Please subscribe to our YouTube channel @ https://www.youtube.com/@DevoxxForever
Subscribe to LinkedIn @ https://www.linkedin.com/company/voxxed-days-amsterdam
Follow us on Twitter @ https://twitter.com/voxxedamsterdam

This 15-minute demo shows how AI helps you spot and shrink the hidden energy and CO₂ footprint in “mature” Java code.

We use a simple, transparent theoretical model for sustainable decision making.

Micro-ops map to CPU work at one to five gigahertz, which we convert to milliwatt-hours bandwidth. Multiply that by your grid’s carbon intensity to estimate CO2. This matters because much of the marginal electricity on the grid still comes from gas turbines, so every milliwatt-hour you avoid cuts indirect emissions.

The AI annotates code line-by-line with uOps and mWh, flags smells/SOLID issues, predicts boost/threads risks, etc.

You’ll see how this works on small, widely used open-source Java libraries (think helpers, IO, logging, JSON).

We apply a lightweight playbook focused on under-utilization, waiting patterns, and bottlenecks. Using our ten DevOps++ open-source principles, like eliminate idle compute, right-size memory, prioritize I/O before scaling, prune work at the source, etc.

Key takeaways
A practical AI-assisted method to estimate a theoretical energy from code.
A repeatable playbook for low-risk patches that save watts without sacrificing speed.
A governance hook to make “less energy” a default quality bar in your CI/CD pipelines.

Target audience
Java engineers, tech leads, SRE/DevEx, FinOps/Sustainability owners, or anyone who wants measurable, low-effort energy wins in code and CI.

Devoxx

A conference talk exploring whether generative AI can estimate the energy consumption of code before direct measurement. Using a custom theoretical model (mapping CPU operations to energy units), the speaker demonstrates how an AI assistant called Sparky can analyze code patterns and classify waste as underutilization, waiting patterns, or bottlenecks. Two concrete examples are shown: an OpenJDK C++ bug fix that reduced file I/O overhead by ~40x, and a Java JSON serialization optimization switching from StringWriter to StringBuilder, saving roughly 0.31 kWh per terabyte serialized. The core argument is that while AI estimates are imperfect and can hallucinate, they can guide developers toward energy-significant optimizations at scale across large fleets of containers and VMs.

Can GenAI predict code’s energy use and why should we care? by Wilco Burggraaf