AI Assisted Software Development - Portland Java User Group - 2026-04-14

Portland Oregon
Portland Java User Group
April 2026

Transcript

1. A.I. Assisted Software Development Sean Sullivan April 14 2026 Portland Java User Group
2. About me software engineer Portland Oregon skier
3. A.I. Milestones 🔵 November 2022: ChatGPT launched 🔵 November 2024: MCP introduced by Anthropic 🔵 February 2025: Claude Code launched
4. Tidal wave
5. $650 billion in 2026 Capex spending
9. Building without AI right now is not a smart use of your time. Full stop. Aruna Kumar Building with AI means genuinely re-learning and re-tooling — not copy-pasting prompts and hoping.
10. This presentation Tutorial Deep dive A.I. concepts A.I. tools
11. Topics 🔶 Software Development 🔶 AI boom / AI hype 🔶 Coding with AI 🔶 Claude Code and beyond
17. May 2025
18. Stack Overflow 2026
19. The rise of generative AI coding assistants has fundamentally altered how software gets built Sam Keen alteredcraft.com
20. brokk.ai
21. “Progress happens slowly, then all at once”
23. DORA report 2025 “AI adoption has become nearly universal”
24. “Something big is happening” Matt Shumer 9 February 2026 https://shumer.dev/something-big-is-happening
25. Augment your existing software process vs Build a software factory
29. • Gemini • Claude • Copilot • Codex • Amp
30. Thank you very much, Mr. Roboto. For doing the jobs that nobody wants to
31. CLASSIC CODING AGENTIC CODING
34. Things I work on • Java frameworks • Gradle plugins • Testing frameworks • OpenRewrite recipes • Kotlin, Scala • Migrations
36. My approach 🟢 make small changes 🟢 write tests before refactoring 🟢 separate behavior changes from structural changes
37. AI assisted development • implement new features • adding test cases in a test suite • troubleshooting bugs in legacy code • updating engineering docs • code review • fixing a broken build
38. Pair programming?
41. 2025 entrepreneur.com
43. “Vibe coding and AI-assisted engineering are two distinct mindsets” Addy Osmani
44. “Vibe coding is top-down and exploratory: you start with a broad idea and let the implementation emerge” Addy Osmani
45. “Al-assisted engineering is systematic and iterative” Addy Osmani
46. vibe coding optimizes for velocity in the short term AI-assisted engineering optimizes for sustained velocity and reliability
48. 🔷 Craft Oriented developers 🔷 Delivery Oriented developers Sam Keen
51. LLM MCP A2A UCP ACP AP2 RAG AGI ASI
52. Model Context Protocol (MCP) is an open standard that defines how applications share context with large language models (LLMs). MCP provides a standardized way to connect AI models to different data sources and tools, enabling them to work together more effectively.
53. AI superintelligence (ASI) is a hypothetical, future form of artificial intelligence that surpasses human intelligence across all fields, including creativity, general wisdom, and social skills.
55. - CLAUDE.md - skills - plugins - hooks - routines
56. Example
58. --dangerously-skip-permissions Do you want to run this command?
59. Auto mode 24 March 2026
60. claude --enable-auto-mode
63. does Claude Code support LSP?
65. Claude Code Setup 1. Enable the LSP Tool
66. 2. Install the Java Language Server brew install jdtls Claude Code Setup
67. 3. Install the Claude Code Plugin Claude Code Setup /plugin install jdtls-lsp@claude-plugins-official
68. does GitHub Copilot support LSP?
70. GitHub Copilot -- March 2026
75. author: George Fahmy
76. author: George Fahmy
77. author: George Fahmy
79. private runtime for autonomous AI agents
80. provides sandboxed execution environments that protect your data, credentials, and infrastructure
81. agents run with exactly the permissions they need and nothing more
83. nono.sh Nono project
85. Docker blog
86. Docker Sandboxes docker.com
88. blog.nvidia.com
89. blog.nvidia.com During inference, an AI receives a prompt that it translates into a series of tokens. The model processes these input tokens, generates its response as tokens and then translates it to the user’s expected format.
90. blog.nvidia.com Some token pricing plans offer users a set number of tokens shared between input and output.
91. Claude Code consumes tokens for each interaction. Costs vary based on codebase size, query complexity, and conversation length. The average cost is $6 per developer per day, with daily costs remaining below $12 for 90% of users.
92. For team usage, Claude Code charges by API token consumption. On average, Claude Code costs ~$100-200 / developer per month with Sonnet 4.6
94. Managing costs effectively - track token usage - set spending limits - set rate limiting - reduce costs with context management - reduce costs with model selection
95. Rate limiting - Tokens per minute (TPM) - Requests per minute (RPM)
96. Token costs scale with context size the more context Claude processes, the more tokens you use
97. /cost /clear /compact
100. cnbc.com
103. NY Times April 2026
105. anthropic.com
106. anthropic.com Context engineering refers to the set of strategies for curating and maintaining the optimal set of tokens (information) during LLM inference, including all the other information that may land there outside of the prompts.
107. anthropic.com - progressive disclosure - compaction - agentic memory - sub-agent architecture
109. RPI pattern Research -> Plan -> Implement
110. Planning mode
116. 🚧 guardrails for humans 🚧 guardrails for AI coding agents
117. 🚧 Guardrails for a GitHub repository - Github Push Protection - GitHub Protected Branches - GitHub PR Status Checks - CODEOWNERS file
118. 🚧 Guardrails : hooks - git commit hooks - Claude Code hooks - Copilot CLI hooks
119. 🚧 Deterministic guardrails for Java - PMD - Checkstyle - JaCoCo - SpotBugs - ArchUnit - maven-enforcer-plugin - revapi plugin - Java compiler options
121. “Your coding agent is only as good as what it knows about your code” moderne.ai
123. Ben Hall
124. Ben Hall
125. Ben Hall
126. Ben Hall
127. Ollama blog
128. apple.com Apple Silicon
130. Working with agentic tools really is a trust game. You're relying on them to actually do the right thing, not just sound like they are. That trust is hard to build and easily eroded; something to keep in mind as we hand them more and more autonomy. Trust Gunnar Morling
132. October 2025
134. Cassandra Expert skills
136. DORA report 2025
137. DORA: Measuring Software Delivery throughput
138. DORA: Measuring Software Delivery instability
139. DORA 2025
140. DORA AI Capabilities Model
141. DORA AI Capabilities Model
144. cnbc.com
145. 🔷 OpenClaw 🔷 NemoClaw 🔷 NanoClaw 🔷 ClawRunr
146. Java
147. ClawRunr
148. ClawRunr
153. • don’t outsource thinking • human in the loop • move fast, don’t break things • use guardrails
154. “with great power comes great responsibility”
155. The End