Screenshot of a web-based object navigator targetting the JVM management interface, to reflect out the process classpath and navigate into the contents of an uberjar. This application is defined in about 200 lines of Clojure (.clj) for the java reflection (for some simple parsing and weaving together of a set of queries that relate but were not intentionally built to relate), and absolutely no frontend code. The navigator UI is fully abstracted away by a saas, the UI is not in the same process as the JVM being interrogated, it is remote. They talk via a reverse websocket with the Electric wire protocol. For more info see clojure.net/ "Network your REPL", a new experiment from Hyperfiddle.

People still don't understand how the mental model for building AI agents is closer to a frontend than a backend service The call graph is dynamic, depending on what the LLM or user saw / decides. They need a BFF that shields them from service topology and product joins.

Thank you Guillem I believe development that focuses on user experience and rich features is what enables this fast pace. Technically the dev speed is enabled by #Electric from the Hyperfiddle team (led by @dustingetz) and the associated Electric-specific LLM skills.

Endive 1.0 is out!🚀🚀🚀 Pure-Java WebAssembly runtime, zero native dependencies, now a @bytecodeallies project. Now with native JVM GC integration for WasmGC references and tail call optimizations! Chcek it out: endive.run/blog/endive-1.0

“every failure has a moment where you saw it and looked away. that moment is the reason”

do we believe this is true? i think it is true but not in the way he implies i think there will be non-ai breakthroughs in languages and protocols and this will be the foundation of future systems that operate like in Star Wars

Agentic architecture is a concurrency and resource management problem Concurrency and resource management is unsolved at scale even for the best of developers. We’re talking the same problem space that led to kubernetes and orchestrators, except at 1000x higher frequency. Responsiveness measured in milliseconds not minutes. 10000 nodes in the system choreography, not 100. Vibecoders are not even remotely equipped to solve an architectural semantics problem that they cannot begin to even see let alone label.

So many people are stuck in the "first principles trap" Your principles are valuable only to the extent that they help you produce timely outcomes Feynman: "The first principle is that you must not fool yourself and you are the easiest person to fool."

Very impressed with Clojure Electric v3. Tried v2 a long time ago and found it very hard to wrap my head around the bad error messages. Now, especially with help from AI, it's very easy to write apps with a fraction of LOC than a react or clj program.

github.com/hyperfiddle/elect… 久しぶりに触って前より理解できる。solidjsやsvelteのようにシグナルの値とそうでない値が混在するのではなく、全ての値がシグナルならリアクティブを意識する必要もなくなるって感じか。 同じ関数内にクライアントサイドのコードとサーバーサイドのコードを混在させられるのはやっぱすごい

Meetup alert! TOMORROW, Tuesday March 10 I'll be presenting clojure.net/, a new Clojure UI experiment from Hyperfiddle. Embed service management console UIs in enterprise microservices that only have a backend, with zero frontend code! Get a great management console UI for your enterprise services with absolutely no client-side framework, no REST APIs, no Dockerfiles! Join us TOMORROW (March 10) at 2:30 ET / 1830 London for this virtual event: meetup.com/london-clojurians…

"exponentials feel like nothing for a very long time and then feel like everything all at once"

The world underestimates how difficult frontier software products are to build. Take the PKM category, ALL major players are FROZEN with ZERO product velocity, including Notion and their reported $600M ARR! What agentic coding methods seem to have done is reduced the cost of replicating previous-generation methods. Indie devs (and VCs) building pixel art nintendo games. Soon games built on top of Unreal Engine, because they can program the Unreal framework about as well as they can program Python, which is spectacularly. But without developing new frameworks, the agentic methods will simply reproduce the same set of flaws and issues that the current generation of growth stage products have. Which means they are (1) uncompetitive, and (2) late. The fundamental problem of software is that it becomes difficult to change once you have users and revenue. Being 100x faster to generate code does not solve that problem that landing the code in prod breaks the user's workflow and generates churn. Agentic coding methods also accelerate research and development, so I anticipate a new generation of more agile products. But still, these will be built by more competent founders, not less. AI reduces switching costs and accelerates learning. The advantage right now is to R&D teams that figure out how to take advantage of the new capability frontier to accelerate their very slow R&D lifecycle.

there really isn't anything out there like Missionary (by @36Rleonoel) if you want dynamic reactive computation at runtime with async & I/O built-in