Glossary – Bernd Zuther

A

Agent Communication: How several specialized AI agents exchange tasks, results and status with each other – e.g. via structured messages instead of free-form text.
AWSAmazon Web Services: The cloud platform many of the projects described here run on in a “cloud-first” way – from compute to managed services.
Agile Methods: Umbrella term for iterative approaches like Scrum, Lean Canvas or Design Thinking that favor fast learning over rigid upfront planning.
Access TokenAPI Token: A time-limited access key a system uses to authenticate against an external platform (e.g. the Instagram API) without knowing the actual password; needs regular renewal.
Approval Gate: A mechanism that only actually executes an automatically prepared action once several conditions are met at the same time (e.g. status “approved” + an enabled live switch + the scheduled time reached) – a core principle from stage 5 of this blog’s own staging model.

Canary Release: Deployment strategy where a new version is first rolled out to a small subset of users, so risks surface early.
CI/CDContinuous Integration / Continuous Delivery: Automated pipeline that continuously tests code changes and ships them to production, instead of rare, large releases.

DDDDomain-Driven Design: Architecture approach that aligns the software structure strictly with the business domain rather than technical layers.
Design Thinking: User-centered problem-solving method with iterative steps: empathize, define, ideate, prototype, test.

ELK StackElasticsearch, Logstash, Kibana: A trio of search index (Elasticsearch), log collector (Logstash) and visualization (Kibana) for centrally searching and analyzing large volumes of logs – a classic homebrew alternative to commercial tools like Splunk.

Gradle: A build automation tool (notably for Java projects) that can, for example, turn an application into a standalone, native-feeling application bundle.

Human in the Loop: A human deliberately sits as a review-and-decision point in the workflow instead of the system deciding fully autonomously – e.g. approving social media posts before publication.
Heap Dump: A snapshot of a running Java application’s entire object memory, used via dominator trees and metrics to track down memory leaks (OutOfMemoryError).

AI Infrastructure: Your own, self-operated technical base for AI workloads (e.g. local LLMs, ComfyUI) instead of relying solely on external APIs.
Idempotent Retries: Retries of a failed or only partially successful automated run, built so that running it again doesn’t cause duplicate publications or other side effects.

Lean Canvas: A one-page template to quickly sketch the core assumptions of a business model or product (problem, solution, audience, channels).
LLMLarge Language Model: A large language model that understands and generates text – the core technology behind most AI agents on this blog.
LoRALow-Rank Adaptation: A training method that teaches an existing AI image model something new (e.g. a character or style) without retraining the whole model – far cheaper and faster than full fine-tuning.

Microservices: Architecture style where an application is split into small, independently deployable services instead of one large monolith.
Multi-Agent System: Several specialized AI agents (e.g. social, writer, coordinator agent) collaborating on a task through division of labor.
Monitoring & Observability: The ability to understand a system’s internal state from the outside – through logs, metrics and traces, not just “up / down”.

Prompt: The text instruction used to direct an AI for a task – rewritten for every request at stage 1, locked once into a system prompt at stage 2.

RAGRetrieval-Augmented Generation: A technique where an AI looks up relevant information in its own knowledge base before answering and feeds that into the prompt – instead of relying solely on what was baked into the model during training.
RCARoot Cause Analysis: Systematic troubleshooting that doesn’t stop at the first symptom but digs down to the actual, often less obvious underlying cause – the standard approach for incidents on this blog.

Scrum: Agile framework with fixed roles (e.g. Product Owner), timeboxes (Sprints) and rituals (Daily, Review, Retro) for iterative product development.
SpringSpring Framework: The most widely used Java framework for enterprise applications – Dependency Injection, Spring Boot, Spring Cloud as the base of many backend systems.
Staged Autonomy Model: This blog’s own, practical model describing how much autonomy an AI system has at a given point – from stage 1 (single request) to stage 6 (agents review agents), inspired by MIT’s autonomy scale for AI agents.
systemd Timer: A scheduled run triggered by the Linux systemd service (e.g. every five minutes) that kicks off the actual automation – the mechanism behind the automatically started runs in stage 5.

Terraform: Infrastructure-as-Code tool for managing cloud resources (servers, networks, …) as versioned code instead of manual click-config.