Mastodawn

New post: building “articulate,” a YouTube-to-article tool, using six different AI workflows along the way.

Covers: AI-assisted search for research, Copilot Cloud agent scaffolding a Spring Boot 4 / Spring AI app from a screenshot and docs, working through YouTube transcript-access blockers, a docs-first prompting pattern for integrating Supadata, Gemini Flash for generation, and prompt tuning via side-by-side comparison.

End result: 60-min videos become 5-min reads.

https://robintegg.com/2026/06/12/youtube-videos-to-blog-articles-with-ai.html

From YouTube Videos to 5-Minute Blog Articles with AI

Java Developer by trade, AI Explorer by choice, and a Spring Boot Wizard at heart. I specialize in building robust, scalable applications and systems while actively integrating the power of Generative AI into modern software workflows. Beyond the IDE, I'm a firm believer in sharing knowledge and fostering the developer community. I thrive at the intersection of enterprise reliability and cutting-edge automation, always looking for the next challenge that requires a mix of clean code and creative intelligence. ☕🚀

Robin Tegg

Robin Tegg Jun 3

Most agent mistakes in multi-step workflows are transition mistakes. The model can usually call an endpoint. The harder part is knowing if that call is valid for the current state.

A static tool catalog doesn’t encode runtime state. A hypermedia response can. That shifts decision pressure from prompt engineering into server-published affordances.

HAL-FORMS in Spring HATEOAS is a practical starting point for Java/Spring teams.

https://robintegg.com/2026/06/03/hypermedia-apis-for-autonomous-agents.html

Hypermedia APIs for Autonomous Agents

An extended write-up of the hypermedia API hypothesis for agent workflows with Spring HATEOAS and HAL-FORMS.

Robin Tegg

Robin Tegg Jun 3

Great news if this is true? Free? Hopefully won’t need to worry about using my copilot tokens? https://x.com/israfill/status/2062108803437404520?s=46

Isra (@israfill) on X

Microsoft launched their own coding AI and it's FREE inside GitHub Copilot starting today 😳 not rented from OpenAI. built by Microsoft. it's called MAI-Code-1-Flash and it's rolling out to ALL copilot tiers right now part of microsoft's new MAI model family: -

X (formerly Twitter)

Robin Tegg Jun 1

My May 2026 AI retrospective is out. Practical notes on how I actually use AI day-to-day as a developer.

The big shift: discovery now has its own phase before planning. Load ticket context, run discovery, write plan, implement.

Also: standardising my prompts and agents into an installable toolkit repo, thoughts on GitHub Copilot's new Agents Window and integrated browser, and links worth your time.

https://robintegg.com/2026/06/01/may-2026-ai-retrospective.html

#AI #DevTools #GitHub #Copilot #SoftwareDevelopment

May 2026 AI Retrospective

Discovery-first workflows, toolkit curation and Copilot still rules.

Robin Tegg

Robin Tegg May 25

Local folders were drifting and every new repo meant rebuilding the same Copilot setup from scratch.

So I built ai-toolkit: a versioned repo of prompts, agents, skills and custom instructions, organised as installable bundles.

"I can version the full set, review changes in Git, and avoid copy-pasting snippets between random local directories."

Install in seconds with JBang:
jbang AiToolkit.java install discovery

https://robintegg.com/2026/05/24/curating-your-own-ai-toolkit-repository.html

Curating Your Own AI Toolkit Repository

Why I created ai-toolkit to manage prompts, agents, skills, and instructions, and how a published catalog plus JBang installer helps me start faster.

Robin Tegg

Robin Tegg May 13

Just published: A Discovery Agent for your VSCode workflow 🔍
Plan mode jumps to solutions. Ask mode is too shallow. Discovery is the phase missing between them — investigate the codebase before committing to a plan.
The agent runs 5 phases: Problem Framing → Context Discovery → Alignment → Synthesis → Refinement. Output is a Discovery Brief, not a plan. The input to a plan.
🔗 https://robintegg.com/2026/05/13/a-discovery-agent-for-your-vscode-workflow.html
#VSCode #AIAgents #DevTools #CopilotChat

A Discovery Agent for Your VSCode Workflow

Robin Tegg

Robin Tegg May 6

My April 2026 AI Retrospective is up 🧵
Covering: persistent knowledge bases (wiki-over-RAG via Karpathy’s gist), hands-on time with Granola for meeting notes, evaluating Emdash & Supacode for parallel agent workflows, and the multi-repo problem that nobody’s solved well yet.
Also: two posts on AI writing consistency and MCP with Spring AI.
👉 https://robintegg.com/2026/05/06/april-2026-ai-retrospective.html
#AI #Java #SpringAI #Agents #Retrospective

April 2026 AI Retrospective

Consolidation, knowledge bases, and trying out the new wave of harness tools, a personal retrospective on April 2026.

Robin Tegg

Robin Tegg May 6

New article: Exploring MCP with Spring AI 🧵

MCP (Model Context Protocol) is becoming a key primitive in the AI tooling landscape — giving LLMs structured access to your application via Tools, Resources, Prompts, and Apps.

Spring AI wraps the MCP Java SDK with Boot starters and annotations that feel natural to any Spring developer.

https://robintegg.com/2026/04/26/exploring-mcp-with-spring-ai.html

#Java #SpringAI #SpringBoot #MCP #AI

Exploring MCP with Spring AI

MCP gives your LLM access to external systems via your agent. Spring AI makes it straightforward to build MCP servers in Java using annotations.

Robin Tegg

Robin Tegg Apr 23

"The writing skill helps because it turns 'style' from a vague intention into explicit rules."

As a developer writing long-form content with AI, I kept ending up with inconsistent, AI-flavoured posts. My fix: a reusable voice skill that the agent applies every time I draft or edit.

Split structure rules from voice rules. Composable, maintainable, and it actually keeps my writing sounding like me.

https://robintegg.com/2026/04/23/consistent-voice-with-a-writing-skill.html

Why a Consistent Voice Matters (and How a Writing Skill Helps)

A short note on why voice consistency matters as content output grows, and how a writing skill plus an agent-driven quiz loop keeps posts aligned.

Robin Tegg

Robin Tegg Apr 6

This month has been all about harnesses and orchestration — building my own agent harness on the Copilot SDK to understand the ecosystem from the inside out.
Also sharing what’s changed about how I approach research and why I’m building a persistent knowledge base inside the repo itself.
https://robintegg.com/2026/04/05/march-2026-ai-retrospective.html
#AI #Java #GitHubCopilot #SoftwareDevelopment

March 2026 AI Retrospective

Harnesses, orchestrators, and the growing ecosystem of AI workflow tooling, a personal retrospective on March 2026.

Robin Tegg