Yes, agentic coding is machine learning. Not because of the underlying language models, but because it is gradient descent, where the test harness provides a gradient for the agent to traverse.
I suggest you encoding your invariants in the harness. Architectural invariants that can be mechanically checked, including which modules are approved, which dependencies, etc.
Yes. It’s astonishing how I see individuals with these enormous egos centered on their intellectual, technical achievements, who have such egregious blindspots surrounding their own unethical attitudes and behavior.
Sirs, please turn the apparatus upon itself. Question yourself as a part of a larger system. What is your function among members of your own species? How can you falsify your hypotheses on this subject? Please don’t be so incurious.
We set up something with a registry of AgentCards and a messaging system with SSE streaming—then we realized we didn’t need an agent on the other side so now we have this weird hybrid pattern.
I’m in the first weeks of designing nub: Not Unix, Basically. The motto is “general-purpose, provably correct software, for everyone”. The idea is to build, on top of seL4, a software stack that builds compile time capability guarantees and other high-assurance primitives into every piece of software by default.
“nub” is a good name because it can also be read as “noob”, which I am, when it comes to PL, type systems, and OS design. But I’m loving the deep connections that I am learning to draw across subjects like computability theory, functional programming, and brass-tacks software development.
For example, I find myself understanding the purpose of the borrow checker in Rust (“to make race conditions impossible” is my understanding: feel free to correct me / add color).
No website or GitHub repo yet. I’m brushing up on classical logic and learning Idris. Installed QEMU today.
Still haven’t succeeded! I’m on my 8th business since being a teenager. I have lost quite a bit of money. I’ve earned some through consulting. I think for those of use for whom it doesn’t come easy or natural (and maybe, for everyone) you have to identify as an entrepreneur—wanting and even trying isn’t enough. Then let go of all outcomes. Then just move forward every day.
But I haven’t succeeded yet, so I might not be giving the best advice!
I think it’s worth having a tiny model that can convert a short sentence into a structured tool call. Now you can add an orchestration layer and you have separation of concerns. That orchestration layer can be a mixture of deterministic and probabilistic systems. Break the foundation model into several single use tools, please!
happy to chat about crazy futuristic ideas, things you're writing, software engineering, music etc