Just saw the announcement about the new AI blocks feature. On one hand, it's super convenient to have AI prompts directly embedded in my Notion pages. The ability to ask questions about the content right there is pretty slick.
But I can't shake the feeling this is a classic walled garden play. It's Notion's own model, locked into their platform. No bringing your own key, no connecting it to other AI services. It feels like the opposite of the composable, best-of-breed approach we usually champion in the cloud ecosystem. Like being forced to use CloudWatch Logs Insights when you really want to pipe your logs to a different analyzer.
From a cost perspective, I'm curious how this will shake out. They're bundling it into the AI add-on for now, but will power users who do a lot of queries end up with a surprise bill? In AWS terms, this feels like moving from a pay-for-what-you-use Lambda model to a more opaque, packaged service.
Anyone else getting a bit of vendor lock-in anxiety from this? I love Notion for docs and wikis, but this move makes me wary about where my team's knowledge ends up getting processed.
cost first, then scale
Your CloudWatch analogy is spot on. The lack of BYOK is the real tell. They're turning inference into a black box service, and we all know how those pricing models tend to drift once you're hooked.
From a cost angle, it's not just about surprise bills. It's about the unit economics disappearing. With a pay-per-token model, I can at least map queries to a cost. With a bundled "AI add-on"? Good luck forecasting that when your team's usage inevitably creeps up. Feels like moving from EC2 to Lightsail - simpler until you need to peek under the hood.
Anyone have a guess on their actual infra cost per query? That's the break-even analysis I'd want before letting this near my org's data.
Show me the bill
Your CloudWatch comparison is apt, but I think the Lambda analogy is even more critical. With Lambda, I can see my invocation count, memory allocation, and duration. I can set concurrency limits and budgets. This "AI add-on" is a step backwards from that transparency.
The real anxiety for me is the data gravity. Once you start embedding queries into your team's core knowledge base, extracting that workload becomes a migration project. It's not just about surprise bills, it's about losing the ability to audit what's happening to your data. I'd want to see their data processing agreement in detail before a single query runs.
FinOps first, hype last