Ben Shoemaker | Data, AI, AR, Tech, Product

Two big releases today – Claude Sonnet 4.5 took all the headlines, but I also found the Lovable Cloud & AI release to be super interesting. Some thoughts below.

Claude Sonnet 4.5

The announcement of Sonnet 4.5 led with a single sentence:

Claude Sonnet 4.5 is the best coding model in the world.

I find that very interesting, for a few reasons. I get why – coding is of significant interest to the folks who work at places like Anthropic, and it’s also a wedge to get into Enterprises & get into the data, billing, etc. at companies. On the other hand – most AI usage is not coding – Chat GPT says 4.2%; other surveys say 12% – but it’s the distinct minority. I think these companies think software engineering is an easy application of LLMs, and that if you win developers you win software in general.

Will Claude Code & Sonnet 4.5 take back the belt?

For this announcement – I also think Anthropic has been smarting about how Codex has taken a lot of shine from Claude Code lately – backed by some Google Trends data –

Some early returns – Simon Willison says “probably” at least right now – he got an early preview. There’s some interesting HN discussion about it as well here.

Other tidbits

Increasing support for Claude for Chrome – this is for Max users only right now, which I am not. It’s unnerving to watch an AI agent go through Google Docs, email, etc. I am going to require significant guardrails to feel confident in a feature like this. Skynet, anyone?

Lovable Cloud and AI

Extremely cool pitch and obscured by Sonnet 4.5, unfortunately – I found it incredible that there was virtually no HN discussion of this. Where my vibe coders at?

It’s a very cool value prop – backend, auth, storage, and AI features all wired up onto the vibe coding tools people have come to enjoy. The cloud portion is less interesting – it’s kind of been done.

AI apps via prompt

The AI piece is pretty new. I think the best analog to this today is v0 + AI Gateway – but that still requires writing code yourself or via agent. Integrating AI into your app is a lot harder than people think – as someone who’s been building a LOT with AI coding tools, it is a lot harder to get an AI model to do something in an app than it is with chat-based models, at least initially.

My thoughts

This will come down to speed vs. portability and lock-in imo. I don’t typically like tools like Replit, v0, etc. in general for anything more than a skeleton, because you lose control over what you’re doing. Maybe I’ll talk about it in a separate post more, but my typical flow today uses these tools sparingly – I’ll use Figma Make to generate a UI; then I’ll export it & have Codex wire up a backend, for example.

Right now, Lovable’s documentation for Supabase export says:

This is technically possible since you own the code in GitHub, but not straightforward. We are looking into making this easier.

If Lovable can make this simple enough that the lock-in is worth it, it’ll be transformative. I think they won’t be able to – and I doubt they’ll focus on interoperability.

I will probably test this out in the coming days – code a simple app that uses AI with Lovable Cloud + AI, vs. my typical Codex CLI + sparing usage of other prompt-based tools. If anyone has input on an app I should try – let me know!