Nicholas Cottrell

Coding as a problem is largely "solved." That's not my claim — it's from Boris Cherny, Head of Claude Code at Anthropic, in a fantastic interview on Lenny's podcast. Some standout takeaways: Even as a senior leader, Boris still codes daily — but now 100% with AI. He's not writing code by hand anymore (though a human is still in the loop for quality review). He believes that within 18-24 months, it won't even be relevant to understand what's happening under the hood. His analogy: the printing press. Before it, only a select few could read and write — employed by nobility. The press democratized that skill. The transition wasn't painless, but it was unstoppable. My take: We've been talking about "Tech eating Ops" for a while, and the general thrust is right. But I think the better frame is this: Ops is evolving into Technologists, empowered by AI coding tools. Operations experts have deep domain expertise and have spent their careers synthesizing across complex, interconnected systems. That makes them uniquely positioned to build and manage an AI-powered workforce. This won't happen automatically, though. It takes individual initiative and strong leadership to guide people into this moment. And one more thing: I'm betting on generalist models over specialist ones. Specialist models carry too much of the baggage we saw with RPA done wrong — overly complex architecture, cascading failures that are hard to trace, and a total cost of ownership that quietly kills the ROI. The printing press didn't create more scribes. It created more thinkers. That's where we're headed. What do you think — are you betting on generalists or specialists? 🔗 https://lnkd.in/gVYWi6pk

24

7 commentaires

Quick thought: It’s not the tech that holds us back — it’s the translation. Most of the value I create these days doesn’t come from designing systems. It comes from helping others understand why they matter. I've seen smart solutions stall because the impact wasn’t clear to the people making decisions. Not because the idea was wrong — but because the message didn’t land. The moment I started framing work in terms of outcomes — revenue, efficiency, customer experience — everything got easier. Priorities aligned. Conversations moved faster. One stat that sticks: nearly 70% of enterprise IT projects miss their objectives. Often not due to bad tech — but because the value gets lost in translation. Communicating clearly is the work. Curious — how do you make complex ideas resonate with non-technical teams? #TechLeadership #AI #ProductStrategy #Communication

2

Sarah's been working on using AI to generate both code and content on this project. She's described it as taking an estimate from weeks to hours, especially when she has foundational knowledge, but lacks detailed knowledge of say a given programming language. Making me excited to set aside some time to work alongside each other. Super cool to see how she's building this in real time.

1

If you aren't following Dachary Carey on LinkedIn, then you should start now. In this latest interview, she and I chat about some of the research she's been doing on how AI agents interact with documentation, and the results are fascinating. The interview is posted on YouTube. If you watch it, please do like and comment. It really helps me out with the algorithm.

18

🚀 First Release of GoloScript - Golo Interpreter in Go I'm pleased to announce the first release of GoloScript, an interpreter that brings the Golo programming language to the Go ecosystem. What is GoloScript? GoloScript is a reimplementation of the Golo language (originally created by Julien Ponge for the JVM) using Go. Golo is a dynamic, weakly typed programming language designed to be simple, expressive, and fun to work with. Highlights: ✓ Full Golo language implementation in Go ✓ Cross-platform binaries (Linux, macOS, Windows) ✓ VSCode syntax highlighting support ✓ Docker-based playground for immediate experimentation ✓ Simple CLI with script execution capabilities Quick Example: ```golo module hello.World function main = |args| { println("Hello, GoloScript! 👋") } ``` Whether you're interested in language implementation, exploring alternative programming paradigms, or just curious about "dynamic" languages, GoloScript offers an accessible platform to experiment. 🏔️ Repository: https://lnkd.in/ekZyFcdz 📦 Download: https://lnkd.in/eBnRuikB 📖 Playground: https://lnkd.in/eWYmYduC Special thanks to Julien Ponge for creating the original Golo language that inspired this project. #SoftwareDevelopment #Programming #OpenSource #Golang #LanguageDesign #Golo

31

4 commentaires

After a moment of reflection (and catching up with emails) following two incredible industry events — LocWorld53 in Malmö and the TAUS Massively Multilingual AI Conference in Dublin — one thing is clear: AI is no longer a future trend. It's a present priority. This was clear with the number of attendees at both events, it was clear that there was an appetite to see what innovation was coming and how l10n teams are adopting tools and processes in all things 'AI'. I had the opportunity to present on the our very own Agentic AI Co-Pilot (thank you Anne-Maj van der Meer), and the feedback has been both energising and validating. From conversations with localisation leaders to direct engagement with C-level decision makers, a common theme emerged: 👉 "What’s your AI strategy?" 👉 "How do we unlock value now?" 👉 " Whats the ROI?" These are no longer abstract questions. They are top-down mandates. It’s clear that the localisation space is ready — and in many ways, demanding — a shift toward smarter automation, intelligent orchestration, and meaningful augmentation of human workflows. Agentic AI isn’t just a tool; it’s a catalyst for process efficiency Thank you to the organisers, our partners, and everyone who joined the conversations over the last couple weeks. If you're exploring how to bring AI into your localisation tech stack in a tangible, scalable way — let’s connect. #LocWorld53 #TAUSDublin2025 #Localization #AI #AgenticAI #XTM #XTRF #LanguageTechnology #Innovation

61

3 commentaires

Check this blog out from Ravi McArthur This part of Qualify, our Automated QC tool. Telestream Qualify Now Speaks Your Language — Automatically. In global media workflows, identifying the spoken language in content is often a manual, time-consuming task — especially when accents, dialects, or poor audio quality are involved. Telestream’s latest update to Qualify, its automated QC platform, uses AI to solve this challenge with automatic language detection built right into the workflow. Now, Qualify can detect which languages are spoken in a file — even when multiple languages are used or the speech is hard to understand — and output results as metadata for downstream automation. This makes it easier to route content for the right subtitling, dubbing, or localization workflows. It’s already integrated with systems like Vantage and Iconik, and the same AI engine also powers profanity detection, Dolby Atmos QC, and upcoming audio/subtitle sync checks. According to Product Manager Ravi McArthur, “Qualify can handle regional dialects and noisy environments with impressive accuracy.” For broadcasters, studios, and streaming platforms, this means faster turnaround, fewer errors, and a major leap toward fully automated global content delivery. Ful Blog https://lnkd.in/gBtiwDm3 Availlable today on Encoding.com (Vantage Gateway)

16

6 commentaires

Something I’ve noticed after working with prompts for a while… They slowly get messy. Every time something doesn’t behave the way we expect, we add one more instruction. One more rule. One more “except when…”. Each change feels reasonable. But over time, the prompt becomes long, harder to read, and harder to reason about. I’ve started calling this prompt slop. Just like code, prompts need occasional spring cleaning: Step back and read them end-to-end Remove things that are no longer needed Merge overlapping guidance Recheck whether the prompt still reflects the original intent If we don’t do this, we end up fixing problems locally while the overall quality quietly degrades. Prompts aren’t set-and-forget.They’re living artifacts.

58

5 commentaires

Why We Still Need Dedicated Agentic Assistants in the Age of Powerful LLMs With the rise of exceptional general-purpose LLMs like Claude, Copilot, and Windsurf, many teams ask: “Do we still need dedicated agentic assistants for building IaC playbooks and blueprints?” Surprisingly… yes, more than ever. While generic LLMs are incredible at reasoning, writing, and problem-solving, they aren’t optimized for the specialized, structured, high‑precision workflows required in infrastructure automation. Dedicated agentic assistants solve this gap by providing: 🔹 1. Faster Onboarding for New Users They simplify the first steps, giving users a guided path rather than an open‑ended chat. No need to learn complex syntax or tooling — the assistant shapes the experience. 🔹 2. Superpowers for Expert Users Experienced engineers no longer need to aggregate samples, wire up multiple agents, or manually validate patterns. Everything is streamlined into a single, purpose‑built workflow. 🔹 3. 100% Ready‑to‑Use, High‑Quality Blueprints Generic LLMs draft good content; dedicated assistants deliver production‑ready, validated IaC blueprints aligned with best practices and organizational standards. 🔹 4. Seamless DAP Integration Instead of context-switching, users get a unified, deeply integrated experience — from generation to implementation. 🔹 5. Consistency, Reliability, and Governance Dedicated assistants enforce conventions, ensure compliance, and maintain repeatability — areas where general LLMs inevitably vary. 🔹 6. Reduction of Cognitive Load They remove ambiguity, automate decision trees, and guide users through the exact workflow required. In short: 🔹General LLMs are powerful. 🔹Dedicated agentic tools are precise. Together, they unlock real enterprise-scale automation. See more in my Medium article: https://lnkd.in/dWjyMmMA #iwork4Dell #IaC #Tosca #Blueprints #Terraform #Ansible

13

🚀 Proud moment! AT4.0 is a leap forward in making speech technology more inclusive and effective (especially for non-native English speakers). The team at Sanas has pushed the boundaries with: - A 27% ASR accuracy boost for Indian-accented English - Natural, clearer speech output - Compatibility across even the most lightweight devices This isn’t just an upgrade; it’s a serious milestone in how humans and machines understand each other. Incredibly excited to be part of this journey! #AccentTranslation #SpeechAI #ASR #SoundsLikeInnovation #SoundsLikeSanas #MarchToABillionVoices

6

🚀 Big improvements to #GitHub Actions this past calendar year We’ve made some meaningful upgrades to help teams scale automation, reduce duplication, and move faster with confidence. A few highlights: ✅ YAML anchors to cut down repetition in complex workflows 🔒 Non-public workflow templates for secure, internal reuse across your organization 🔁 Deeper reusable workflows — now up to 10 nested and 50 total 📦 Larger caches to better support dependency-heavy and large-scale builds 🎛️ More flexibility with workflow_dispatch, increasing inputs to 25 These enhancements are all about making CI/CD more powerful, reusable, and enterprise-ready—without adding complexity. 👉 Full details here: https://lnkd.in/gk5VBrMT #github #actions #automation

12

1 commentaire

You asked for it, and we listened. MULTILINGUAL Canary v2 and Parakeet v3!! 🌏 25 European languages 🏆 SotA on Multilingual Open ASR Leaderboard 🔥 600x and 2000x faster than real-time 🕰️ Timestamps! 🗣️ Speech translation (Canary) 🃏 Granary: all data is open, train it yourself! Which model is best for what? nvidia/parakeet-tdt-0.6b-v3: blazing fast and accurate ASR inference with PnC and timestamps nvidia/canary-1b-v2: top accuracy with fast inference, ASR and translation, with PnC and timestamps Both are commercially-friendly licensed: CC-BY-4.0 Granary: the dataset used to train Parakeet-v3 and Canary-v2. It has 1.000.000 hours with commercially permissive licenses. 25 European languages. It is pseudo-labeled with Whisper and Qwen-2.5, and filtered for quality. Built in collaboration with partners from CMU and FBK. This comes with a new release of Multilingual Open ASR Leaderboard starting with 5 languages, to be expanded soon. Canary-1B-V2: https://lnkd.in/gCnah5ud Parakeet-TDT-0.6B-V3: https://lnkd.in/gDkJNQd7 Granary: https://lnkd.in/gRxPFEH9 Leaderboard: https://lnkd.in/eqp3jhjj Finally kudos to the entire NVIDIA NeMo team for pulling it off, with a special shout out to Nithin Rao Koluguri and Monica Sekoyan for their incredible effort.

354

12 commentaires

Eric Eggert challenges the idea of blindly following so-called “best practices.” Instead, he argues for critical thinking, context, and thoughtful application when building for accessibility and the web. It’s a timely reminder: what’s “best” isn’t always universal, and may even hold us back if applied without reflection. What do you think, are “best practices” a helpful guide or a barrier to real progress? https://buff.ly/8uJQncJ #DigitalAccessibility #InclusiveDesign #UX #WebDevelopment #A11y #AccessibilityMatters #BestPractices #Accessibility #WCAG

1

Anthropic’s Claude Cowork Plugins Just Proved What We’ve Been Saying: Traditional Enterprise Software Is Dead. On January 30, 2026, Anthropic quietly released plugin support for Claude Cowork — their agentic desktop collaborator. They open-sourced 11 ready-to-use plugins, including one called simply “Legal.” Within days, the market has reacted. Bloomberg reported a $285 billion and going sell-off across software, legal-tech, data services, and professional information companies. This is not hype. It was the sound of monolithic enterprise software hitting its expiration date and Rimini Street taking off as a result. What Anthropic Actually Shipped From the official announcement (https://lnkd.in/dES_zBFE): “Plugins let you bundle any skills, connectors, slash commands, and sub-agents together to turn Claude into a specialist for your role, team, and company.” No rip-and-replace.
No forced upgrades.
No vendor lock-in. Just autonomous agents that read, reason, act, and iterate — on top of the systems you already own. This Is Dario Amodei’s “Adolescence of Technology” Playing Out in Real Time that I posted on just yesterday. In his January 2026 essay, Anthropic CEO Dario Amodei warned that superhuman AI would arrive in 1–2 years and that our current systems are nowhere near ready. The market just priced that warning in — violently. Legacy software that charges premium prices for static, upgrade-dependent functionality is now competing against infinitely customizable agentic layers that work with whatever you already have. Exactly What Rimini Street Has Been Saying Since October 2025 The Bottom Line. The adolescence has begun. Organizations that treat their current ERP and enterprise systems as stable systems of record — while layering intelligent, agentic AI on top — will win. Those still planning the next big rip-and-replace will be left paying for yesterday’s technology while their competitors move at the speed of AI. We’ve been preparing for this day for over a year. Our customers are already living it. The question now is simple: 
Which side of history is your enterprise software strategy on? Let’s talk about how Agentic AI ERP can transform your organization — without the disruption the market just punished so harshly. 
→ Learn more: https://lnkd.in/dkQuaKZd #AgenticAI #ERP #AI #DigitalTransformation #EnterpriseSoftware #ClaudeCowork #RiminiStreet

6

Most people think AI is just about automation. But really I see it in three buckets: - Do the things you were doing, but faster ⚡ - Do things you couldn't do before, through access to knowledge and tools 🌐 - Automate your life with agents 🤖 If you’re early in your AI journey, start with bucket one. Don’t rebuild your life, just do the same work, 2–4x faster. No need to install OpenClaw🦞! My favourite tool in this area is Wispr Flow. I've been using it for about 8 months now, it's the cheapest of the AI tools I pay for and, if I'm honest, the most impactful on my work. Simply put, it's an AI-powered dictation tool. Hold down a key on your keyboard and just speak your messages. Release, and bam - it's understood context, removed your umms and ahhs, anywhere you misspoke and added punctuation. Then when you combine this with other AI tools, dictating your prompts, "brain-farting" your thoughts... 🤯 Don't just take my word for it thought, I recently shared this tool internally at Mews and here's what people had to say about it: "Wispr Flow changed my life" - Craig Bonafont "I've had this for 4 days and its changed my life thank you" - Jen Carlton "This is an excellent tool" - Rob Ralston "[With Wispr] I have gone 10x in a week" - Michael Enslin "Been using this for a while... it is excellent" - Andrew Cullen If you want to try it out you can get a month of premium with my referral code: https://lnkd.in/dc9Xia3X (Although I already have 2 years free because of the amount of people I've referred so if anyone wants to drop their links in the comments please go for it as I really don't need it 😅)

30

2 commentaires

Very excited to share more about Service Publishing, now available on Tanzu Platform 10.3 🎉 Service publishing streamlines collaboration between dev teams, by taking an application running on the platform and publishing it directly to the marketplace. This enables other teams to easily discover and self-serve API access with a simple create and bind experience 🧑‍💻 You can find out more in our blog below, which includes a quick demo video of the end-to-end flow👇 Huge shout out to the team who made this happen 🙌 Virgil Mihailovici Brenden Blanco Dilip Tadepalli Abhishek S. Daniel Fein Morgan Fine Ben Hale Ajit Roy Haripriya Rajagopal Siva Kommuri Yogesh Katreddy Jie T. Lucas Martiniano

22

If you’ve ever opened a multilingual RFP and thought, “Wait… didn’t we already answer this in another language last quarter?” then join us on 27 November at 14:00 CET. Our special guest this time is Serge Robyns from OpenWay ✨ We’ll walk you through how structured content and light automation can make multilingual RFPs far less painful. In this short session, we’ll cover: • Why FS questionnaires get so complex so quickly • The real bottlenecks (spoiler: it’s rarely the translation) • How to keep answers consistent across markets • Where automation actually helps, and where it doesn’t • What vendors expect in cross-border submissions If your team handles multilingual questionnaires, this session can genuinely save you time, rework, and a bit of sanity.

14

2 commentaires