How we let our engineers choose their AI tools

Today I demoed a refactor. Someone asked what percent was Copilot. I laughed it off, then felt a little hollow. On the way home I reread a note I wrote earlier and it pulled me back to my North Star. Tools can help with typing. Our craft is judgment under constraints. What mattered today was choosing seams, clarifying interfaces, protecting invariants, writing tests that let others move without fear, and planning a safe rollout. None of that is measured in percentages. I am sharing this because the AI conversation can make good engineers feel like their work has shrunk to autocomplete. It has not. Tools speed us up. We stay accountable for design, trade-offs, and reliability. If you are feeling a bit lost in the AI hype, this helped me realign. Link in first comment. #AIinDev #GenerativeAI #Copilot #SoftwareEngineering #Refactoring #EngineeringLeadership

💡 The real question isn’t “build or buy.” It’s: What’s the cost of building? In our latest article, we uncover the hidden cost of building AI infrastructure yourself: ⏰ 6–12 months in development before users see value 👩💻 1+ full-time engineer just to maintain it (≈$200K/year) ⌛ Lost time building infrastructure instead of improving the product 🤕 And the biggest risk of all: success — because that’s when the system starts to break This isn’t just Aquant’s story. It’s the reality for every team moving from AI prototype to production. 🔗 Read: https://lnkd.in/gW7Ti3cZ

🚀 Faster Code, Slower Delivery — The AI Velocity Paradox Harness’s State of AI in Software Engineering 2025 report reveals that while 63% of teams ship code faster with AI, 72% have faced production incidents, and just 6% have fully automated delivery. The fix: extend AI beyond coding into testing, deployment, and security to turn speed into lasting advantage. Learn more: https://lnkd.in/dsipDMKY

🚀 Faster Code, Slower Delivery — The AI Velocity Paradox Harness’s State of AI in Software Engineering 2025 report reveals that while 63% of teams ship code faster with AI, 72% have faced production incidents, and just 6% have fully automated delivery. The fix: extend AI beyond coding into testing, deployment, and security to turn speed into lasting advantage. Learn more: https://lnkd.in/gS3_9Qds

If you’re an aspiring AI engineer, here’s the shift you can’t afford to miss: We’ve officially moved from Prompt Engineering → Context Engineering. Prompt engineering was about crafting better inputs. Context engineering is about designing smarter systems. As models get larger and more capable, the real differentiator isn’t how well you can write a prompt, it’s how well you can manage, structure, and serve context. In our next hands-on AI workshop, we’ll dive deep into: - What “context” really means in LLM systems and why it determines how powerful your model can become - Techniques like Summarization, Query Rewriting, Context Offloading, Isolation, and Memory - How to balance accuracy, cost, and reliability in production use cases - A live demo bringing these methods together in real time 📅 November 8 ⏱️ 1.5 hours 🎙️ With Arvind Narayanamurthy & me 🎟️ Save your spot: https://lnkd.in/dvFmMiRr

🚀 Faster Code, Slower Delivery — The AI Velocity Paradox Harness’s State of AI in Software Engineering 2025 report reveals that while 63% of teams ship code faster with AI, 72% have faced production incidents, and just 6% have fully automated delivery. The fix: extend AI beyond coding into testing, deployment, and security to turn speed into lasting advantage. Learn more: https://lnkd.in/e3GC7h2X

🚀 Faster Code, Slower Delivery — The AI Velocity Paradox Harness’s State of AI in Software Engineering 2025 report reveals that while 63% of teams ship code faster with AI, 72% have faced production incidents, and just 6% have fully automated delivery. The fix: extend AI beyond coding into testing, deployment, and security to turn speed into lasting advantage. Learn more: https://lnkd.in/g6TRDYGv

I'm extremely bullish on Platform Engineering, even more so with AI in the picture. You discover late in a project a few clicks you want to track? If you have good documentation and a well defined code structure to do that, it's a breeze to delegate it to a PR from an AI agent, review the functionality and merge. This applies to many capabilities. These golden paths you've defined for engineers are highways for AI agents. In turn, engineers can focus on deeper and more interesting work.

I think Software Engineers will have an easier time and more fun working with AI agents when they familiarise themselves with the principles. There's on overlap in application e.g with terms like context caching, types of memories, AI agents acting as the main application or the backend to one, or a package/library for a series of them. Model architecture and shaping of end products isn't a lazy option or an entirely unfamiliar process or option. There's so much opportunity to learn, create, extend, shape, monitor and enhance the value of different AI options. Some great livestreams from the 5-Day AI Intensive Course by Google https://lnkd.in/djYwWb4h

Thought I'd start sharing some of the engineering articles I've been writing about for years. The latest one is about deployment on edge, architectural constraints, power, latency. It's a pragmatic look at why deploying AI on the edge goes far beyond model optimization.