Achshar

I believe skills are much more than a simple markdown file with instructions. They are a very powerful script engine. How I organize it is that of course there's a markdown with instructions but I split the work in a hybrid of script + instructions. So all the work that can be deterministic is a python script api surface and all the logical or thinking work is in instructions. and agent is also instructed on how to use the api of the python helper functions. This makes it almost like a normal script but the runtime is a harness and the business logic can be any combination of code + human-level intelligence.

So I like to do all the edge case handling and validation etc via a helper function, and the agent is simply instructed to call the function to do something. It is extremely powerful and a completely different way of automating things. I am constantly forced to re-think how computers are supposed to work and its limitations.

I believe the knowledge of multiple languages brings something onto the table that having narrow language knowledge does not. The LLM will be able to better generalize a problem, apply universal concepts and avoid mistakes that would otherwise be too "echo chambery". Even as a human, knowing multiple languages has made me better at my primary language.

I am working on a vs code extension for antigravity that will let me designate one thread as a "controller" and it can spawn and chat with multiple other agents in ag. I have a separate extension that lets me control any chat agent in ag from a telegram bot. So that should ideally allow me to replace my human work of checking on agents. It should also allow me to pit claude against gemini and use low models for less smart tasks and costlier models for tasks that require more thinking. I have more or less defined my development workflow in stages where first I write a comprehensive wiki + empty test files with descriptions, then I write the code and then recursively fix the code to meet the test criteria set earlier. I want to automate this with the "controller" thread. I have come to realize a knowledge graph via wiki is the only way I am able to get AI to generate production ready stuff, otherwise its too flaky.

I feel like I am living in a bubble, no one seems to mention Antigravity in these discussions and I have not had any issues with Ultra subscription yet. It seems to go on forever and the Interface is so much better for dev work as compared to CC. (Though admittedly my experience with cc is limited).

I strongly believe google's legs will allow it to sustain this influx of compute and still not do the rug-pull like OAI or Anthropic will be forced to do as more people come onboard the code-gen use case.

I also have a small dev team of 25 that wants to move to gitlab paid model but the upfront cost is too much. We need a monthly payment plan even if it costs more than annual one because I can't justify a high upfront cost to my ceo. We're rapidly expanding and expect to be 100 within 2021.

The auto update certainly falls in the former you mean? These (older, less tech savvy) people clearly lack the imagination. That's why auto update has to be mandatory. This is again for consumer facing OS and not for the server infra.

Disagree. “If I had asked people what they wanted, they would have said faster horses.” - Henry Ford. Customer is mostly right. Not always. Use your best judgement as a business owner. Innovate where necessary and stick to customer feedback where you think it makes sense.

I use cmder on windows. It's pretty good you should check it out.

No. Just no. Don't install a library and two others just to show a tree view. Make your own with whatever features you need and just those features, nothing else. Less than 200 lines of code and one day's work for a competent developer. Don't install three libraries in your webpage just to show a tree view.

How does once check for the presence of a VM from inside the VM? Doesn't that defeat the purpose of the VM to begin with?

We do a bunch of stuff superset doesn't. Mostly related to filters and how they're implemented. We never mess with user's query. We only supply filters to it. Superset doesn't let users write optimized queries. We replaced superset for one of our largest clients and they are extremely happy with the move. This is pure js as well, nothing like superset tech stack wise.

Out of those I only have substantial experience working with Superset. We replaced Superset for one of our biggest client a few months ago. We built a ton of features in the process. My take was that they ended up meddling with the user's query which raised issues with SQL injection and performance. For example filters only going on the end of the query.

In Kato filters can be replaced anywhere. Plus we offer a ton of integration and data manipulation features. One of the biggest things that the client really liked over Superset is the ability to set up drilldowns where when you click on a bar chart on any graph you can see further details about that bar chart in a table, or any other graph. And this drilldown can go however many levels deep. We can link different reports from entirely different datasources with drilldown. I don't think that's available in any of those mentioned above and it completely changes the user experience.

We have 3 different levels of storages to avoid hitting remote. Redis cache for query, storing the results on our server and letting users edit the data as well and finally front end cache with service workers.

I think I'll start writing out the feature comparisons.

That... would be the marketing team for you. We do spend quite a bit of our time adding stuff that I don't think I've seen anywhere else but is extremely useful for some of our biggest clients.