Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Architectural Intelligence: Ist AI die bessere ...

Architectural Intelligence: Ist AI die bessere Softwarearchitekt:in

Wie heißt es so schön? Nicht die KI wird dir den Job nehmen, sondern diejenigen, die verstehen, wie man sie nutzt. Übertragen auf das Berufsbild der Softwarearchitekt:innen bedeutet das, sinnvolle KI-Szenarien zu entwickeln, die uns dabei unterstützen, unsere Arbeitsweise als Architekt:innen und die von uns zu fällenden Architekturentscheidungen zu verbessern. Leichter gesagt als getan. Denn dazu gilt es zunächst einmal die Möglichkeiten und Grenzen von KI zu verstehen und das, was uns als Mensch grundlegend von ihr unterscheidet. Was kann die KI bereits heute leisten und was eher nicht? Welche KI-basierten Tools können uns bei der Architekturarbeit unterstützen? Inwieweit sind die Ergebnisse der KI dabei belastbar und vertrauenswürdig? Und last but not least die alles entscheidende Frage: Liefern wir Menschen am Ende überhaupt noch den entscheidenden Mehrwert?

Avatar for Lars Roewekamp

Lars Roewekamp PRO

November 07, 2025
Tweet

More Decks by Lars Roewekamp

Other Decks in Technology

Transcript

  1. Das neue Normal (Gen)AI und CODING • Code-Vervollständigung • Code-Generierung

    • Code-Review • Code-Dokumentation • Testgenerierung • Bug Fixing GenAI Voodoo
  2. Das neue Normal (Gen)AI und CODING • Code-Vervollständigung • Code-Generierung

    • Code-Review • Code-Dokumentation • Testgenerierung • Bug Fixing GenAI Voodoo
  3. Das neue Normal (Gen)AI und SOFTWARE ARCHITECTURE „Wie wirkt sich

    (Gen)AI auf Softwarearchitektur aus?“ 1. „Wie verändert (Gen)AI die Art und Weise, wie wir als Softwarearchitekten arbeiten?“ 2. „Wie verändert (Gen)AI die Dinge, die wir architektonisch gestalten?“
  4. („The six core activities for software architects.“, ISAQB) 3 2

    4 5 6 Clarification of Requirements and Constraints
  5. („The six core activities for software architects.“, ISAQB) 3 4

    5 6 Clarification of Requirements and Constraints Design of Structures
  6. („The six core activities for software architects.“, ISAQB) 4 5

    6 Clarification of Requirements and Constraints Design of Structures Design of cross-sectional Concepts
  7. („The six core activities for software architects.“, ISAQB) 5 6

    Clarification of Requirements and Constraints Design of Structures Design of cross-sectional Concepts Evaluation of Architectures
  8. („The six core activities for software architects.“, ISAQB) 6 Clarification

    of Requirements and Constraints Design of Structures Design of cross-sectional Concepts Evaluation of Architectures Communication of Architectures
  9. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation („The six core activities for software architects.“, ISAQB)
  10. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation („The six core activities for software architects.“, ISAQB) Wie kann GenAI uns dabei helfen?
  11. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation („The six core activities for software architects.“, ISAQB) Was kann GenAI gut* und was nicht so gut?
  12. Die Idee: Frage und Antwort liegen SEMANTISCH dicht beieinander. Die

    semantische Nähe wird MATHEMATISCH mit Hilfe von Vektoren abgebildet. LLM Voodoo Wie generiert das LLM eine Antwort?
  13. Hund Katze Tiger Maus Hamburg Berlin Madrid Amazonas Embedding Model

    0.6 0.3 0.1 … 0.8 0.5 0.3 … 0.4 0.2 0.9 … Embeddings (semantische Vektoren)
  14. Hund Katze Tiger Maus Hamburg Berlin Madrid Amazonas Embedding Model

    0.6 0.3 0.1 … 0.8 0.5 0.3 … 0.4 0.2 0.9 … Embeddings (semantische Vektoren) Berlin: Stadt, Europa, Hauptstadt von Deutschland, …
  15. Hund Katze Maus Tiger Hamburg Berlin Madrid Amazonas Embedding Model

    0.6 0.3 0.1 … 0.8 0.5 0.3 … 0.4 0.2 0.9 … Hund Katze Tiger Maus Hamburg Berlin Madrid Amazonas Berlin: Stadt, Europa, Hauptstadt von Deutschland, …
  16. Hund Katze Maus Tiger Hamburg Berlin Madrid Amazonas Embedding Model

    0.6 0.3 0.1 … 0.8 0.5 0.3 … 0.4 0.2 0.9 … Hund Katze Tiger Maus Hamburg Berlin Madrid Amazonas
  17. Hund Katze Maus Tiger Hamburg Berlin Madrid Amazonas „Kannst du

    mir ein paar europäische Hauptstädte nennen?“ Embedding Model 0.8 0.5 0.4 … „Kannst du mir …“: Stadt, Europa, Hauptstadt, … FRAGE:
  18. Hund Katze Maus Tiger Hamburg Berlin Madrid Amazonas Embedding Model

    0.8 0.5 0.4 … „Kannst du mir …“: Stadt, Europa, Hauptstadt, … „Kannst du mir ein paar europäische Hauptstädte nennen?“ FRAGE:
  19. Hund Katze Maus Tiger Hamburg Berlin Madrid Amazonas Embedding Model

    0.8 0.5 0.4 … „Kannst du mir …“: Stadt, Europa, Hauptstadt, … Unser Ziel? „Kannst du mir ein paar europäische Hauptstädte nennen?“ FRAGE:
  20. Hamburg Berlin Madrid Amazonas Unser Ziel? Alles dafür tun, dass

    „Frage“ und „Antwort“ möglichst dicht beieinander liegen!
  21. man woman king queen queen – woman + man =

    king Ok, aber was ist jetzt das Tolle daran? Wir können damit rechnen und „verstehen“.
  22. man woman king queen Ok, aber was ist jetzt das

    Tolle daran? Wir können damit rechnen und „verstehen“. doctor – man + woman = ?
  23. man woman king queen Ok, aber was ist jetzt das

    Tolle daran? Wir können damit rechnen und „verstehen“. doctor – man + woman = nurse
  24. Transformer (LLM Layer) John wants his bank to cash the

    … ? Transformer (LLM Layer) John wants his bank to cash the … (verb) (verb) (John‘s) (finance) John wants his bank to cash the … ? (verb) (verb) (male) Enough context information to be able to „guess“ the next word. cheque ( ... ) Generative AI unter der Haube Tokens transformiert in Embeddings context aka hidden state
  25. Large Language Model John wants his bank to … Large

    Language Model Next Token increase cash close … Probability 0.2 0.4 0.1 … Next Token account cheque money order … Probability 0.15 0.75 0.05 … John wants his bank to cash the … Generative AI unter der Haube
  26. Generative AI unter der Haube CONTENT GENERIEREN CONTENT BESCHREIBEN CONTENT

    BEWERTEN CONTENT KORRIGIEREN Texte Bilder Sprache
  27. Generative AI unter der Haube CONTENT GENERIEREN CONTENT BESCHREIBEN CONTENT

    BEWERTEN CONTENT KORRIGIEREN Quellcode Diagramme Dokumente Tabellen Skripte Module/Klassen Abhängigkeiten Dokumentation Beschreibungen Änderungen Code Analysen Kritikalität Inkonsistenzen NaN Fehleranalysen „Bug“ Fixing Refactoring Bereinigung Synthetic Data Self Healing Texte Bilder Sprache
  28. Role Instruction Example 1 Context Example n Question Wie sollte

    ein guter Prompt aufgebaut sein? Who am I (or the assistant)? What is my intention? What are helpful examples? Are there any additional information? BTW: what is the task I ask for? GenAI als Sparringspartner
  29. You Can you suggest a good software architecture? LR You

    Acting as an experiences software architect, you want to develop a new SaaS application for healthcare billing with the following requirements [YOUR REQ. HERE]. Can you suggest an architecture and its main buidling block as a starting point for further discussion? The output format should be in markdown and follow the ok arch. design rules. The following examples should serve a a guide when generating the architecture proposal: … LR Output Example Context Question Role Acting as an experiences software architect, you want to develop a new SaaS application for healthcare billing with the following requirements [YOUR REQ. HERE] Can you suggest an architecture as a starting point for further discussion including the main building blocks ? The output should be in markdown and follow the ok arch. design rules. The following examples should serve as a guide when generating the architecture proposal: …. . GenAI als Sparringspartner
  30. User LLM Answer System Prompt User Prompt Prompting Patterns: •

    Pros and Cons • Stepwise Interaction • Flipped Interaction • Discussion of Experts / Think Tank GenAI als Sparringspartner Prompting Pattern
  31. User LLM Answer System Prompt User Prompt Judge AI Metrics

    LLM Metrics RAG Metrics GenAI als Sparringspartner AI as a Judge
  32. LLM User Retrieval Answer (Vector)DB Lookup Relevance Ranking Chunk Selection

    Content Trimming Prompt Augmention System Prompt User Prompt Context Augmention GenAI als Sparringspartner RAG Systems
  33. User System Prompt User Prompt Answer AI Agent(s) Workflow Automation

    Function Calling GenAI als Sparringspartner Agentic AI
  34. Agentic AI vs. Generic AI CONTENT GENERIEREN CONTENT BESCHREIBEN CONTENT

    BEWERTEN CONTENT KORRIGIEREN Quellcode Diagramme Dokumente Tabellen Skripte Module/Klassen Abhängigkeiten Dokumentation Beschreibungen Änderungen Code Analysen Kritikalität Inkonsistenzen NaN Fehleranalysen „Bug“ Fixing Refactoring Bereinigung Synthetic Data Self-Healing
  35. Agentic AI vs. Generic AI CONTENT GENERIEREN CONTENT BESCHREIBEN CONTENT

    BEWERTEN CONTENT KORRIGIEREN Quellcode Diagramme Dokumente Tabellen Skripte Module/Klassen Abhängigkeiten Dokumentation Beschreibungen Änderungen Code Analysen Kritikalität Inkonsistenzen NaN Fehleranalysen „Bug“ Fixing Refactoring Bereinigung Synthetic Data Self-Healing „intelligente“ Koordination
  36. Gen AI AI Agents Agentic AI CREATES Generiert neuen Content,

    wie z.B. Texte, Bilder oder Quellcode. ACTS Führt Aufgaben automatisch auf der Basis von Zielen und Umgebung aus. THINKS Koordiniert eine Reihe von KI-Agenten, um komplexe Ziele und Vorhaben zu erreichen.
  37. LLM Task Output Result Gen AI AI Agents Goal Agent

    Tools Output Objective Sub-Agents Output Tools & Memory Agentic AI
  38. AGENT MODEL SERVICES Long-Term Memory Vector Datastore Execution Loop in

    out Plan Action Memory [ST] Tools Content Data Devices Code Services Human APPLICATION Function Calling LLM A #2 A #n . . . FLIGHTWHEEL Agentic AI vs. Generic AI
  39. CONTENT GENERIEREN CONTENT BESCHREIBEN CONTENT BEWERTEN CONTENT KORRIGIEREN Quellcode Diagramme

    Dokumente Tabellen Skripte Module/Klassen Abhängigkeiten Dokumentation Beschreibungen Änderungen Code Analysen Kritikalität Inkonsistenzen NaN Fehleranalysen „Bug“ Fixing Refactoring Bereinigung Synthetic Data Self-Healing „intelligente“ Koordination Was kann Agentic AI? Friendly Reminder
  40. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation („The six core activities for software architects.“, ISAQB) Die 6 Kernaktivitäten Friendly Reminder
  41. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation CONTENT GENERIEREN CONTENT BESCHREIBEN CONTENT BEWERTEN CONTENT KORRIGIEREN
  42. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation CONTENT GENERIEREN CONTENT BESCHREIBEN CONTENT BEWERTEN CONTENT KORRIGIEREN
  43. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation „That sounds like a great project!“ #WTF*, du bist eine Maschine! *(what a terrible fraud)
  44. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation „…, let‘s define key … .“ Was heißt hier „let us“?
  45. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation potenzielle fachliche Anforderungen benennen
  46. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation potenzielle fachliche Anforderungen benennen Unique set of requirements?
  47. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation potenzielle fachliche Anforderungen benennen Some pleasant surprises included!
  48. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation potenzielle fachliche Anforderungen benennen Some pleasant surprises included, too!
  49. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation potenzielle nicht-fachliche & technische Anforderungen benennen large, many, fast, …
  50. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation nützliches Bonus Material (du hast zwar nicht gefragt, aber …) Reasoning via for, if, … Explaining „business“
  51. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation OMG! This is great, isn‘t it? nützliches Bonus Material (du hast zwar nicht gefragt, aber …)
  52. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Lücken & Mehrdeutigkeiten in Anforderungen erkennen
  53. Clarification of Requirements and Constraints Design of cross-sectional Concepts Evaluation

    of Architectures Communication of Architectures Design of Structures Support during Implementation Building Blocks identifizieren und Strukturen ableiten
  54. Clarification of Requirements and Constraints Design of cross-sectional Concepts Evaluation

    of Architectures Communication of Architectures Design of Structures Support during Implementation Building Blocks identifizieren und Strukturen ableiten Good Service Design? Good Module Design?
  55. Clarification of Requirements and Constraints Design of cross-sectional Concepts Evaluation

    of Architectures Communication of Architectures Design of Structures Support during Implementation Building Blocks und Strukturen visualisieren
  56. Clarification of Requirements and Constraints Design of cross-sectional Concepts Evaluation

    of Architectures Communication of Architectures Design of Structures Support during Implementation Building Blocks und Strukturen visualisieren
  57. Clarification of Requirements and Constraints Design of cross-sectional Concepts Evaluation

    of Architectures Communication of Architectures Design of Structures Support during Implementation Building Blocks und Strukturen visualisieren
  58. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Übergreifende Konzepte „entwerfen“ und dokumentieren
  59. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Communication of Architectures Evaluation of Architectures Support during Implementation Architektur bewerten gemäß einer vorgegebenen Methode 1. Present ATAM 2. Present business drivers 3. Present the architecture 4. Identify architectural approaches 5. Generate quality attribute utility tree 6. Analyze architectural approaches 7. Brainstorm and prioritize scenarios 8. Analyze architectural approaches (with the added) 9.Present results
  60. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Communication of Architectures Evaluation of Architectures Support during Implementation Architektur bewerten gemäß einer vorgegebenen Methode
  61. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Communication of Architectures Evaluation of Architectures Support during Implementation Architektur bewerten gemäß einer vorgegebenen Methode
  62. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Communication of Architectures Evaluation of Architectures Support during Implementation Architektur bewerten gemäß einer vorgegebenen Methode
  63. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Template für Architekturdokumentation vorschlagen
  64. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Dokumentation auf Basis vorhandener Quellen erzeugen
  65. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Dokumentation auf Basis vorhandener Quellen erzeugen
  66. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Dokumentation auf Basis vorhandener Quellen erzeugen
  67. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Dokumentation auf Basis vorhandener Quellen erzeugen
  68. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Source Code auf Basis von Beschreibungen generieren
  69. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Basis für den ersten Service generieren
  70. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: Value Objects und Builder Pattern
  71. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: Domain Events and Event Publisher
  72. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Bug Fixing: Tests laufen nicht (mehr) korrekt
  73. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Bug Fixing: Tests laufen nicht (mehr) korrekt
  74. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: aktuelle Java Features nutzen
  75. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: JPA einbinden (bisher nur In-Memory Persistence)
  76. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: JPA Testing via Quarkus & Testcontainer
  77. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: Bootstraping mit generierten Demo Daten
  78. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: weiteren Service erzeugen
  79. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: weiteren Service erzeugen
  80. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: weiteren Service erzeugen
  81. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: Docker Capabilities hinzufügen
  82. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: Docker Capabilities hinzufügen
  83. Clarification of Requirements and Constraints Design of cross-sectional Concepts Design

    of Structures Evaluation of Architectures Communication of Architectures Support during Implementation Refactoring: Docker Capabilities hinzufügen
  84. Support during Implementation Clarification of Requirements and Constraints Design of

    cross-sectional Concepts Design of Structures Evaluation of Architectures Communication of Architectures
  85. „It is comparatively easy to make computers exhibit adult level

    performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.“ ( Moravec's paradox, 1988)
  86. „We know how to read a situation, body language and

    gauge appropriate behaviour. That is an area where AI is currently very poor at.“ ( Avraham Poupko, 2024)
  87. vs.

  88. genAI Mensch günstig schnell skalierbar antwortet „fragt“ scheut das Risiko

    kennt „Fakten“ teuer langsam einmalig fragt fühlt antwortet liebt das Abenteuer versteht Kontext
  89. genAI Mensch Der Mensch versteht Zusammenhänge und kann dadurch kontext-spezifisch

    entscheiden. Die genAI (er)kennt Fakten und Muster und kann daher statistisch relevant Antworten geben.
  90. Es werden keine konstruktiven Fragen gestellt. Es findet kein wirklicher

    Dialog statt. Es werden keine Entscheidungen getroffen. Es findet keine Anpassung an den Kontext statt. Old John versucht nicht den Kontext zu verstehen. Seine Argumente sind sehr allgemein gehalten. Fragen zu den Anforderungen des Systems fehlen völlig. Es werden keine Erfahrungen in die Diskussion eingebracht. Annahmen werden nicht hinterfragt oder gar korrigiert. Und genau das macht uns Menschen aus!
  91. GenAI ist gut darin … • Texte für verschiedene Zielgruppen

    aufzubereiten • uns als Sparrings-Partner zu dienen • mit uns Brainstorming zu betreiben • Design-Optionen aufzuzeigen • Architektonische Trade-offs zu erklären • für uns Architectural Decision Records zu verfassen • parallel mehrere Szenarien durchzuspielen • und vieles mehr … GenAI kann uns aber unterstützen!
  92. vs.

  93. „LLM is a tool and not a replacement for your

    critical thinking. Always use it as a colaborator not as a decision-maker.“ vs. und Adel Ghlamallah, Software Architect, Book Author
  94. KI-Tools sind deine neuen „Kollegen“, die dich gerne bei deiner

    Architekturarbeit unterstützen. Nicht mehr, aber auch nicht weniger. vs. und Lars Röwekamp, Software Architect, AI Euphorist & Realist
  95. GenAI ist ein Tool. Menschen nutzen Tools. GenAI kennt „Fakten“.

    Menschen verstehen. GenAI unterstützen. Menschen entscheiden.
  96. „The software architect will not be replaced by AI but

    by another software architect who makes smart use of AI.“
  97. #WISSENTEILEN #WISSENTEILEN BILDNACHWEIS Folie 21: © photoplotnikov - istockphoto.com Folie

    23: © Mix und Match Studios - shutterstock.com Folie 23: © Mix und Match Studios - shutterstock.com All other pictures, drawings and icons originate from • pexels.com, • pixabay.com, • unsplash.com, • flaticon.com or are created by my own.