iAgentX for IBM i — Demo Speech Script

00

Hook

Opening — set the scene

"This is what happens when you ask your AI assistant about IBM i code today. Error. Error. Error. It's not that the AI is bad — it's just completely blind to your system. That's the problem we're solving."

💡 Let the terminal animation play out — the comedic timing of each error landing one by one does the work. Pause briefly before the last line.

~20 sec

01

Title — Bring AI Intelligence to Your IBM i

What iAgentX is

"iAgentX for IBM i bridges that gap. It connects Claude Code, GitHub Copilot, and IBM Bob directly into your IBM i — reading source members, querying DB2, pulling job logs. Free to use. Secure. Available right now on the VS Code Marketplace."

💡 Keep it short — the title slide is a visual moment. Let the terminal animation on the right breathe.

~25 sec

02

The Challenge

Why AI can't help IBM i developers — yet

"The real issue is awareness. Copilot can actually call Code for IBM i extension commands — but only if you explicitly tell it those tools exist. Without that context, it won't go looking. Claude Code and IBM Bob have no IBM i access at all. And the one vendor adding real integration — IBM Bob — is making it a paid feature launching June 24th. iAgentX solves this by automatically giving the AI full context of every available IBM i tool — no prompting required."

💡 Pause on "only if you tell it" — this is the nuance that surprises developers who tried Copilot and assumed it worked automatically.

~40 sec

03

Market Gap

How the options stack up

"Here's how the options stack up. Copilot gets partial credit — it can reach IBM i if you prompt it correctly. But that's the catch: you have to know which tools exist and mention them every time. Bob today — nothing. Bob after June 24th — some access, at a premium. iAgentX — full access, proactive tool context, free, today."

💡 Run your finger along the last column — the green checkmarks speak for themselves. Don't over-explain the table.

~30 sec

04

Introducing iAgentX

The solution

"iAgentX is a VS Code extension. Install it — that's it. It auto-configures Claude Code, Copilot, and IBM Bob on first launch. No manual MCP setup. And crucially, it reuses the SSH connection you already have from Code for IBM i — no new credentials, no extra setup."

💡 Emphasise "reuses your connection" — this is the biggest friction-removal point for enterprise teams with strict credential policies.

~30 sec

05

Architecture

How it all connects

"Under the hood it's a lightweight MCP server running inside the VS Code extension host, bound only to localhost. AI agents call it over HTTP — it forwards requests through Code for IBM i's existing SSH session. Nothing is exposed to the network. One server instance serves all your VS Code windows."

💡 Point to the diagram layers as you speak. You don't need to explain every layer — just trace the path: AI → iAgentX → SSH → IBM i.

~35 sec

06

Security

Built-in, not bolted-on

"Security was designed in from day one. The AI runs under your own IBM i user profile — no service accounts, no elevated privileges. SQL is strictly read-only — SELECT, WITH, VALUES only — enforced at the extension layer, not just a convention. CL commands are allowlisted to display and diagnostic verbs only. And editor writes only touch files you already have open."

💡 This slide often generates questions in enterprise settings. Be ready for "what if someone modifies the allowlist?" — answer: it's auditable source code, enforced server-side, and licensed under Elastic License 2.0 so the community can review it.

~40 sec

07

29 IBM i Tools

At your AI's fingertips

"That translates to 29 purpose-built tools — source members, objects, IFS, SQL, job logs, spool files, CL diagnostics, and full VS Code editor integration. Every tool available to all three AI clients simultaneously, the moment you install."

💡 You don't need to read every chip — gesture at the categories and let the audience scan. Keep the pace up here, the demos are next.

~25 sec

08

Demo 1 — Job Stuck in Message Wait

Diagnose a silent batch job with ibmi_find_jobs + ibmi_get_job_log

"First demo. A batch job went completely silent — no error, no output, just hanging. I ask Claude Code: 'Is any of my jobs in message wait? Get the root cause.' It calls ibmi_find_jobs to scan for MSGW jobs, then pulls the full job log with ibmi_get_job_log. Root cause in seconds: the job is actually healthy — it completed two SQL steps successfully, then sent an inquiry message to QSYSOPR waiting for a human decision before running the extract. One prompt. Two tool calls. Full picture."

💡 If running live — paste the prompt and point out the two tool calls appearing in sequence in chat. That two-step chaining ("find it, then read it") is the moment the audience sees the AI reasoning, not just looking something up.

~45 sec

09

Demo 2 — IFS Directory Analysis

Explore & explain an IFS directory with ibmi_list_ifs_directory + ibmi_get_ifs_file

"Demo 2. A developer inherits a home directory full of mystery files — shell scripts, Python, configs, logs. One prompt: 'What are all these files in my home directory and what do they do?' It lists the IFS directory, reads the key scripts and config files, and gives a plain-English breakdown of every object — including which ones are active cron jobs and when they run. What would normally take 30 minutes of cat-and-grep: done in one shot."

💡 Emphasise the two-step: list first, then selectively read files for context. The AI is making a judgment call about which files are worth reading — that's what makes it genuinely useful rather than just a grep wrapper.

~40 sec

10

Demo 3 — Source Member Analysis

Explain a CL program & generate a flow diagram with ibmi_get_source_member

"Demo 3. There's an 89-line CL program nobody has touched in 3 years. I ask: 'Analyse BATCHFTP in QCLLESRC. Explain what it does and create a program flow diagram.' It reads the source member directly and returns a clean explanation — plus an ASCII flow diagram you can share with the team right now. No copy-paste. No 5250 session. No manual reading every MONMSG."

💡 The flow diagram is the visual payoff here — point to it on screen. Teams often screenshot this immediately for documentation. Mention that the diagram is generated from the actual code, not from memory.

~35 sec

11

Demo 4 — Row Count Analysis

Cross-table row count across a full schema with ibmi_run_sql

"Demo 4 — a SQL power move. I ask Claude Code: 'Get row counts for all main tables and their history tables in XYZLIBRARY.' It calls ibmi_run_sql, queries the system catalog, dynamically builds the COUNT queries for every table pair, and returns a clean result — 18 table pairs in seconds. Manually writing that SQL would take most developers 20 minutes."

💡 If running live — paste the prompt and let it run in real time. The tool call appearing in chat is the "wow moment." Point it out explicitly. The key thing: the AI inferred the naming convention (trailing H) from the prompt alone.

~40 sec

12

Demo 5 — Fix FK Hierarchy Function

Debug & fix a SQL table function with ibmi_get_active_editor + ibmi_run_sql

"Final demo — a real debugging scenario. A SQL table function is returning level 1 for every table, ignoring all the FK relationships. I ask Claude Code to investigate. It reads the open source member directly from my editor using ibmi_get_active_editor — no copy-paste — then queries the FK catalog. Root cause: wrong column name in the recursive CTE join. It finds it, fixes it, and gives me the CREATE statement ready to deploy to library DK. That's the key point — the AI saw my open file without me doing anything."

💡 Stress "without me doing anything" — the audience should feel the friction that just disappeared. This is the most technically impressive moment of the full demo. End here with a beat of silence before moving to the impact slide.

~55 sec

13

Before & After iAgentX

Impact on the day-to-day workflow

"The difference in workflow is significant. Before: manual copy-paste, context-switching between VS Code, a 5250 terminal, and a chat window — the AI working completely blind. After: one chat window, the AI reads your source directly, sees your open editors in real time, has full access to job logs and diagnostics. Zero extra credentials. Twenty-nine tools. Three AI clients. Zero cost."

💡 Read the four stat boxes slowly and with weight — "zero dollars" always lands well as the final note before the closing slide.

~40 sec

14

Get Started

Up and running in 3 steps

"Three steps. Install Code for IBM i if you haven't already. Install iAgentX for IBM i from the VS Code Marketplace or Open VSX Marketplace. Open Claude Code — it's already configured. That's it. The extension is free to use under Elastic License 2.0. Links are on screen. Happy to take questions."

💡 End on the closing tagline on screen — "Your IBM i. Your AI. Connected." — then go quiet and open the floor. Don't rush past it.

~30 sec