Paris AI Forum ideas

added prompt engineering slide

TO-DO. start on demo slides and notes

01-claude-ocp/node_modules/.package-lock.json

@@ -1,6 +0,0 @@
-{
-  "name": "01-claude-ocp",
-  "lockfileVersion": 3,
-  "requires": true,
-  "packages": {}
-}

01-claude-ocp/package-lock.json

@@ -1,6 +0,0 @@
-{
-  "name": "01-claude-ocp",
-  "lockfileVersion": 3,
-  "requires": true,
-  "packages": {}
-}

01-claude-ocp/package.json

@@ -1 +0,0 @@
-{}

01-claude-ocp/presentation.md

@@ -42,7 +42,7 @@ What is MCP?
 <!-- new_line -->
 <!-- alignment: center -->
 <!-- font_size: 2 -->
-**MCP gives tools to AI**
+MCP gives tools to AI
 ===
 ![image](robot.png)
 <!-- no_footer -->
@@ -123,9 +123,11 @@ sequenceDiagram
     LLM->>ArgoCD: Get GitOps application
 	LLM->>Git: Get Git repo
     LLM->>LLM: Plan fix
+	LLM->>You: Here is the plan. Requesting Approval
+	You->>LLM: Approved
     LLM->>Git: Push fix to Git
     LLM->>OCP: Verify change applied
-	LLM->>You: The issue was X. I resolved by doing Y
+	LLM->>You: Applied Succesfully
 ```
 <!-- pause -->
 <!-- column: 1 -->
@@ -135,6 +137,9 @@ sequenceDiagram
 <!-- new_line --> 
 <!-- new_line -->
 <!-- new_line -->
+<!-- new_line --> 
+<!-- new_line -->
+<!-- new_line -->
 ![image:width:100%](homerspin.gif)
 
 
@@ -176,14 +181,32 @@ Tools for the demo
 <!-- new_line -->
 * `argocd-mcp-server` - ArgoCD (aka Openshift GitOps) application management
 <!-- new_line -->
-* `minio-mcp-server` - S3-Compatible Object storage operations
 
 <!-- no_footer -->
 <!-- speaker_note: Explain the tools in the wider context -->
 <!-- speaker_note: openshift - complete CRUD access -->
 <!-- speaker_note: gitea - ability to read and commit, create repos etc. -->
 <!-- speaker_note: ArgoCD - add and keep deployments in a specified state -->
-<!-- speaker_note: MinIO - self-hosted S3 -->
+<!-- end_slide -->
+
+
+
+
+
+<!-- font_size: 4 -->
+Prompt Engineering
+===
+<!-- font_size: 2 --> 
+Intent vs. Instruction
+===
+<!-- font_size: 2 --> 
+![image:width:100%](table.png)
+
+<!-- pause -->
+<!-- new_line -->
+<!-- new_line -->
+<!-- font_size: 2 --> 
+Ambiguity is the enemy of automation. If you don’t define ‘safe’… the AI will.
 <!-- end_slide -->
 
 

01-claude-ocp/test.md

@@ -1,7 +0,0 @@
-![image:width:50%](happy.gif)
-
-```bash +exec_replace
-mpv --no-config --vo=kitty  --profile=sw-fast --really-quiet  happy.webm
-```
-
-

02-paris-ai-forum/IDEAS.md

@@ -0,0 +1,38 @@
+# Paris AI Forum: Hyper-Agentic Demo Ideas
+
+**Goal:** Demonstrate the difference between "Chatting with AI" (Hope) and "Agentic Workflows" (SOP/Action).
+**Constraints:** Must be impossible/hard for a standard web LLM. Must show tools, file manipulation, or system state changes.
+
+---
+
+## 1. The "Instant Briefing" (Research Loop)
+**Concept:** Turn a vague question into a tangible artifact in < 2 minutes.
+**The Prompt:** "I have a meeting with the CTO about [Emerging Tech Topic] in 5 minutes. Build me a briefing pack."
+**The Agent Run:**
+1.  **Search:** Scours the web for the latest 3-5 credible sources (filtering spam).
+2.  **Synthesize:** Writes a structured `EXECUTIVE_BRIEF.md` (Key Trends, Risks, Opportunities).
+3.  **Generate:** Uses `presenterm` (or `pandoc`) to auto-generate a PDF/Slide deck from that markdown.
+4.  **Deliver:** Pushes the PDF to a shared repo or Slack channel live.
+**The "Wow":** Moving from "text output" to "file delivery" without human copy-pasting.
+
+## 2. The "Infrastructure Surgeon" (Self-Healing)
+**Concept:** AI that can touch the metal.
+**The Setup:** A broken service on a demo OpenShift namespace (e.g., a pod crash loop or wrong env var).
+**The Prompt:** "The `demo-app` is down. Fix it."
+**The Agent Run:**
+1.  **Diagnose:** Runs `kubectl get pods`, `kubectl logs`, `curl localhost`. Identifies the error (e.g., "Port 8080 refused").
+2.  **Reason:** "Config map points to 8081, app listens on 8080."
+3.  **Act:** Runs `kubectl patch` or edits the deployment manifest.
+4.  **Verify:** Runs `curl` again to confirm 200 OK.
+**The "Wow":** Breaking the "Read-Only" barrier. The AI *changed state* to solve a problem.
+
+## 3. The "Meeting Artifact Generator" (System Integration)
+**Concept:** Turning unstructured voice/text into structured systems of record.
+**The Setup:** A raw transcript or audio file of a chaotic meeting.
+**The Prompt:** "Process this meeting dump. Update our systems."
+**The Agent Run:**
+1.  **Parse:** Extracts Action Items, Decisions, and Deadlines.
+2.  **Ticket:** Creates actual Issues in Gitea (or Mock Jira) for each action item.
+3.  **Schedule:** Generates an `.ics` calendar invite for the follow-up.
+4.  **Notify:** Sends a summary message to a Telegram/Slack group tagging the owners.
+**The "Wow":** The AI interacting with *multiple* disparate APIs (Git, Calendar, Chat) to orchestrate a workflow, not just summarize text.

SCRATCHPAD.md

@@ -0,0 +1,114 @@
+# Presentation Ideas Scratchpad
+
+Raw ideas and scenarios for future AI presentations. Capture now, refine later.
+
+---
+
+## Presentation #1: MCP in Practice (OpenShift/GitOps)
+
+**Audience:** Internal Axway team (small, reports to Conan)
+**Tone:** Serious content, Simpsons humor, relatable frustration → payoff arc
+
+### Core Thesis
+The value of MCP isn't "AI can run kubectl" — it's **flipping the discovery loop**.
+
+- **Without MCP:** Human works *for* the AI (copy-paste errors, run commands, provide context, be the middleware)
+- **With MCP:** AI works *for and with* the human
+
+### Narrative Arc
+
+1. **What is MCP?** — Open standard, Anthropic 2024, gives tools to AI
+2. **Why MCP? (Without)** — Grandpa Simpson walks in, sees GPT, walks out
+   - *Pitch:* Complex problems without tool-bearing AI = frustrating, time-consuming, upside-down. People "bounce off" AI for real work.
+3. **Why MCP? (With)** — Setup for the payoff
+4. **Tools for the demo** — openshift-mcp, gitea-mcp, argocd-mcp
+5. **Prompt Engineering** — "Intent vs. Instruction" / "Ambiguity is the enemy of automation. If you don't define 'safe'… the AI will."
+6. **DEMO** — Three scenarios, progressively impressive:
+   - **Scenario 1:** "Run a healthcheck of my cluster"
+     - AI examines multiple resources
+     - Key: produces **INSIGHT**, not just dry information
+   - **Scenario 2:** "Is my NAS slower than it should be?"
+     - AI launches ephemeral test pod (permitted via system prompt)
+     - Reasons against hardware expectations (2-bay 5400RPM, 1Gb network)
+   - **Scenario 3:** "Lock down security in this namespace"
+     - AI investigates current posture
+     - Plans path to privileged-v2
+7. **Summary** — *(to be drafted)*
+   - Land the loop-flip insight
+   - "What you didn't see" — all the commands, errors, iterations handled silently
+8. **Closer** — Homer spinning in chair (callback)
+   - "Taking us out of the discovery loop leaves us free to be productive elsewhere"
+
+### Demo Technique
+During demo: **actually do other work** while AI runs. Makes the "frees you up" point concrete — audience watches AI work, watches presenter work, realizes neither needed the other.
+
+### Key Moments
+- "Holy shit, it just figured it out" — if presenter had these moments building the demo, audience will too
+- Transition from "tech demo" to "I want this"
+
+---
+
+## Presentation #2: Ideas (TBD)
+
+**Context:** Axway heritage — Sopra Group → Sopra Steria. Enterprise software, API management, governance. Audience cares about security, compliance, "don't let the AI rm -rf."
+
+### Option A: Security / AI Gateway
+- Take the security line further than system prompts
+- Show how an **AI Gateway** solves remaining concerns:
+  - Rate limiting
+  - Action allowlists
+  - Audit trails
+  - Policy enforcement
+- Narrative: "Your enthusiastic AI colleague needs guardrails in enterprise"
+
+### Option B: Hyper-Agentic First, Then Governance
+- Go full agentic — show the power (browser control, cross-platform nodes, compound tool use)
+- *Then* pivot: "And this is how you keep your enthusiastic colleague from doing something you didn't want"
+- AI Gateway as the answer to the "wait, but what if..." concerns the demo just raised
+
+### Potential Beats
+- Show the Apple Notes/Reminders cross-platform demo (container → Mac node → AppleScript)
+- Browser automation
+- Multi-step reasoning with real consequences
+- Then: governance layer, audit, policy, enterprise controls
+
+---
+
+## Loose Ideas (Capture as they come)
+
+### Cross-Platform Automation Demo (Apple Notes/Reminders)
+**Already documented above — usable for Presentation #2 "hyper-agentic" section**
+
+- AI in secure container reaches out to Mac node
+- Original tool (memo CLI) too interactive
+- AI adapts: discovers AppleScript as native alternative
+- Creates notes/reminders via osascript
+- "Figures it out on its own" moment
+
+### Meta-Demo: The Session Itself
+The conversation that led to this scratchpad is itself demo-worthy:
+- Casual experiment → discovery
+- Presenter mentions "might use this for a demo"
+- AI offers to capture it (not requested)
+- Writes structured doc to the right repo
+- That's working *with* someone, not commanding a tool
+
+---
+
+*Add ideas below as they emerge*
+### The 3x Multiplier (Key Talking Point)
+
+Common framing: "AI will replace all our jobs" — scary, paralyzing, defensive.
+
+Better framing: **"The person leveraging AI will outcompete the person who isn't."**
+
+**Example from practice:**
+- Conan wanted to do a manual task (building muscle memory)
+- Asked AI: "While I'm doing this, write an MD with X, Y, Z bullet points and anything else you see fit"
+- Result: well-structured doc + *additional valuable points the AI added on its own*
+
+**The math:**
+- 2x = doing two things at once (parallel work)
+- 3x = parallel work + AI's *additive* contribution (compounded output)
+
+This is the multiplier people miss. You're not just parallelizing — you're getting value you didn't explicitly ask for.