Finding Patterns: From Middleware to Modern AI

An airhacks.fm conversation with Prof. Dr. Michael Stal (linkedin.com/in/drstal) about:
writing a Star Trek Basic (en.wikipedia.org/wiki/BASIC) game from memory after a context reset,
the en.wikipedia.org/wiki/Zilog_Z80 chip and assembly (en.wikipedia.org/wiki/Assembly_language) programming via a disassembly and debugging monitor,
buying Turbo Pascal (en.wikipedia.org/wiki/Turbo_Pascal) directly from a Munich garage 500 meters from home,
studying computer science with C and Turbo c64-wiki.com/wiki/Pascal,
building a functional interpreter for a compiler generator with visual UI at university,
joining Siemens (siemens.com) for a research project,
working on the C++ (en.wikipedia.org/wiki/C%2B%2B) standardization proposal for meta-information and reflection through source-code instrumentation,
the influence on the C++ ISO standard,
the path to writing the POSA (Pattern-Oriented Software Architecture) (wiley.com/en-us/Pattern-Oriented+Software+Arch…) book in 1996 in parallel with the Gang of Four (en.wikipedia.org/wiki/Design_Patterns) book,
coordinating with Erich Gamma to avoid overlap,
finding patterns in middleware platforms like COM,
DCOM,
CORBA (en.wikipedia.org/wiki/Common_Object_Request_Br…),
and RMI rather than inventing them,
broker (en.wikipedia.org/wiki/Broker_pattern),
microkernel (en.wikipedia.org/wiki/Microkernel),
pipes and filters (en.wikipedia.org/wiki/Pipeline_(software)),
blackboard (en.wikipedia.org/wiki/Blackboard_(design_pattern)),
layers,
publisher-subscriber,
forwarder-receiver,
client-dispatcher-server,
and whole-part patterns,
learning Java (java.com/en) from Java in a Nutshell by David Flanagan,
skepticism about Java replacing C++ that proved wrong,
Java Competence Centers at Siemens and Sun Microsystems (en.wikipedia.org/wiki/Sun_Microsystems),
Java mobile edition for handsets and the rise of enterprise Java,
Siemens application server limited to stateless EJBs,
embedded systems at Siemens written in C++ and rust-lang.org for ICE trains and medical tomographs,
Java for enterprise services and monitoring,
LLM disillusionment on the Gartner hype cycle (en.wikipedia.org/wiki/Gartner_hype_cycle),
the plateau in transformer-based models,
the brute-force compute investment vs specialized models,
AGI skepticism rooted in transformer architecture limits and missing context memory and consciousness,
physical AI cooperation between Nvidia and Siemens,
the n times square root of n attention scaling,
the practical context window limits below the advertised million tokens,
errors during code generation with Claude Sonnet 4.5,
the keep-it-simple principle and Occam's Razor (en.wikipedia.org/wiki/Occam%27s_razor),
simplicity as an art,
the role of naming,
applying patterns to existing systems rather than coding them as building blocks (wiki.c2.com/?SmalltalkBlocksAndClosures),
over-abstraction of HTTP behind custom adapters as an anti-pattern,
Convention over Configuration (adam-bien.com/roller/abien/entry/ioc_and_conve…) in Quarkus (quarkus.io) with JAX-RS (projects.eclipse.org/projects/ee4j.jaxrs) and Model Context Protocol (modelcontextprotocol.io/docs/getting-started/i…) annotations,
grounding LLMs with publicly available (sourceforge.net/projects/javagroups) Java and Jakarta EE (jakarta.ee) specifications written in normative RFC 2119 (rfc-editor.org/rfc/rfc2119) language,
the Ivar Jacobson (en.wikipedia.org/wiki/Ivar_Jacobson) BCE (youtube.com/watch?v=grJC6RFiB58) (Boundary Control Entity) structure as a stable scaffolding for code generation,
vibe coding limited by specification coverage

Prof. Dr. Michael Stal on LinkedIn: linkedin.com/in/drstal