Probably it is the case for top world universities (ivy league, oxford, etc.).

But, as far as I know, if you attend a second‑tier but still solid university, a phd may be a very miserable experience.

My entire digital ecosystem is almost completely google free now: Proton for mail and vpn + Mega NZ for cloud storage.
Recommend if it fits your budget

However, we also use the same word when referring to classes that execute the agent loop (LLM → tool → LLM) inside this application, such as the `Agent` classes from CrewAI.

This ambiguity sometimes makes communication harder.

Everything would be in a standardized format.

With such a standard in hand, we could build a one‑fit‑all monitoring solution that tracks a crew of agents and presents valuable information and warnings to operators.

Or maybe it already exists and I’m just not aware of it?

4/4

- the tools exposed to each agent in the app.
- a description of each agent (system prompt, etc.).
- endpoints from which you can read streams of logs for each agent and its LLM/tool calls.
- …etc.

3/4

Imagine an agentic AI application exposing a standard http endpoint from which you can retrieve json (built using a known schema) that provides:

- the inner structure of the app, such as instances of the `Agent` class from CrewAI/Google ADK and their dependency graph.

2/4

I think the easiest way is to simply ship all the logs/metrics to the same collector. e.g., EKL sitting in the cloud and accessible from both your mac and from servers running other agents.

Coolwarm

Given that I’m a privacy geek, it will find its place in my toolbox.

But paying $10 for Lumo Plus… I don’t know, man, I’m not sure it’s worth it. I’ll stay on the free limited version for now.

Is 49 k cars a significant amount for the Canadian EV market?

It turned out to be more effective to store only variable data in databases and implicitly embed domain knowledge in code, rather than using an ontology.

Simply put, none of the engineers developing systems for yet another bank, logistics firm, etc., have concluded, “yeah, it’s definitely a good idea to create an ontology of business rules for my customers and write soft that retrieves data from these ontologies and automates processes based on it.”

I believe that the adoption of technologies created in the academic world depends solely on whether they appear useful in real software built for enterprise customers.

Given that ontologies are a very niche area that is not widely used outside academia, I doubt whether there is any user friendly software for them at all.

this is what we've used in uni during KRR course: protege.stanford.edu

Filesystem is a good point. Ok, maybe stdio is mostly for AI agents aiming to work on users’ local machines 🤷

Yeah, most probably it's important part of the protocol for historical reasons indeed. But it was strange for me to see it used in this Udemy course and in the Google ADK docs: google.github.io/adk-docs/too....

I don't really remember ever using stdio as a transport anywhere.