<NESway/>Let's Build Together

Field notes

What we learn shipping AI, data, cloud and security to production. Every number is sourced in the margin.

95%of generative AI pilots show no measurable P&L impactAgentic AI & Multi-Agent Systems4 min readWhy AI pilots stall before production, and the five checks that get them outMost AI pilots stall because nobody designed for production: no agreed success metric, data the system can’t reliably reach, no cost per task, no permissions model, and no monitoring. MIT found 95% of generative AI pilots show no measurable P&L impact. Settle those five things before the pilot starts and it has a route to production.Read the note

Earlier notes

  1. 29%of IaaS and PaaS spend is wasted, up for the first time in five yearsCloud Modernization & FinOps3 min readYour GPU bill is the new cloud bill: FinOps for AI spendFinOps for AI applies cloud cost discipline to GPUs, model APIs and inference: tag every AI workload, allocate spend to a use case, set budgets with alerts, switch idle capacity off and track cost per task. Flexera estimates 29% of cloud spend is now wasted, the first rise in five years, as AI workloads make costs harder to forecast.
  2. 24hto send an early warning on an actively exploited vulnerability under the CRADigital Provenance & Trust4 min readTwo EU deadlines just passed: what Article 50 and the CRA now requireSince 2 August 2026, EU AI Act Article 50 requires chatbots to disclose they are AI and generative AI output to be machine-readably marked. Since 11 September 2026, the Cyber Resilience Act requires manufacturers to report actively exploited vulnerabilities within 24 hours. Both are engineering problems: content provenance in your AI pipelines, and an SBOM-driven reporting process.

Reading about it is the slow way.

Tell us what's stuck. One call with a senior engineer gets you a straight answer on what it would take.

Let's Build Together