Google, IBM & Meta Certificates – 40% Off
One Coursera Plus subscription covers most Professional Certificates on Coursera.
Unlock All Certificates
Running one AI agent is easy. Running six at once and trusting what comes back is the hard part. Merge cost grows with output, workers collide on shared files, a model asked three times agrees with itself, and a pipeline that never says "not run" quietly reports green. This course builds the discipline that makes parallel agent work worth its cost, in Rust, on real tools.
Key Course Features:
- Estimate fan-out speedup from a measured serial fraction and merge cost, and see when widening stops paying
- Partition work into disjoint write sets under a context budget, with one owner per section
- Let a gate the worker did not write decide whether its work passed, following a textbook ported to Rust under contract
- Run a local model as a resident lane and report its latency only from a receipt
- Orchestrate a fleet with a traffic cop, fan out and fan in, and build a heterogeneous quorum where any single FAIL blocks
- Stop the line on a red gate and trace the defect to a mechanism
Perfect for:
- Rust engineers who already use coding agents and want to run several of them in parallel without losing control
- ML and platform engineers building review, evaluation or CI pipelines around language models
- Technical leads deciding when a multi-agent workflow is worth its overhead
Every claim in the course is checked by something other than the agent that made it. By the end you will be able to design a fan-out, a reducer and a quorum that fail closed when evidence is missing.
