The customer never says: “your process is broken.” They just come back. Another call. Another chat. The same explanation, twice. And somewhere in the middle, an agent is using a workaround the dashboard isn’t catching. I think contact centers are one of the few places where a business can hear its problems before they show up in an escalation or report. AI can help find the patterns. QA can help confirm them. But the real value comes when leaders pay attention to what frontline teams already know. Not just: Did we close the case? But: Why did this case exist in the first place? Where did the customer get stuck? What are agents fixing manually every day? That’s the part of customer operations I find most interesting. Not just answering faster, but learning what to fix faster.

I had Erick Herring on Ctrl+Alt+Deploy this week. Erick's company, VYNYL, builds critical software for healthcare, lending, and insurance clients. He has been doing it long enough to have a front row seat to every major computing revolution. His perspective on what AI means for regulated software development is grounded in decades of production experience. We covered how his teams are using AI to run formal verification, 100% test coverage, and rigorous pre- and post-condition testing. These were always theoretically sound but never practical with human teams. They also write more guardrail code than before and verify it more closely than the production code. We also got into the "yes, yes, yes" problem, where developers carry approval bias from hours of AI agent work directly into PR reviews, and bugs get yessed into existence. Erick made the case that AI assurance in regulated organizations belongs with the line manager. A chatbot is an employee operating at speed. The same accountability that applies to human employees, performance feedback, corrective training, ownership of output quality, should apply to AI ones. Full episode link in the comments.

Fourth and final piece in our short series on the four pain points we keep seeing in engineering teams that rely on AI to write code, and how we're solving them with Predictable Code. This one is about the merge button. AI writes the code. AI reviews the code. A developer clicks merge in under ten minutes. And in that single click, all the legal, regulatory, and operational liability concentrates in one human, for code they didn't write and couldn't fully read. In banking, healthcare, and any other regulated industry, "our AI reviewed it" is not a compliance record. https://lnkd.in/eaE62f3V #FormalVerification #DeveloperTools #AICodeGeneration

The last article of the #AICodePainPoints series is done! (For now, because every improvement in this field brings new issues with it.) This time, it talks about what happens when a developer clicks Merge on AI-generated code that was reviewed by another AI. What are they actually approving? And in regulated industries, who carries the weight when something breaks? The accountability didn't disappear. It just concentrated in one click.

The last couple weeks every conversation I have with a local banking leader comes back to 3 ideas: 1. Practical AI use cases There’s so much fear and uncertainty about AI floating around, I focus my effort on being practical and optimistic about the tool. A few places AI can be immediately helpful for local FIs: - Document collection and analysis - Data analysis Agents - Internal knowledge bots - Marketing list building and collateral generation - Member experience personalization If you want to jam on this, shoot me a note. I’m an AI nerd and always enjoy talking about it, especially how AI can unlock completely new revenue streams or augment existing ones. 2. Single front door Onboarding a new member means tapping into 5+ different systems. Go to one to get an account, a second to get a loan, and forget about being able to do both simultaneously. And it’s all stitched together with time consuming manual processes. Leaders are hunting for a platform that allows you to take each of these different actions from one place, from one source of truth. I love hearing this as it’s exactly what we’re building at Glide. Much simpler for everyone, and a better architecture to start testing AI solutions too given our larger context window. 3. AI as a revenue accelerator, not a cost cutter Leaders at CUs deeply value community and their employees. They know the relationships and service they provide clients is their biggest differentiator. They’re tired of vendors pumping AI as a cost cutting measure. They’re excited about AI that expands revenue.

Testing at the speed of AI: How Inspired Testing brought order and intelligence to a global financial software platform Inspired Testing The engagement began not with a contract, but with a conversation. Inspired Testing was initially brought in to conduct a structured assessment, designed to understand exactly what a leading global fintech needed and what Inspired Testing could offer. https://lnkd.in/dBF5pthk

📊 96% of banks are experimenting with AI. Fewer than 20% see production impact. What’s holding the industry back? At the QA Financial & eCommerce Forum, we’ll break down how financial institutions can move from AI pilots to agentic, autonomous workflows across DevOps, testing, fraud, and risk, turning AI into real, measurable business value. Let’s close the gap between experimentation and execution: https://lnkd.in/gP387kbF #AgenticAI #QAFinancialForum #DigitalBanking #AIinFinance

If your payments transformation is just a processor replacement, you’re upgrading the engine while keeping the operating model broken. Many programs go live successfully and still struggle with: • Manual investigations • Duplicated controls and data • Fragile AI pilots • Poor end‑to‑end visibility across rails In our latest POV, we outline a more pragmatic path forward: ✅ Start with high‑friction domains like investigations, reconciliation, disputes, and servicing. ✅ Build a shared event and evidence fabric early. ✅ Introduce AI through a governed, layered model stack, not opaque automation experiments. Authors Sridhar Bhagawan & Vivek Dwivedi opine that the goal isn’t to rebuild everything at once - it is to reduce complexity and create optionality for what comes next. Read the full POV to see how this plays out in practice: https://lnkd.in/dg_xf6wr #PaymentsTransformation #OperationalExcellence #EnterpriseAI #FutureOfBanking #infosysfinancialservices

The hardest call I made as a founder was saying no to the bigger market. When we started Covecta, every AI playbook said go horizontal. Bigger TAM. Faster pilots. More logos. We went the other way. Banking only. Starting with Lending workflows and then expanding into deposits, Fraud, K&C/AML etc. One vertical, deep. Deloitte's State of AI in the Enterprise 2026 puts a number on the gap. Only 1% of enterprises describe themselves as AI-mature. The rest are stuck somewhere between pilot fatigue and real production. That gap isn't a model problem. It's a depth problem. The teams converting AI into outcomes aren't running broader platforms. They're going deeper into the work itself. Task by task. Process by process. Founder lesson: when capability is commoditising, depth is the bet. Sometimes, you can question your decision. Then the data shows up. #FounderJourney #VerticalAI #BuildingInPublic

AI agents are being deployed in production. Very few teams have a structured way to test them. That gap is the reason I developed the PACE Framework, introduced in Chapter 5 of my AI QA Practitioner Series, as a method for decomposing an agent's observable workflow into testable units before a single test case is written. PACE stands for: 🔹 Purpose - What is the agent designed to accomplish? 🔹 Actions -What tools, APIs, or services does it invoke, and in what sequence? 🔹 Conditions -What decision points exist? When does the agent take one path vs. another? 🔹 Exit - What does task completion look like? What happens when something goes wrong? This isn't abstract theory. In regulated environments like fintech, an agent that exits incorrectly or takes an unintended action path is a compliance risk not just a quality issue. PACE gives QA teams a structured map of the system before testing begins. That map tells you where the interesting test cases are. The full series (6 chapters) covers adversarial LLM testing, prompt injection taxonomies, and AI agent QA built from experience. 🔗 Link in comments. #AITesting #QualityEngineering #LLMSecurity #TestAutomation #AIAgents #AItesting #Zenodo