GenAI is fun… until you try to keep it running in prod
GenAI is fun… until you try to keep it running in prod 😅
I’ve been seeing tons of GenAI demos lately and yeah, they look great. But every time I end up thinking, okay cool, but how do you operate this thing after the demo?
Recently AWS started talking more seriously about GenAIOps.
GenAI just doesn’t behave like normal apps. Same prompt, different output. “Works” but not always right. Tokens quietly draining money. Stuff breaks in weird ways.
Funny thing is, just recently I found myself using shell scripts and multi-stage Azure DevOps pipelines to build some guardrails and ops around GenAI workflows. Not fancy, but very real. And that’s when it hit me, yeah, this absolutely needs its own ops mindset.
AWS is basically saying the same: treat prompts, models, agents like deployable artifacts. Monitor quality, not just uptime. Add safety, cost controls, evals. It’s like MLOps… but leveled up for GenAI chaos.
This feels less like hype and more like reality catching up. We’re clearly moving from GenAI experiments to GenAI systems. And systems always need ops.
Good reads if you’re curious: https://aws.amazon.com/blogs/machine-learning/operationalize-generative-ai-workloads-and-scale-to-hundreds-of-use-cases-with-amazon-bedrock-part-1-genaiops/
I hope you are happy now @mods. 😜
#AWS #GenAIOps #GenerativeAI #DevOps #MLOps #CloudEngineering
https://redd.it/1pt3b7w
@r_devops
GenAI is fun… until you try to keep it running in prod 😅
I’ve been seeing tons of GenAI demos lately and yeah, they look great. But every time I end up thinking, okay cool, but how do you operate this thing after the demo?
Recently AWS started talking more seriously about GenAIOps.
GenAI just doesn’t behave like normal apps. Same prompt, different output. “Works” but not always right. Tokens quietly draining money. Stuff breaks in weird ways.
Funny thing is, just recently I found myself using shell scripts and multi-stage Azure DevOps pipelines to build some guardrails and ops around GenAI workflows. Not fancy, but very real. And that’s when it hit me, yeah, this absolutely needs its own ops mindset.
AWS is basically saying the same: treat prompts, models, agents like deployable artifacts. Monitor quality, not just uptime. Add safety, cost controls, evals. It’s like MLOps… but leveled up for GenAI chaos.
This feels less like hype and more like reality catching up. We’re clearly moving from GenAI experiments to GenAI systems. And systems always need ops.
Good reads if you’re curious: https://aws.amazon.com/blogs/machine-learning/operationalize-generative-ai-workloads-and-scale-to-hundreds-of-use-cases-with-amazon-bedrock-part-1-genaiops/
I hope you are happy now @mods. 😜
#AWS #GenAIOps #GenerativeAI #DevOps #MLOps #CloudEngineering
https://redd.it/1pt3b7w
@r_devops
Amazon
Operationalize generative AI workloads and scale to hundreds of use cases with Amazon Bedrock – Part 1: GenAIOps | Amazon Web Services
In this first part of our two-part series, you'll learn how to evolve your existing DevOps architecture for generative AI workloads and implement GenAIOps practices. We'll showcase practical implementation strategies for different generative AI adoption levels…
👍1