I am not a fan of AI agents when used poorly either, but this quote is a bit of a logical trap. No customer or user wakes up and says "I hope I get to talk to a human customer service agent today" either. Having to contact customer service at all is already a friction point. If AI or a human or a monkey pulling levers can fix my problem quickly, I don't really care. It's when my qualm is unable to be solved quickly that I will start finding flaws in the process - and because humans lack empathy for machines, it's easy to blame the machine rather than the process.
AI being used to advertise a brand is a different story - absolutely it comes across as insincere and lazy.
That must have been expensive! Thanks for running the benchmark and sharing.
I tested Coducky (my AI review macos app) on the full 50-PR Martian benchmark using qwen3.7-plus via OpenRouter as the reviewer with a lightweight pre-save precision gate with deepseek-v4-flash. The score (gpt 5.2 judge) was 43.0% precision / 35.8% recall / 39.0 F1. That puts it about inline with CodeRabbit. This cost around $7 to run the full 50 PRs.
Your post inspired me to set up a test harness for my app to continue to test model combinations. Coducky allows you to select whichever models/subscriptions you like to run reviews, but it could make sense to build a collection of model combinations that work well for this.
I'm self hosting umami for https://loose-tongues.com/. It's simple, fast, and I have full control over it. It's using postgres so I can do whatever I want with the data. https://umami.is/
AI models on their own are raw, undirected, and inherently probabilistic. A "harness" acts as a control layer wrapped around the model, designed to steer it toward deterministic outcomes. It achieves this by equipping the model with actionable tools like web search or file I/O, and by orchestrating an evaluation loop that runs until an acceptable result is produced. (various analogies work here - an astronaut and a space suit, a rocket and the launch pad/mission control, okay I'm out of analogies that aren't car engines)
You can see this in practice by looking at the leaked Claude Code source code. It is a harness around Anthropic's model built for writing code. It relies on heavily engineered (and sometimes brittle) steering mechanisms. These range from highly specific situational prompts to deterministic, hard-coded logic that executes based on the model's output.
Getting a harness right is incredibly hard and feels like whack-a-mole at times.
I've just now had a look at a couple of YouTube tutorials on Magic Mask for video and I think I understand where I got lost. It looks like the tools I was looking for are on the Color tab, and I first need to drag my images in to the timeline at the bottom of the screen and then jump to the Color tab to work with them.
All the tools are here for a good photo editing product - they just need to be extracted and arranged in a way that is intuitive for photo editing, and this would be a legitimate LR alternative.
edit: This is a great video on the Photo tools: https://www.youtube.com/watch?v=HuKgfytA0lg. I do feel a little more confident that I could use Resolve for editing after watching this.
Yes, but I think it's important to make that clear. It doesn't appear to attempt to target photographers who are not coming from a video editing background, and photographers will probably be disappointed.
I do hope they split this out to a separate focused product, as the photo editing space is in dire need of more options.
It is unclear. Their marketing material at least does imply that they are targeting photographers:
> Whether you’re a professional colorist looking to apply your skills to fashion shoots and weddings, or a photographer who wants to work beyond the limits of traditional photo applications, the Photo page unlocks the tools you need
I actually downloaded this and tried it. Am I the first one here to do that?
As someone who hasn't touched DaVinci products before (but a lot of experience with LR) - I am immediately confused by the integration of photo editing here. It feels very much like video editing software with photo editing tacked on. I can imagine that this would be much more intuitive for people who are already used to using DaVinci for video editing.
I can intuit from the interface that there are a lot of powerful editing opportunities here, but I feel lost in the software. I spent 15 minutes or so trying to figure out how to do simple masking, but I could not find any way to do it for photos.
Obviously this is just a beta and hopefully the workflow will be improved, but unless the photo editing features are extracted in to their own software package, I don't think it's enough yet to sway me from LR (and I want so desperately to be swayed)
Indeed, the hardest part is inoculating without introducing contaminants that will out-compete the mycelium (typically trich).
However, this is relatively simple and cheap to do at home — especially using a technique called uncle Ben’s tek. As it happens, pre-packaged rice is the perfect sterile environment for mycelium growth.
Once successfully inoculated, it’s basically a matter of mixing the mycelium with substrate (wood chips or coir depending on the strain) in a container with a lid that allows minimal airflow (a plastic tub for example) and waiting.
The process is really quite simple, and I’ve harvested kilograms of mushrooms with an initial outlay of around $30. Once you have mushrooms, you have spores, and the recurring cost is only substrate (and uncle Ben’s bags or grains, jars, pressure cooker if you want to sterilise them yourself).
During my life as a vegetarian/vegan I regularly consume “fake meat” products because they are simple to prepare and allow me to use recipes that are not vegetarian. For example, it’s extremely hard to find a good vegetarian burger — the beyond burger was a revelation the first time I tried it. I still like burgers, even though there are many wonderful vegetarian dishes. I didn’t stop eating meat because I don’t like the taste of meat.
Imagine that many of your favourite dishes, those you grew up with, were suddenly unavailable to you because some ingredient no longer existed. If a new ingredient hit the market that was not quite the same but was a convincing substitute in taste and texture, would you consider it silly and avoid it?
Tofu and wheat gluten have been used for centuries as meat analogues in many cultures (especially those that mostly eat vegetarian). It’s not a silly concept at all, and opens up a world of delicious culinary options to those of us who don’t eat animal flesh.
It’s very common to use it on the HTML element, less common to use it on elements further down the tree. Like many accessibility features of the web, it’s often neglected by developers.
Flashing high/low/high/low from behind here in Austria at least means “you are annoying me but not in a way that immediately threatens my safety” (beeps are reserved for slow starters at lights or near-collisions). If going in the other direction it means cops ahead.
This outputs static HTML that gets hydrated on the client (as opposed to server rendering the HTML and then hydrating it on the client), which I don’t think what was being asked.
I think what the OP was asking was more along the lines of partial hydration (where only parts of the DOM are hydrated by React/other framework) or no hydration (no JavaScript is loaded at all).
AI being used to advertise a brand is a different story - absolutely it comes across as insincere and lazy.