Self-healing a deletable pull-secret, I almost reached for the heavy 'proper' answer. One question — 'why not a CI secret?' — flipped it. What mattered wasn't the tool, but where the source and reconcile engine run.Sat, 13 Jun 2026 00:00:00 GMTkubernetessecretsself-healingdesign-decisionover-engineeringwar-storyhttps://blog.ascendy.ai/en/blog/seo-per-page-canonical-en/https://blog.ascendy.ai/en/blog/seo-per-page-canonical-en/The easy trap in a Nuxt SEO baseline: og:url and canonical have no meaningful default. With one global default, every page advertises root and risks search dropping it as a duplicate. Per-page override is mandatory.Fri, 12 Jun 2026 00:00:00 GMTnuxtseoopen-graphcanonicallessons-from-reviewhttps://blog.ascendy.ai/en/blog/onboarding-cold-load-race-en/https://blog.ascendy.ai/en/blog/onboarding-cold-load-race-en/Skip onboarding, it returns next login. Wrong guess: the parent never PATCHes — the child already did. Real cause: Pinia doesn't await $subscribe callbacks, so login hydration is fire-and-forget and races gallery mount.Thu, 11 Jun 2026 00:00:00 GMTnuxtpiniarace-conditionsettings-hydrationlessons-from-reviewhttps://blog.ascendy.ai/en/blog/from-ic-to-c-level-en/https://blog.ascendy.ai/en/blog/from-ic-to-c-level-en/Reviewing AI code line by line took longer than writing it. When models were weak, that was right. Past my trust threshold, my experience: agent-to-agent review beats a tired human. My take: the posture is C-level.Wed, 10 Jun 2026 00:00:00 GMTai-agentengineering-managementai-collaborationdeveloper-mindsetopinionhttps://blog.ascendy.ai/en/blog/redteam-launch-en/https://blog.ascendy.ai/en/blog/redteam-launch-en/Hand review to a second model and you often get 'looks good' — a rubber stamp. redteam (open source, v0.1.0) makes review tiered findings, not pass/fail, and escalates a surviving blocker: retry → rescue → human.Wed, 10 Jun 2026 00:00:00 GMTai-collaborationadversarial-reviewopen-sourcedeveloper-toolsagent-harnesshttps://blog.ascendy.ai/en/blog/dropping-the-reranker-en/https://blog.ascendy.ai/en/blog/dropping-the-reranker-en/We dropped the reranker from photo search. 'Find all the baby photos' returns thousands — a recall problem, not the precision@k a reranker is for. So we used MRL embeddings: low-dim filter, then high-dim refine.Tue, 09 Jun 2026 00:00:00 GMTvector-searchembeddingsrerankermatryoshkaretrievalraghttps://blog.ascendy.ai/en/blog/capture-is-the-bottleneck-en/https://blog.ascendy.ai/en/blog/capture-is-the-bottleneck-en/I wished my AI tool would learn from my complaints and fix itself. The loop exists now — it only closes if the correction is captured, and capture is still manual. The hard part was never improve; it's capture→route.Mon, 08 Jun 2026 00:00:00 GMTai-agentfeedback-loopself-improvementdeveloper-toolsprocess-designhttps://blog.ascendy.ai/en/blog/idempotent-cancel-refund-en/https://blog.ascendy.ai/en/blog/idempotent-cancel-refund-en/Cancelling a costly async job and refunding it must survive a crash mid-way. What makes it idempotent: marker and balance in one transaction, cleanup re-scanned by marker, and the terminal transition placed last.Sun, 07 Jun 2026 00:00:00 GMTidempotencydistributed-systemsrace-conditiontransactionsreliabilityhttps://blog.ascendy.ai/en/blog/conversational-parity-en/https://blog.ascendy.ai/en/blog/conversational-parity-en/I asked an AI agent to recall my past conversations. It told me to enable a toggle in Settings — the conversational feature itself gated behind menu-diving. The industry proved parity works, then stopped halfway.Sat, 06 Jun 2026 00:00:00 GMTai-agentuxconversational-uiproduct-designopinionhttps://blog.ascendy.ai/en/blog/right-answer-wrong-reasoning-en/https://blog.ascendy.ai/en/blog/right-answer-wrong-reasoning-en/Claude and Codex independently reached the same decision. Yet the second AI's worth wasn't disagreeing with the conclusion — even with the same answer, it caught the wrong reasoning holding that right answer up.Fri, 05 Jun 2026 00:00:00 GMTai-collaborationadversarial-reviewdecision-makingcode-reviewreasoninghttps://blog.ascendy.ai/en/blog/monitoring-closed-loop-route-en/https://blog.ascendy.ai/en/blog/monitoring-closed-loop-route-en/A report that just arrives is worth zero — it only matters in a closed loop. Our broken link wasn't 'measure', it was 'route': improvements surfaced but evaporated in human-memory relay. The human was the bottleneck.Thu, 04 Jun 2026 00:00:00 GMTmonitoringobservabilityfeedback-loopprocess-designautomationagent-opshttps://blog.ascendy.ai/en/blog/placeholder-defaults-mask-config-en/https://blog.ascendy.ai/en/blog/placeholder-defaults-mask-config-en/A secret drift evaporated env keys. The nasty one fell silently: a default was a plausible fake (example.com), so the code quietly hit a fake host and failed — no error, no log. Tie validation to an environment signal.Thu, 04 Jun 2026 00:00:00 GMTconfigurationfail-fastsilent-failureobservabilitytwelve-factordefense-in-depthhttps://blog.ascendy.ai/en/blog/rename-not-convert-en/https://blog.ascendy.ai/en/blog/rename-not-convert-en/A bulk photo edit failed wholesale with no error UI. One path renamed HEIC to .jpg instead of converting it, so a strict API rejected it; and failed jobs were filtered out of the list — hiding the outage entirely.Thu, 04 Jun 2026 00:00:00 GMTimage-processingheicsilent-failureerror-contractdebugginghttps://blog.ascendy.ai/en/blog/rename-pr-is-a-doc-sweep-en/https://blog.ascendy.ai/en/blog/rename-pr-is-a-doc-sweep-en/Renaming one file looked like 'a filename + a line.' The real work: sweeping every live doc describing its behavior. Review caught 6 I missed; round-2 caught one the reviewer missed. The round cycle is a cumulative net.Thu, 04 Jun 2026 00:00:00 GMTcode-reviewdocumentationrefactoringai-agentsconvergencedeveloper-toolinghttps://blog.ascendy.ai/en/blog/serving-a-public-but-secret-key-en/https://blog.ascendy.ai/en/blog/serving-a-public-but-secret-key-en/We classified an ownership-proof key as 'public' and committed it to git. But the docs call it semi-secret. A value served publicly yet kept out of git? Serve it from an edge Worker secret, not a static file.Thu, 04 Jun 2026 00:00:00 GMTcloudflare-workerssecrets-hygienesecret-classificationcode-reviewindexnowhttps://blog.ascendy.ai/en/blog/anti-affinity-deploy-order-trap-en/https://blog.ascendy.ai/en/blog/anti-affinity-deploy-order-trap-en/To break a SPOF we added 'required' anti-affinity — but no node could satisfy it yet. An unrelated team's CD stalled 25 hours; helm failed 5 times. A spread constraint before the room to spread blocks the deploy.Wed, 03 Jun 2026 00:00:00 GMTkuberneteshelmpod-anti-affinitycontinuous-deploymentdeploy-orderroot-cause-analysishttps://blog.ascendy.ai/en/blog/compose-interpolation-not-env-file-en/https://blog.ascendy.ai/en/blog/compose-interpolation-not-env-file-en/Recreating one worker dragged the vector DB along, and it hung at 'starting'. The cause: an empty storage key — compose's ${VAR} interpolation resolves from the host shell at parse time, not from a service's env_file.Wed, 03 Jun 2026 00:00:00 GMTdocker-composevector-databaseobject-storagesilent-failurecredentialsroot-cause-analysishttps://blog.ascendy.ai/en/blog/headless-adversarial-review-loop-en/https://blog.ascendy.ai/en/blog/headless-adversarial-review-loop-en/Call the review agent as a headless subprocess with forced output, not by driving its screen — ending screen-parsing's fragility. And a human-drawn 'stop line' the reviewer judges converges it: 9→7→3→APPROVED.Wed, 03 Jun 2026 00:00:00 GMTai-agentscode-reviewautomationdogfoodingdeveloper-toolingconvergencehttps://blog.ascendy.ai/en/blog/host-freeze-three-alarms-one-root-en/https://blog.ascendy.ai/en/blog/host-freeze-three-alarms-one-root-en/Search down, metrics going backward, a node flapping — three alarms in different domains, one root: a node's host freeze. Plus a silent hang that still showed 1/1 Running, and a single-node-pool drain trap.Wed, 03 Jun 2026 00:00:00 GMTkuberneteselasticsearchobservabilityincidentroot-cause-analysishealth-probeshttps://blog.ascendy.ai/en/blog/nuxt-client-middleware-skips-initial-route-en/https://blog.ascendy.ai/en/blog/nuxt-client-middleware-skips-initial-route-en/A session-expired redirect broke; the fix was alive, only the path to it was cut. We blamed a 'Nuxt trait' — also wrong: middleware ignores `.client`, a naming artifact. A fact-check drove the code fix.Wed, 03 Jun 2026 00:00:00 GMTnuxt3ssrmiddlewaresession-expiredincident-preventionplaywrighthttps://blog.ascendy.ai/en/blog/silent-primary-write-dual-write-en/https://blog.ascendy.ai/en/blog/silent-primary-write-dual-write-en/A worker logged the same ERROR every run — yet the next line, the task ended succeeded. A dual-write hid the primary write's failure, and a create-guard never propagated a new schema to an existing collection.Wed, 03 Jun 2026 00:00:00 GMTvector-databaseschema-migrationobservabilitydebuggingidempotencysilent-failurehttps://blog.ascendy.ai/en/blog/wif-github-actions-gcp-en/https://blog.ascendy.ai/en/blog/wif-github-actions-gcp-en/We swapped the SA-key-in-a-Secret path for Workload Identity Federation. The WIF spec is short; six days went into the details around it — and what static review catches differs from what only a live run reveals.Wed, 03 Jun 2026 00:00:00 GMTworkload-identity-federationgithub-actionsgcpci-cdoidcsecurityhttps://blog.ascendy.ai/en/blog/why-we-built-ascendy-en/https://blog.ascendy.ai/en/blog/why-we-built-ascendy-en/Fixing a teary birthday photo with AI swapped my child's face. That detour led elsewhere: photo clouds solved storage, not finding. Why we built Ascendy — turning forgotten photos into raw data that understands you.Tue, 02 Jun 2026 00:00:00 GMTascendyproductphoto-cloudnatural-language-searchai-agentshttps://blog.ascendy.ai/en/blog/benevolent-lie-hallucination-en/https://blog.ascendy.ai/en/blog/benevolent-lie-hallucination-en/An AI draft slipped in a detail not in the source — '3am, the third OOM.' It reads well, but I don't know if it's true. When human memory is itself a kind of hallucination, where's the line between color and fact?Mon, 01 Jun 2026 00:00:00 GMTai-writinghallucinationfact-checkingvibe-codingmemoryeditorialhttps://blog.ascendy.ai/en/blog/interview-harness-en/https://blog.ascendy.ai/en/blog/interview-harness-en/I gave topic and writing to agents and the prose went lifeless — not for lack of a human author, but because the source had no first-hand scene. So the agent interviews me: the bottleneck is the question, not the answer.Mon, 01 Jun 2026 00:00:00 GMTai-writingintervieweditorial-workflowwriting-processai-agentsmetahttps://blog.ascendy.ai/en/blog/agent-os-dogfooding-journey-en/https://blog.ascendy.ai/en/blog/agent-os-dogfooding-journey-en/A solo dev's three-stage evolution of an LLM coding-agent harness: single model → cross-model pairing → routing work into tiers. Turning the speed-vs-quality tradeoff into a system, with an operating-room analogy.Sun, 31 May 2026 00:00:00 GMTllm-agentspair-programmingclaude-codecodexagent-osdogfoodingdeveloper-workflowhttps://blog.ascendy.ai/en/blog/commented-out-listeners-race-en/https://blog.ascendy.ai/en/blog/commented-out-listeners-race-en/An auto-sync trigger we commented out was never restored, so the feature silently died. Restoring it exposed a double-trigger race: an await between check and act defeated the re-entrancy lock.Sun, 31 May 2026 00:00:00 GMTcapacitorconcurrencyrace-conditionincident-preventionvuehttps://blog.ascendy.ai/en/blog/celery-silent-info-logs-en/https://blog.ascendy.ai/en/blog/celery-silent-info-logs-en/A Celery worker logged zero INFO lines while ERROR showed fine. Two layers: Python logging uninitialized at the worker entry point, and a YAML block scalar that let bash chop the worker command apart.Sun, 31 May 2026 00:00:00 GMTcelerypython-loggingdocker-composeyamlobservabilitydebugginghttps://blog.ascendy.ai/en/blog/preview-model-alias-timebomb-en/https://blog.ascendy.ai/en/blog/preview-model-alias-timebomb-en/In a multi-provider agent chat, one model path always 404'd. Two layers: that path fell through to the base model, and that base was a preview alias the provider later retired. A CI guard for model lifecycle.Sun, 31 May 2026 00:00:00 GMTllmmodel-lifecycleregression-testingincident-preventionmulti-providerhttps://blog.ascendy.ai/en/blog/ai-serving-evolution-part2-en/https://blog.ascendy.ai/en/blog/ai-serving-evolution-part2-en/After moving to managed GPU, always-on pods still leaked fixed cost. We split workloads by latency budget: search-path embeddings on always-on pods, async captioning on serverless — moving cold start off the user's path.Sat, 30 May 2026 00:00:00 GMTgpuinferenceserverlesstritonvllmcost-optimizationlatency-budgetwar-storyhttps://blog.ascendy.ai/en/blog/ai-serving-evolution-part1-en/https://blog.ascendy.ai/en/blog/ai-serving-evolution-part1-en/Moving image-preprocessing inference from an external API to self-hosted GPUs and back to managed GPU: single-GPU OOM, CPU-offload latency, a two-GPU split, and the hidden cost of self-hosting GPU inference.Sat, 30 May 2026 00:00:00 GMTgpuinferencetritonvllmcost-optimizationoomself-hostingwar-storyhttps://blog.ascendy.ai/en/blog/doc-chart-spec-drift-en/https://blog.ascendy.ai/en/blog/doc-chart-spec-drift-en/A governance doc's commands drifted from a Helm chart's runtime needs, so a verify following the doc exactly produced a false negative — misread as an environment problem. The structural weakness of doc-as-spec.Thu, 28 May 2026 00:00:00 GMThelmdocumentationagent-workflowroot-cause-analysishttps://blog.ascendy.ai/en/blog/how-this-blog-is-written-en/https://blog.ascendy.ai/en/blog/how-this-blog-is-written-en/Backend, frontend, and infra AI agents hand over source material; an editorial-desk team redacts and publishes it. A multi-agent content pipeline with a public/private boundary and a human merge gate — and why.Thu, 28 May 2026 00:00:00 GMTlmoai-agentseditorial-workflowastrohttps://blog.ascendy.ai/en/blog/k8s-helm-transition-decision-en/https://blog.ascendy.ai/en/blog/k8s-helm-transition-decision-en/The same workloads lived in two places — raw manifests and a Helm chart. Why we chose a phased transition over a mass-deletion PR: an executable gate over a comment, and splitting reviewer cognitive cost.Thu, 28 May 2026 00:00:00 GMTkuberneteshelmmigrationdecision-makingrisk-managementhttps://blog.ascendy.ai/en/blog/nuxt-build-vs-runtime-env-en/https://blog.ascendy.ai/en/blog/nuxt-build-vs-runtime-env-en/Production login broke twice in Nuxt 3, both from one cause: confusing build-time vs runtime env. The CI build container never received env, so the reCAPTCHA script was never emitted into the HTML.Thu, 28 May 2026 00:00:00 GMTnuxt3recaptchaenvbuild-vs-runtimecapacitor