What happened
Researchers from MIT CSAIL and Microsoft Azure Research introduced Murakkab, a system that lets developers describe agentic workflows in plain language instead of manually wiring together models, tools and hardware. Murakkab then identifies the best models and tools, parallelizes components, and handles scheduling automatically.
In reported tests, it met user requirements using roughly 35% of the compute, 27% of the energy and under 25% of the cost of comparable methods, corresponding to about a 65% compute reduction, a 73% energy reduction and a cost reduction of more than 75% against baselines.
Why it matters
Manual assembly of agentic pipelines is slow and often wasteful. Letting a system optimize model choice, hardware and scheduling from a plain-language description could make efficient agent workflows accessible to more teams.
The energy and cost figures are notable at a time when AI spending and power draw are under scrutiny. The work is set to be presented at USENIX OSDI 2026.
MintedBrain take
The efficiency gains are striking, but they come from a research prototype, so treat the numbers as a target rather than a guarantee. The plain-language-to-optimized-pipeline idea is the part worth tracking.
Discussion
Sign in to comment. Your account must be at least 1 day old.