DORA's 2025 Report: AI as a Diagnostic for Software Development

In software, hype fades fast but data never lies. When Google Cloud’s DevOps Research and Assessment (DORA) team dropped the 2025 State of AI-Assisted Software Development report on Sept 22, the industry took notice. This wasn’t another puff piece on AI #codingtools, it was the first DORA report built entirely around AI, a decade-in-the-making audit of how machine intelligence is reshaping engineering. Nearly 5,000 pros across 80 countries and 40 industries. 100+ hours of interviews. No noise, just evidence. Led by Nathen Harvey, DORA Lead at Google Cloud, and advised by Nicole Forsgren, the team reached a blunt conclusion: AI isn’t a magic wand. It’s an amplifier. Strong orgs get sharper. Struggling ones get exposed. Harvey’s crew turned ten years of #DevOps research into a lens for AI. 90% of tech pros use AI at work; 80% say it boosts #productivity; 59% see #codequality improve. But gains hold only for teams with clean data, user focus, solid version control, and quality platforms. Without that, AI accelerates instability. Mature platform orgs were 3× more likely to see positive outcomes. Enter Jellyfish, the Boston-born intelligence layer for engineering orgs. Founded in 2017 by Andrew Lau, Philip Braden and David Gourley, the company has raised $114.5M+ across Series A–C, including a $71M C led by Accel, Tiger Global and Insight Partners. Lau, an MIT CS grad and ex-Endeca exec, built Jellyfish to make engineering measurable. Now, with AI redrawing every workflow, that mission looks prophetic. As gold sponsor and analysis partner behind this DORA release, Jellyfish didn’t just fund the conversation, it framed it. On Oct 30 at 1 PM ET, Jellyfish hosts the official DORA deep dive with Nathen Harvey, advisor Adam Ferrari, Engineering Manager Marilyn C. Cole and Sr Sales Eng Milan Thakker. Ferrari, known for Engineering Together, joins Cole and Thakker to translate DORA’s data into execution. They’ll unpack the 7-part AI Capabilities Model: clear AI stance, healthy data, AI-ready info, version control, user focus, quality platforms and small batches, the multipliers separating high performers from hype chasers. The future won’t belong to those who adopt AI fastest, but to those who know what it amplifies. Orgs investing now in #dataintegrity and #teamculture will scale AI without the crash. Next up: AI agents and autonomous workflows, the next wave of measurable intelligence in software delivery. The DORA 2025 report doesn’t predict the future, it documents the pivot already underway. AI isn’t changing how we build software; it’s revealing how we’ve always built it. And for teams ready to measure what matters, Jellyfish just handed them the mirror. #AI #Data #DataDriven #SoftwareDevelopment #SoftwareEngineering #Cloud #CloudComputing #EngineeringLeadership #Technology #Innovation #TechEcosystem #StartupEcosystem #TechNews If software engineering peace of mind is what you crave, Vention is your zen.

Architecting AIOps Pipelines — Sense, Reason, Act, Learn Loops for CloudOps Modern CloudOps isn’t struggling with lack of data — it’s struggling with lack of understanding. Dashboards show, alerts shout, and scripts execute — yet real intelligence still hides in the noise. The next generation of operational excellence depends on how we design AIOps pipelines — not as toolchains, but as learning systems. The new architecture mantra is clear: Sense → Reason → Act → Learn. In traditional CloudOps, monitoring is reactive — we see what happened. AIOps makes it proactive — we anticipate what might happen. But in cognitive CloudOps, we go further — systems evolve by learning why things happen. This requires closing the loop between telemetry, inference, and feedback. Let’s break it down: 1. Sense — Observability as Perception The pipeline begins with telemetry: logs, metrics, traces, and events. But AIOps doesn’t just collect — it interprets. Through correlation and anomaly detection, AIOps agents convert raw signals into situational awareness. It’s not just uptime; it’s understanding context. 2. Reason — Causality as Cognition Here, LLMs and ML models collaborate to identify relationships, not just patterns. Causal graphs, Bayesian networks, and vectorized reasoning let systems explain impact — “this latency spike stems from policy drift in region X.” Reasoning agents turn observability into operational intuition. 3. Act — Automation as Execution Once a probable cause is found, the action layer orchestrates resolution — scaling, rolling back, remediating, or alerting. But unlike old playbooks, actions here are context-aware and FinOps-governed — meaning automation respects cost, policy, and intent. 4. Learn — Feedback as Evolution The final loop is what makes AIOps cognitive. It analyzes remediation outcomes and updates models, thresholds, and rules dynamically. The pipeline doesn’t just react faster; it gets smarter. Architects must design AIOps pipelines as neural feedback systems — continuously learning, self-improving, and self-explaining. They integrate telemetry (Sense), inference (Reason), automation (Act), and knowledge (Learn) under a single governance plane that unites FinOps, SecOps, and CloudOps. When designed this way, CloudOps transforms from operations to cognition in motion. AIOps stops being a product — it becomes a practice of reasoning. The future of reliability is not resilience alone — it’s reflex. #AIOps #CloudOps #FinOps #LLMOps #AIArchitecture #SolutionArchitecture #CognitiveOperations #Automation #Observability #ThoughtLeadership

Google's latest DORA report reveals that 90% of software teams now use AI daily, but success depends more on organizational adaptation than tool adoption. By Darryl Taft feat. Nathen Harvey

Is your observability stack drowning in telemetry? ControlTheory says there's a better way. VMblog sat down with Bob Quillin, CEO of ControlTheory, ahead of #KubeCon + #CloudNativeCon North America 2025 in Atlanta to discuss a paradigm shift in how we think about monitoring and observability. The challenge is real: Kubernetes adoption continues to accelerate, but operational complexity and signal overload have outpaced our ability to make sense of it all. The traditional approach—collect everything, store everything in a data lake, query everything—is breaking down under the weight of modern cloud-native infrastructure. ControlTheory's answer? Distill at the edge, not after the fact. At Booth 1570, they're launching the public preview of Dstl8, an AI-powered observability platform built on OpenTelemetry that uses continuous AI (called Möbius) to: 🔷 Analyze telemetry locally and extract only essential information 🔷 Correlate patterns across clusters automatically 🔷 Explain incidents in plain English, not complex queries 🔷 Reduce log volume by 40-60% while improving detection accuracy Their open-source Gonzo TUI has already gathered 2,000+ GitHub stars in two months, and now they're bringing enterprise-grade continuous AI to the observability challenge. In our exclusive Q&A, Bob shares: ✅ Why "edge intelligence" is the future of Kubernetes observability ✅ How AI agents are becoming first-class citizens in the SRE stack ✅ What the convergence of observability, FinOps, and SecOps means for 2026 ✅ Real customer outcomes: 50% reduction in MTTR, maintenance-free dashboards The bottom line: Organizations are moving from observability to understanding—and continuous AI is leading the way. Read the full interview to see how ControlTheory is helping teams scale Kubernetes without scaling confusion or cost. https://lnkd.in/gthfu8NE #Kubernetes #Observability #AIOps #AI #OpenTelemetry #PlatformEngineering #SRE #DevOps

KubeCon is just around the corner — and VMblog asked us what we’re bringing to Atlanta. Short answer: Continuous AI + Observability Distillation. Here’s founder and CEO Bob Quillin's take on the future of monitoring and incident understanding. #KubeCon #CNCF

How to Land a Founding AI Engineer Role in 2026? EMBRACE THE CHAOS - I say! I’ve been asked this so many times lately and honestly, it’s a great question. Founding AI Engineer roles are popping up everywhere, but breaking into one takes more than just knowing how to fine tune an LLM. Here’s what I tell people Build real things: -Don’t just read about AI ship it. -Build end-to-end systems with LLMs, RAG, or agents. Put them into production or at least simulate a real environment. The best example portfolio is something that works. Get hands-on with Cloud & DevOps: -Deploy on AWS or GCP. Use containers, Kubernetes, monitoring, scaling the boring-but-critical parts that make AI systems actually run. Go deep on new AI patterns: -Prompt engineering, vector databases, multi-agent orchestration these are fast becoming core skills for AI engineers. Think like an architect: -Strengthen your system design chops. Learn to reason about trade offs: latency vs cost, accuracy vs speed, complexity vs maintainability. 5. Level up your soft skills: -Founding engineers don’t just code they communicate. -You’ll need to explain AI decisions to non-technical founders or investors, and understand the ethics (data bias, model risk, regulation). 6. Get close to startups: -Network with founders, join hackathons, contribute to early-stage projects. -Nothing beats firsthand exposure to the chaos of “0 → 1”. 7. Stay on top of regulation: -AI governance is no longer optional. Knowing how compliance and risk fit into AI systems will make you a trusted engineer. 8. Build domain depth: -AI + domain knowledge is a superpower. Whether it’s fintech, healthcare, or construction pairing AI skills with industry understanding sets you apart. 2026 will be the year of the Founding AI Engineer the ones who can build, deploy, and reason about AI systems end-to-end. If you’re serious about it, start now. And if you’re already on that path drop your learnings in the comments 👇 Let’s connect: karl.ramsaran@wtalent.com EMBRACE THE CHAOS!

🚀 𝗠𝗟𝗢𝗽𝘀 𝗶𝘀𝗻’𝘁 𝗮 𝗯𝘂𝘇𝘇𝘄𝗼𝗿𝗱, 𝗶𝘁’𝘀 𝘁𝗵𝗲 𝗺𝗶𝘀𝘀𝗶𝗻𝗴 𝗹𝗶𝗻𝗸 𝗯𝗲𝘁𝘄𝗲𝗲𝗻 𝗔𝗜 𝗽𝗿𝗼𝘁𝗼𝘁𝘆𝗽𝗲𝘀 𝗮𝗻𝗱 𝗿𝗲𝗮𝗹-𝘄𝗼𝗿𝗹𝗱 𝗶𝗺𝗽𝗮𝗰𝘁. Here’s the brutal truth: 💡 Most ML models never make it to production. Not because they’re bad but because they’re orphans of poor infrastructure. That’s where 𝗠𝗟𝗢𝗽𝘀 = 𝗠𝗟 + 𝗗𝗲𝘃 + 𝗢𝗽𝘀 changes the game. It’s not “just automation.” It’s the bridge between data science and reliable, scalable systems. Here’s what that actually looks like 👇 🔹 𝗗𝗮𝘁𝗮 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 build the foundation: CI/CD for data, feature stores, reproducibility. 🔹 𝗗𝗮𝘁𝗮 𝗦𝗰𝗶𝗲𝗻𝘁𝗶𝘀𝘁𝘀 push commits, not notebooks: models are versioned, validated, and deployment-ready. 🔹 𝗠𝗟𝗢𝗽𝘀 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 connect it all: orchestrating deployment, monitoring, governance, and compliance. ✅ Continuous integration meets continuous training. ✅ GitOps meets model lifecycle management. ✅ DevOps mindset meets data-driven systems. The result? A pipeline that doesn’t just ship models, it learns, adapts, and improves. 💬 If you’re serious about scaling AI beyond the demo stage, MLOps isn’t optional, it’s oxygen. ⚡ 𝗧𝗟;𝗗𝗥: MLOps = Turning “It works on my notebook” into “It works in production continuously.” ♻️ Repost if you agree MLOps deserves more spotlight in the AI conversation. 🔔 Follow me for Cloud, DevOps & AI insights. #MLOps #DevOps #MachineLearning #AIInfrastructure #DataEngineering #CloudComputing #AIOps #MLOpsCommunity #SoftwareEngineering #AI #TechLeadership

🔧 A Beginner’s Guide to MLOps: From Model Building to Deployment Too many machine learning models sit pretty in testing, but fail the moment they hit production. MLOps is the bridge between "works only in the lab" and "delivers real-world impact." In this guide, you’ll learn how to: • Build models with scalability in mind • Deploy them reliably using production-ready pipelines • Handle monitoring, drift, and updates with confidence • Understand what it takes to move MLOps from effort to outcome Whether you’re starting your AI/ML journey or scaling up to production, this guide helps turn experiments into systems. https://lnkd.in/dVT7RtgA #MLOps #MachineLearning #AIinProduction #DevOps #MLDeployment #TechLeadership #ITIDOLTechnologies

🔧 A Beginner’s Guide to MLOps: From Model Building to Deployment Too many machine learning models sit pretty in testing, but fail the moment they hit production. MLOps is the bridge between "works only in the lab" and "delivers real-world impact." In this guide, you’ll learn how to: • Build models with scalability in mind • Deploy them reliably using production-ready pipelines • Handle monitoring, drift, and updates with confidence • Understand what it takes to move MLOps from effort to outcome Whether you’re starting your AI/ML journey or scaling up to production, this guide helps turn experiments into systems. https://lnkd.in/dhZeUTZm #MLOps #MachineLearning #AIinProduction #DevOps #MLDeployment #TechLeadership #ITIDOLTechnologies

Last week, I was stuck on a complex technical review. Hours of back-and-forth, no clear direction. Then I changed my AI prompt from "review this architecture" to: "Act as a solutions architect. Review this code for performance bottlenecks and suggest optimizations.” Instant clarity. The AI delivered exactly what was needed. 🎯 Here's what I've learned about prompt engineering: The mistake everyone makes: Treating AI like Google. The breakthrough: Treating AI like a brilliant colleague who needs context. My go-to framework that works every time: 1️⃣ Give it a role: "You are a [specific expert]" 2️⃣ Set clear boundaries: "Focus only on [specific area]" 3️⃣ Define the output: "Give me [exact format needed]" Real example from yesterday: ❌ "Help me with CI/CD" → Generic, unhelpful response ✅ "Act as a DevOps engineer. Create a GitHub Actions pipeline for React with testing stages" → Ready-to-implement solution The best part? You don't need to be technical. You just need to be specific. Try this today: Take your worst-performing prompt and add these 4 elements: • Role (who the AI should be) • Context (what you're solving) • Constraints (what to focus on) • Format (how you want the answer) P.S. Compiled my favorite resources here: → Anthropic's templates: https://lnkd.in/gRxNgnQj → Google's free course: https://lnkd.in/gA2fkNjt #AI #PromptEngineering #Productivity #TechTips