+ 1 , agree still prefer composable pipeline architecture for voice agents. The flexibility on switching LLM for cost optimisation or quality is great for scaled use cases.

Something seems off here, OP has an extra $1500 a month for over a year and then finally noticed. They then instead of pausing or migrating “expensive” services, stop paying and then AWS terminate as expected.

Feel there is more to this story than AWS being mean.

You can try emailing garman@amazon.com and complain about the poor AWS customer service with Jeff cc’d.

Sandbox won’t be enough, distroless + “data firewall” + audit

Indeed, but a rock solid sandboxing and isolation strategy is step 0.

Still end up marked. Don’t do it

Huge legal tech business units

But to be useful it’s not in a contained environment, it’s connected to your systems and data with real potential for loss or damage to others.

Best case it hurts your wallet, worse case you’ll be facing legal repercussions if it damages anyone else’s systems or data.

Simon, this is going to produce some nice case studies of your lethal trifecta in action!

It's certainly an opportunity for it to happen publicly! We may see some API key or passwords leaking directly to the forum.

No offence, but this is 5% of what a middle manager does. We automated this task years back without LLMs too.

You don’t need a vector database or graph, it really depends on your existing infrastructure , file types and needs.

The newer “agent” search approach can just query a file system or api. It’s slightly slower but easier to setup and maintain as no extra infrastructure.