This week’s edition of www.datapro.news is a practical guide to running model work like any other production workload, with sessions, artefacts, quality gates, and cost controls. We break down what Anthropic actually shipped at Code with Claude 2026, why Claude Managed Agents resembles a job runner, how Outcomes functions like a rubric-driven test loop, and where multi-agent orchestration helps. You will also get a simple reference architecture you can use to pick a safe pilot, plus the governance questions you should answer before “agent memory” becomes a silent failure mode. Check out the video edition below ⬇️

Yet almost half of them are committing it anyway. That is not a confidence gap. That is a governance crisis hiding inside a productivity narrative. This week's edition of www.Datapro.News digs into what the empirical data actually shows about AI agents and code development, and the findings should give any data professional pause. Here is what caught our attention during the research: 👉🏼 A randomised controlled trial found developers with AI assistance felt 20% faster. Measured productivity showed a 19% slowdown on complex codebases. The gap between perception and reality is almost 40 percentage points. 🧐 Gartner projects that over 40% of agentic AI projects will be cancelled by 2027. The reason cited is not the technology. It is inadequate risk controls and a culture of measuring success through demos rather than production outcomes. 🧑🏽‍💻 Junior developer job postings are down 40 to 50% since early 2024. The apprenticeship pipeline that produced the senior engineers capable of verifying AI output is quietly being dismantled. If your organisation is building an AI-assisted engineering strategy, or governing the data that feeds one, this week's edition is worth your time 👇🏼

Happy Star Wars Day! A couple of years ago, we actually did a podcast on the way the Empire uses data management to keep an iron fisted grip over the Galaxy Far Far away - you can check it out here! And no matter where you are today, may the force be with you! https://youtu.be/AmNuhm-lRZc

We recently put this together based on what we’re seeing across a number of AI projects. A common pattern is that it’s not the model that’s holding things back, it’s the structure and quality of the data underneath. If the foundation isn’t consistent and well organised, AI outputs become hard to trust. This piece goes into that in a bit more detail, including where approaches like Data Vault can help create a more reliable base for AI in a lakehouse setup. Interested to hear if this aligns with what others are seeing or if you’ve tackled it differently. https://ignition-data.com/resources/data-intelligence-series/your-data-architecture-is-holding-your-ai-back

Every model now costs tokens. The free-at-point-of-use era for your most-used coding tool is over in 33 days. And it is not just GitHub. Anthropic has already made the move, quietly restructuring its enterprise agreements to replace flat-rate subscriptions with consumption-based billing, removing volume discounts at the same time. This week's datapro.news explains exactly why this was inevitable across every major vendor, what it means for your pipelines, and what you need to do before the next renewal lands. 👉 datapro.news