signal4i.ai
SIGNAL4i
Agentic Tech Map
Technical Reference  ·  Signal4i  ·  May 2026

The Agentic
Tech Map

Six domains. Three tiers. The complete agentic technology landscape mapped to IBM i specifically — what's foundational to any production deployment, what's emerging, and where most implementations fail.

This is not a tutorial. It is not a vendor comparison. It is a map — the one that tells you what connects to what, what's foundational versus emerging versus noise, and how to sequence your attention.

94%
of GenAI pilots are failing
MIT / Fortune, Aug 2025
1%
at full deployment — those see 3–5× productivity
McKinsey Superagency 2026
77%
have no committed agentic AI strategy
ManpowerGroup, 2026
362
documented AI incidents in 2025, +55% YoY
HAI 2026 AI Index
How tiers work

Foundational — required working knowledge regardless of where your organization is in the transition. These domains underpin every agentic deployment. Emerging — direction is clear, path is still forming. Vocabulary and evaluation posture now, implementation later. Critical Gap — the domain most commonly underbuilt until a deployment fails. Treat it as infrastructure, not afterthought.

The IBM i lens

Each domain includes an IBM i-specific read: what it means for environments running RPG, DB2 for i, and a platform with object-level security and journal-based audit trails built in since 1988. Where IBM i shops have a structural advantage, this map names it. Where the gap needs closing, it names that too.

Solution architecture — how agents connect to IBM i
Orchestrator agent
Python · LangChain / Agno · Routes · Plans · Coordinates domain agents
Think ↓
Domain agents
IBM Cloud PowerVS · Specialized by business function · Dispatched by the orchestrator
Act ↓
MCP interface layer
IBM i MCP server · Any agent · Any framework · 500 tools in 2026
IBM i sovereign core
RPG · DB2 for i · Mapepire · SQL services · CL commands · APIs
Object-level security · Journal audit trails · MI architecture
Observe ↑
Foundational
T H
AI Models & Interfaces
The intelligence your agents and your people interact with directly. Claude, ChatGPT, watsonx, IBM Granite. Prompt engineering. Model selection for domain-specific work.
Which model do I trust for my domain? What does prompt engineering mean when I'm working with 30 years of business logic?
Foundational
T H O
Agentic Coding
Moved from emerging to foundational in April 2026. IBM Bob's GA at PowerUp is the signal for IBM i shops. GitHub Copilot and Cursor for general use. IBM Bob when the context is RPG.
Does IBM Bob change what I write or does it write for me? What happens to my RPG expertise when the agent can generate it?
03
Emerging
T O
Agentic Frameworks
MCP — the protocol that allows an agent to reach into a system, retrieve data, and take action. LangGraph for multi-step orchestration. A2A for agent-to-agent communication.
I don't need to build these. But I need to know what they are so I can evaluate vendors who will use them in my environment.
04
Foundational
T O
Infrastructure
PowerVS as the rational capital allocation for IBM i shops evaluating cloud. GPU access vs. GPU ownership. The on-premise vs. cloud question reframed as a sovereignty question.
Where does my business logic live when agents operate against it at scale? Who controls it — and what's the access model?
05
Foundational
T O
Integration & Data
30 years of business logic in Db2 and RPG. Mapepire as the access layer. Natural language interfaces to your data. The integration architecture is what makes or breaks the agentic transition for IBM i shops.
What does it take for an agent to reach my IBM i data? Where are the seams? This is a knowledge inventory question before it's a technology question.
06
Critical Gap
O T H
Governance & Security
The domain that kills deployments that succeeded everywhere else. Agent identity, audit trails, human-in-loop design, SOX exposure when an agent touches financial logic. EU AI Act enforces August 2, 2026.
Who is responsible when an agent makes a decision that affects a customer? How do I design oversight into a system that's supposed to run without me?
1
Integration & Data
Do First
This is where your expertise is structural and where the transition will succeed or fail. Before any vendor conversation — understand your own access architecture. What does it take for an agent to reach your IBM i data? The practitioner who has done this work controls every subsequent decision.
2
AI Models & Interfaces
Do in Parallel
You don't need to choose between Granite and Claude this week. You need to understand what the distinction means at a working level. Use both. Your domain knowledge makes you a better model user than a cloud developer who doesn't know what a service program is.
3
Agentic Coding
Near-Term Priority
IBM Bob is GA. The window for early adoption advantage is open. You don't need to adopt it tomorrow — but you need to be in a deliberate evaluation posture rather than a waiting one. The shop that understands what Bob can and can't do with RPG by end of 2026 is in a fundamentally different position.
4
Agentic Frameworks
Watch
MCP and LangGraph are infrastructure-layer decisions for most IBM i shops. What you need now is orientation, not implementation. Understand what MCP does. Understand why it matters that agents use it to connect to your environment. The spotlight pieces go deeper.
5
Infrastructure
Evaluate Now · Commit When Clear
Infrastructure decisions carry 3-6 month procurement and security review cycles. Begin PowerVS evaluation and POC work in parallel with step 1 — don't wait. Defer the capital commitment until use case is clear. The distinction matters: the shops that started evaluating early are not the ones still deciding.
6
Governance & Security
Concurrent — Not Optional
For regulated environments — banking, insurance, healthcare, government — this is not step 6. It runs in parallel with steps 1-3 and precedes any agent touching production data. SOX exposure, HIPAA, GDPR, and the EU AI Act August 2026 deadline are not scale concerns. They are deployment prerequisites. IBM i shops already hold the governance primitives. The work is extending them to cover agent identity and agent actions before the first production deployment, not after.
94%
of GenAI pilots are failing
1%
are at full deployment
3–5×
The productivity gap for those who cross

The 1% at full deployment aren't seeing incremental gains — they're seeing 3–5× productivity. McKinsey tested 25 variables across 2,000 organizations. The single highest-ranked factor was workflow redesign, not tool selection. The chasm is not a technology problem. It is a readiness problem. The organizations that cross it first compound their advantage permanently.

77%
have no committed agentic AI strategy today
ManpowerGroup, 2026
26%
scaled AI experiments to production
BCG, 2024
88%
use AI in at least one function — agent deployment in single digits
HAI 2026 AI Index
The IBM i Read

Six domains. One platform's answer.

Every domain in this map intersects with IBM i specifically. This is the layer underneath — what each domain means for the practitioner who has built their career on the platform.

01 · AI Models & Interfaces
Which model do you trust — and have you made that choice deliberately?

Watsonx and Granite give IBM i shops a path that stays inside the IBM relationship, which matters for shops where procurement, security review, and vendor management all live inside that relationship already. Claude and ChatGPT are what your developers are actually using, whether or not they've told you. Shadow AI is already in your environment. The real question isn't which model to standardize on — it's whether you've made the choice deliberately or by default.

02 · Agentic Coding

That distinction matters when the codebase is RPG, not Python. The practitioner who learns to work with Bob now — understanding where it helps, where it misreads the platform, where human judgment still governs — has an advantage that compounds. The shop that waits discovers the gap when it's larger. Steve Will has been direct: the platform moves to a fully agentic development model within the next few years. The question is whether that happens to you or with you.

03 · Agentic Frameworks
MCP is how an agent reaches into your IBM i system.

Before you evaluate any agentic vendor, ask: what protocol does your agent use to connect to IBM i? If they don't know what MCP is, the conversation ends there. The IBM i MCP server is a published, IBM-maintained repository — the access layer is being built. LangGraph orchestration is what makes multi-step agent workflows coherent rather than brittle. Most IBM i shops won't build these layers — but you need the vocabulary to evaluate who is building them on your behalf.

04 · Infrastructure
The on-premise vs. cloud question is a sovereignty question, not a cost question.

PowerVS keeps the IBM i workload inside IBM's infrastructure while opening the agentic surface. Where does your business logic live, who controls it, and what's the access model when agents operate against it at scale? That's the infrastructure question the agentic economy forces. The capex reckoning is real — Morgan Stanley revised 2026 hyperscaler investment from $650B to $805B in four weeks. The shops that transferred hardware risk to IBM are better positioned than the ones still deciding.

05 · Integration & Data
The integration architecture is the competitive moat you're already sitting on.

Every piece of business logic embedded in your RPG programs, every Db2 table encoding decades of operational decisions — the agent that can access that intelligence is more capable than one that can't. Mapepire is part of the access layer. Natural language interfaces to Db2 are part of it. Understanding the access architecture your agents need before you build it is work only you can do — because only you know what the data actually means.

06 · Governance & Security
IBM i environments already have the foundation. These are extension questions, not starting-from-zero questions.

Object-level security, journal-based audit trails, profile management — the governance primitives are there. What the agentic layer requires is extending that logic to cover agent identity and agent actions specifically. Who is the agent acting as? What can it access? Where does the audit trail live when an agent makes a decision that affects a customer? The IBM i practitioner who already understands audit trails and access controls has a shorter path to agent-ready governance than a cloud-native shop building those muscles for the first time.

Spotlight Series — Coming Next
Each domain gets its own deep dive
Coming
Making IBM i agent-ready — the integration architecture
Coming
MCP — the protocol nobody in IBM i is talking about yet
Coming
Governing agents in a compliance-heavy environment
Signal4i Intelligence Brief

Signal-driven research for IBM i operators navigating the AI transition. Free.

— or — Read on Substack ↗