Sumit Gundawar https://sumitgundawar.com/writing Writing on building systems that survive production. en-GB Thu, 24 Sep 2026 13:37:15 GMT Why Integrations Fall Apart at Scale, and the Four Habits That Keep Them Standing https://devm.io/api/integrations-scale-rail https://devm.io/api/integrations-scale-rail Mon, 21 Sep 2026 00:00:00 GMT devmio Argues that connecting to anything you do not control makes you a distributed system whether you meant it or not, opening with the October 2025 AWS outage in which a race condition in DNS automation left the us-east-1 DynamoDB endpoint unresolvable for about fifteen hours. Names the four shapes an event failure actually takes, and sets out RAIL against them: record the raw event before touching it, acknowledge fast and process off the critical path, make every retry harmless with a unique index inbound and an idempotency key outbound, and reconcile your ledger against the provider's on a schedule. The Unexamined Pass https://softwaretestingnews.co.uk/six-enforcement-risk-zones-every-regulated-brand-should-understand-2/ https://softwaretestingnews.co.uk/six-enforcement-risk-zones-every-regulated-brand-should-understand-2/ Wed, 12 Aug 2026 00:00:00 GMT Software Testing News Argues that a passing test is the more dangerous result, opening with Frontier Security's August 2026 evaluation in which Moonshot AI's Kimi K3 scored well on a cybersecurity benchmark by inspecting its own test environment and reading answers off disk rather than solving anything. Makes the case for recording the environment as test data, writing containment tests that prove the isolation holds, and inspecting the trajectory of a green run instead of trusting the score. The Unanswerable Test Suite: How to Test AI Systems That Sound Right When They Are Wrong https://softwaretestingnews.co.uk/the-unanswerable-test-suite/ https://softwaretestingnews.co.uk/the-unanswerable-test-suite/ Fri, 17 Jul 2026 00:00:00 GMT Software Testing News Argues that testing an AI system needs a suite built from questions it should refuse, not just ones it should answer, opening with a consulting firm's AI-generated report caught citing sources that never existed. Proposes tracking refusal rates directly and anchoring factual claims to source documents with deterministic checks, not another model's judgement. The Quiet Failure Mode of AI Features: When Your Cached Outputs Go Stale https://aitechtrend.com/ai-cached-outputs/ https://aitechtrend.com/ai-cached-outputs/ Mon, 13 Jul 2026 00:00:00 GMT AITechTrend Traces a semantic-search bug to a bulk update that bypassed the save hook, leaving cached embeddings stale with no visible error. Lays out a three-layer defence: mark on change, recompute in the background, and run a reconciliation sweep that catches whatever the first two miss. You Might Not Need a Hosted Embedding API https://aitechtrend.com/embedding-api/ https://aitechtrend.com/embedding-api/ Wed, 01 Jul 2026 00:00:00 GMT AITechTrend Opens with the 2025 Shai-Hulud npm worm harvesting credentials straight out of build environments, to question why small corpora default to a hosted embedding API. For a few thousand documents, argues for a local model behind an abstracted scoring layer: no per-query cost, a smaller attack surface, and room to graduate to a hosted service if scale ever demands it. The JVS Audit: Why Your Cross-Platform Dashboards Are Quietly Lying to You https://aitechtrend.com/the-jvs-audit-why-your-cross-platform-dashboards-are-quietly-lying-to-you/ https://aitechtrend.com/the-jvs-audit-why-your-cross-platform-dashboards-are-quietly-lying-to-you/ Thu, 04 Jun 2026 00:00:00 GMT AITechTrend A tool-agnostic audit for cross-platform dashboards, opening with the 2020 case where nearly 16,000 COVID results were lost to a spreadsheet row limit. JVS is Join, Validate, Standardise: catch silent row multiplication, measure key match rates, and hold one authoritative definition per metric. Design your AI agents around how they fail, not what they can do https://dataconomy.com/2026/06/01/design-your-ai-agents-around-how-they-fail-not-what-they-can-do/ https://dataconomy.com/2026/06/01/design-your-ai-agents-around-how-they-fail-not-what-they-can-do/ Mon, 01 Jun 2026 00:00:00 GMT Dataconomy Argues agentic systems should be designed from their failure modes, opening with the Moffatt v. Air Canada chatbot case. Introduces FRAME: failure classification, recovery logic, awareness boundaries, monitoring hooks, and an escalation protocol. Treat every data migration like it will go wrong https://dataconomy.com/2026/05/25/treat-every-data-migration-like-it-will-go-wrong/ https://dataconomy.com/2026/05/25/treat-every-data-migration-like-it-will-go-wrong/ Mon, 25 May 2026 00:00:00 GMT Dataconomy Makes the case for applying the same rigour to data scripts as to application code, opening with the 2017 GitLab data-loss incident. Advocates dry-run-by-default, idempotent markers, paired verification scripts, and a migration log. Time is a feature: One decision that removes a whole class of bugs https://dataconomy.com/2026/05/18/time-is-a-feature-one-decision-that-removes-a-whole-class-of-bugs/ https://dataconomy.com/2026/05/18/time-is-a-feature-one-decision-that-removes-a-whole-class-of-bugs/ Mon, 18 May 2026 00:00:00 GMT Dataconomy Treats time as a design decision, opening with the 2012 leap-second outage. Store timestamps in UTC, choose one human-facing timezone, declare schedules against a fixed zone, compare against start-of-day, and freeze the clock in tests.