First-person views from the world of industrial IoT, data, and digital solutions.
Adopting AI coding tools has increased the velocity of most companies' software development teams, sometimes by tens of percent, sometimes several-fold. The team is faster than ever, but, surprisingly, often less productive than before. Not in lines of code, but in validated value shipped. They build more, yet less of it counts: features ship that nobody asked for, and every one of them quietly joins the maintenance roster for years. Turns out, speed without direction is very expensive. AI is an amplifier: it speeds up execution, but it has no opinion about whether you're executing the right thing. Where the vision is clear, that speed creates leverage. Where it's hazy, it just creates waste, faster. The true competitive advantage was never coding speed but knowing what to build. Before a line gets written comes the (often) harder work of figuring out what's worth building at all. And AI is just as useful there as it is once you're building. Yet most teams fully utilize AI only when implementing: the coding gets the acceleration, while discovery still runs at its pre-AI pace.
Every team is shipping AI agents now, and most have been handed the same docx file of the AI Governance Policy. Not one of those agents can follow it, because an agent doesn't read your policy, it just calls tools. Tool calls (an API, an MCP server, a function) are where data leaves the building and attacks land. They are where your governance isn't. Attackers are already there, from a zero-click leak in Microsoft 365 Copilot to a worm that hijacked AI coding agents outright. No better document would have stopped either one. The fix is to move the control onto every tool call and data retrieval, and make every team inherit it by default. Here is how to turn a policy no agent reads into a control every agent carries.
Effective AI-assisted development depends less on clever prompts and more on the engineering practices around the agent. Clear context, a repeatable spec → grill → plan → implement → verify workflow, and strong review gates help teams catch avoidable failures before they become rework. Agents tend to fail in recognizable ways, so teams need current documentation, reusable skills, and deliberate safeguards around planning, implementation, and review. The real leverage comes from building the scaffolding around AI: context curation, verification, documentation, and human judgment.
.png)
AI coding agents have not removed the need for software developers, but they have changed where developer judgment matters most. Less effort goes into typing implementation code, and more effort moves into writing clear specifications, steering agent work, and reviewing the results critically. This shift makes product clarity, technical intent, review culture, and verification more important than ever. The teams that benefit most from AI agents are not the ones generating the most code, but the ones most disciplined about intent, context, and quality.
Helen is transitioning from combustion-based energy production to a carbon-neutral system by 2030 and aims to end all combustion by 2040. To support this shift, Helen partnered with Brightly to develop a robust, scalable data platform that enables real-time monitoring, forecasting, and optimisation across their energy infrastructure. This blog post describes our partnership and the solution we have built together.
.jpg)
If you’re attending the Data Innovation Summit this year, I’d love to see you there. Whether you’re exploring your next move in data and AI or just want to trade ideas, please come by and say hi. You can meet our team at booth A2, join our session with Helen on district heating, or book a 1:1 meeting to talk more in-depth about your goals.
We've built autonomous AI agents since 2022 —here's what we've learned about clear instructions, tool design, memory management, and choosing the right LLM. Practical insights, real examples, zero fluff.
Explore how agentic AI differs from basic AI assistants, delivering 3-4x ROI through autonomous operation. Practical insights and implementation strategies inside.
Are you curious about the AI landscape, its real-world applications, and where things are headed? This article breaks down key trends and insights, focusing on practical AI tools rather than just theory.
At Santen Pharmaceutical, Brightly designed and implemented ChatAIRI, a generative AI tool that enables quick analysis of a vast amount of regulatory data. Read more about how we used AI to tackle a real-world problem efficiently.
Brightly joins the LifeFactFuture research program to lead the Technology Excellence Work Package in a consortium of university entities, leading life science manufacturers as well as data and technology companies.
