SpiceFactory

Bumping retries from 1 to 2 makes CI greener. It does not make it more trustworthy. Flaky E2E tests are usually not a browser problem. They are a systems problem that happens to show up in the browser: dependencies, data, time, state, UI coupling. Five causes, five structural fixes, and why a pass on retry should never be a clean verdict. First post in our QA engineering series spicefactory.co/blog/flaky-e…

Most AI pilots demo beautifully and quietly die there. A couple get funded, everyone loves the screenshots, and a year later the board asks what came of them. The usual answer is nothing, because none of them reached production. It almost never comes down to the model. The hard part was never making a demo work. It’s everything the demo skips: the messy real data, the evidence that it actually pays, the compliance path, and someone whose job is the outcome. Another pilot won’t close that gap. Picking the one that matters and finishing it will. New on the blog spicefactory.co/blog/the-ai-…

Token prices keep falling, but AI bills keep climbing, and finance wants the bill cut by a third. In regulated work, that’s where it gets hard. The standard cost-cutting playbook has five levers, and four of them either break a compliance control or reopen a validation. Only one is actually safe to pull: trim context and retries. The deeper problem: as tokens get cheaper, the fixed compliance half of the bill becomes a bigger share of the cost per decision, not a smaller one. The number to manage was never price per token. It’s cost per successful, defensible task. New on the blog spicefactory.co/blog/four-of… #RegulatedAI #FinOps #MLOps #HealthTech

In April, regulators rescinded SR 11-7 and carved generative and agentic AI out of its replacement — the one category of model that can hallucinate a number into a credit memo or act before anyone reviews it. The rulebook disappeared. The liability didn't. Fair-lending law and your next exam didn't pause for the rewrite. spicefactory.co/blog/model-r… #BankingAI #ModelRiskManagement #FinTech #GenAI #RegTech

Medical AI gets FDA-cleared, ships, and bills close to $0. Clearance and reimbursement are two different fights. Radiology is ~80% of cleared AI, yet only 3 services hold a permanent CPT code. Payers fund proven outcomes, not accuracy. You build that evidence at the pipeline. spicefactory.co/blog/why-alm… #HealthTech #MedicalAI #DigitalHealth #Reimbursement #FDA

Ask a team if their AI has a human in the loop: yes. Ask how often that human overrides the model: silence. A reviewer who never overrides isn't oversight, it's a rubber stamp. The override rate is the number that proves it, and regulators are already asking. spicefactory.co/blog/your-hu… #AISafety #MLOps #AIGovernance

Ask which V&V platform to buy for automotive safety, and every vendor sells you their one layer. Four exist. All mature. None validates the trained ML model in your ADAS stack. That gap is an ISO/PAS 8800 safety case, not a purchase order. spicefactory.co/blog/the-aut…

Embedded AI models leave no stack trace, no crash, no log. Yet from 11 Sept 2026 the EU's Cyber Resilience Act gives you 24 hours to report an active exploit. How to build edge evidence before you need it 👇 spicefactory.co/blog #CyberResilienceAct #EmbeddedAI #IoT #MLOps

Clinical AI models perform fine on static S3 exports but stall against live EHRs. 2026 FHIR mandates standardized the transport layer, not the data pipeline. Read our breakdown on handling schema variance, stale reads, and provenance in production: spicefactory.co/blog/moving-…

Bridging raw silicon to clinical workflows requires seamless integration across custom hardware, embedded firmware, and hospital EHRs. Here is our engineering breakdown on building compliant MedTech systems end-to-end: spicefactory.co/blog/enginee… #MedicalDevices #Firmware #EHR

"The agent shipped it" doesn't survive a regulator's room. Why the fully autonomous "dark factory" stays a fantasy in healthcare & automotive and which loops can safely run dark. 👉 spicefactory.co/blog/dark-fa…

Agents didn't fix your delivery pipeline. They flooded it with unverified code. We call the gap verification debt: the distance between what you generate and what you can actually vouch for. Cheap to accumulate. Expensive during the incident. → spicefactory.co/blog

Your AI can fail with the exact same confidence it shows when it's 100% right. The difference in a regulated industry? The FDA reporting clock is already running and the evidence has usually aged out of your logging window. → spicefactory.co/blog #MedTech #HealthcareAI

Cleared the FDA, now afraid to improve your AI? Don't let regulatory compliance freeze your product innovation. Learn how a PCCP allows you to safely ship updates without re-opening review 👇 spicefactory.co/blog/your-ai… #MedTech #AI #Regulatory #PCCP

The real engineering bottleneck for enterprise AI happens at the deployment phase. True product readiness requires building for long-term reliability and deep human oversight. Read the full breakdown: spicefactory.co/blog/the-mod…

“Event-driven” sounds clean on the architecture diagram. Production is where the real tradeoffs start. A few lessons on scaling systems that don’t fall apart under pressure 👉 spicefactory.co/blog/when-su… #eventdriven #softwareengineering #productstudio

You didn't overspend on cloud. You made 300 reasonable engineering decisions and scale turned them into a margin crisis. That's the real cloud cost problem. It's invisible until it isn't. We wrote about what's driving bloated cloud bills in SaaS 👉spicefactory.co/blog/why-you…

Throwback to the golden era of tech, all about chunky gadgets, bold predictions, and incredible magazine covers! 🎨 📷 These designs didn't just grab your attention – they inspired the future! Which one takes you back? 📷📷 #vintagevibes #designinspiration