What was your prompt? Literally "Push the project further"? Then the behavior wouldn't be very surprising.
As you probably know, you have the Plan Mode available - personally I'm also a big fan of the OpenSpec workflow. If you've agreed with Claude Code on a much tighter plan, and then it started signing a contract, I'd be concerned.
I've had Claude Code drive DaVinci Resolve before in three ways:
- Generate .otio files that are imported. Unfortunately, .otio doesn't fully support everything Resolve can do.
- Resolve has a scripting engine. Claude can write Lua scripts, then execute the script (through UI Automation)
- Computer Use. It's quite token intense, but it works.
I didn't try it, but for the paid version, someone also built a MCP server based on the scripting engine.
I'm programmatically adding analysis to sim racing lap guides, therefore my flow is probably different from the usual editing flow, but I have to say I found Claude Code + Opus to be quite capable of making edits with computer use.
I had AI write a program to edit with resolve. It does transcription, looks at the transcript and decides where to make cuts. If does a decent job and saves me a bunch of time.
I helped my girlfriend use the MCP server as she had a project where all the media files had been moved to a new source location, and using Codex, it managed to hunt down all the files and stitch it all back together. Ended up saving quite a lot of manual boring labour, and was quite cool to see it work in real time:)
I spent a few tokens last year trying to create a Python lib for native (undocumented) Resolve files. I was making progress, but it was slow (I ended up scripting Resolve in different ways). I imagine that current models could reverse-engineer the native file format relatively quickly.
I used IntelliScript a lot. I give Claude a subtitle export, ask it to cut it and import the cut subtitle track with IntelliScript. Resolve then creates a video version that’s cut based on the subtitle changes
The problem I have: The CEO (and the investors, or the whole industry) is obsessed with ARR per employee. That's the key metric (in their mind).
It doesn't matter how much the employee costs. Let's assume you can hire two Juniors or one Senior for the same salary. If you have to optimize for ARR per employee - you can only hire one person. And if you have the choice between hiring a Junior or hiring a Senior... you're going to hire the Senior.
When you say the "whole industry", which industry are you referring to? The entire technology sector? Venture-backed companies specifically? Growth-stage VC-backed specifically?
This claim is counterintuitive to me in many business contexts - ex. for an established privately-held company issuing dividends to the owners, what incentive do owners have to emphasize ARR per employee rather than net income?
I predict that Anthropic will clamp down on the heavy agentic use of Claude Pro, the cost gap between Claude Code and this approach is HUGE...
Anyway, I'm doing the same - one extra tip is to use the "project" feature of Claude Desktop to give your coding assistant some context - use it similar to .cursorrules.
Why would they? There's already rate limits in place. If you use it too much, you have to stop for a while. Whether you're doing text extraction from large documents, having it write your memoir, or using a diy Claude Code, what's the difference?
There's also some fundamental limitations to the Desktop MCP experience that are probably never getting fixed; Claude Code can spin off subagents and play around with the context, I assume that Claude Desktop's form factor is basically going to stay the way it is until the end of time lol.
> The data exposure was discovered following an internal investigation conducted voluntarily by Kaiser Permanente. The company discovered that online trackers used on its websites and mobile applications were transmitting certain types of personal data when users interacted with its services.
I have respect for the individuals that started this investigation, and the ones that made sure this is publicly disclosed. This could have easily been swept under the carpet.
Actually, that they uncovered this on their own and publicly disclosed it sounds like they have an above-average privacy culture in place.
I know the odds that the person(s) who kicked off this investigation are reading this comment are very low, but if so: Kudos, well done!
To you and me, that's a lot of money, for Google's size that's really not a lot of money. The fact that it just became profitable certainly helps the beancounters decide to keep it around though.
Being in Silicon Valley has warped my sense of money so much.
When I read $300m, what I read is "when you factor in office costs & health care as well as salary/stock grants, an average FAANG software engineer must be ~$1m. That's only 300 employees!"
It's a crazy way to think. Suddenly all prices that end with `m` and not `b` seem pointless.
Google includes the income from paid Gmail accounts in their cloud revenue. From all accounts, that is all of the profit and in fact it generates more profit that offsets the cloud losses.
Answering your technical question: Some companies let you deploy their "building blocks" onto multiple clouds (you typically choose the exact region/availability zone), and then also allow you to build proper multi-cloud deployments. There is usually no switch to flip, but these deployments (if properly setup) should tolerate if one region/cloud goes down.
For higher level services, I can't think of an example. For the examples above, you're still thinking at VM level, whereas you typically don't do that if you purchase a higher level service.
I think your criticism of the term is somewhat valid, in the end these are just back-ends with APIs :)
However, the point the platforms are trying to make is that their APIs have been written with the explicit purpose of building a frontend ontop of them later on.
A lot of the old-school platforms have some APIs as well, but they are meant to e.g. import products or export orders. They are not meant to implement a frontend on top of, because they have a frontend engine as part of their monolith, and many functions you need to write a proper frontend will not be available through their APIs.
So the term "headless" means "we have APIs, and they are designed for you to build a frontend on top of".
A “headless” computer system is just one without a local interface. There’s no monitor (“head”) plugged into it. There’s also no keyboard, mouse, touchscreen, or other local interface for controlling it.
As you probably know, you have the Plan Mode available - personally I'm also a big fan of the OpenSpec workflow. If you've agreed with Claude Code on a much tighter plan, and then it started signing a contract, I'd be concerned.
reply