A Signal4i orientation to the technology landscape every IBM i practitioner needs to navigate — without becoming an expert in all of it.
The terms are arriving faster than anyone can absorb them. MCP. LangGraph. A2A. IBM Bob. Granite. PowerVS. Telum. Spyre. watsonx. Claude. Cursor. GitHub Copilot. EU AI Act. The 49-point gap. Agent supervisors. Atomic units of work. Coasean Singularity.
If you've spent the last six months feeling like you're behind, you're not. You're experiencing something specific: the absence of a hierarchy for what matters.
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 through a landscape that is genuinely, measurably changing.
The Q1 2026 data is instructive: Gartner finds 80% of enterprise applications now ship with at least one embedded AI agent. S&P Global Market Intelligence puts the share of organizations with an agent actually running in production at 31%. A 49-point spread between what's being built and what's being operated. That gap isn't a technology problem. It's an orientation problem. You can't sequence a transition you can't see.
The practitioner who has spent 30 years connecting business logic to technology has exactly the pattern recognition this moment demands. What you need is not more information. You need to know what the domains are, how they connect to your world specifically, and in what order they deserve your attention.
That's what this map is for.
The agentic technology landscape breaks into six domains. Some are foundational — you need them as working knowledge regardless of where your organization is in the transition. Some are emerging — the direction is set but the path is still forming. One is a critical gap that most shops are underestimating until something goes wrong.
This is the entry layer — the intelligence your agents and your people interact with directly. Claude, ChatGPT, watsonx, IBM's Granite foundation models. Prompt engineering. Model selection for domain-specific work.
The foundational fluency here isn't mastery of any single model. It's understanding what differentiates them at the level that matters for your decisions: which one do you trust for regulated data, which one integrates cleanly with your existing IBM relationship, which one your development team is already using informally. That last one matters more than most organizations want to admit. Shadow AI is already in your environment. The question is whether it's visible.
Morgan Stanley's capex projections tell part of the story — $805 billion for hyperscaler AI infrastructure in 2026, rising to $1.1 trillion in 2027. The model layer is not a passing experiment. It's infrastructure that will outlast your current architecture decisions.
This domain moved from emerging to foundational in April 2026. IBM Bob's general availability at PowerUp — announced at COMMON's annual conference in New Orleans — is the signal for IBM i shops specifically. The RPG developer who begins working with an agentic coding assistant now has a meaningful head start on every shop that waits for the technology to prove itself on someone else's codebase.
The question isn't whether agentic coding will affect IBM i development. Steve Will, IBM i CTO, has been direct about the trajectory: the platform moves to a fully agentic development model within the next few years. The question is whether you understand what that means for your team's expertise before it happens to you rather than with you.
The architecture matters here. IBM Bob is not a standalone tool — it runs inside VS Code through IBM Code for i. Adopting Bob means moving to VS Code as your primary development surface. For shops still on RDi, that's a real transition decision, not just a tooling swap. Bob covers RPG, COBOL, CL, SQL, and Python. For shops with workloads beyond the platform, a multi-tool posture is normal: Bob for IBM i work, Copilot or Cursor for everything else.
This is where the connective tissue of the agentic layer lives. MCP — Model Context Protocol — is the protocol that allows an agent to reach into a system, retrieve data, and take action. LangChain and LangGraph are orchestration frameworks that coordinate agents across multi-step reasoning tasks. A2A protocols govern agent-to-agent communication as multi-agent systems become more common.
Most IBM i practitioners haven't encountered these terms yet. That's appropriate — you don't need to build them. What you need is the orientation to evaluate vendors and partners who will use them in your environment. When a vendor tells you their agent connects to your IBM i system, MCP is how that connection works. Knowing the vocabulary is enough. Understanding the architecture is better.
The Deloitte Tech Trends 2026 read is useful here: agent supervisors are emerging as the strategic handoff architecture — the human role that governs what agents do and how they escalate. The framework layer is where that governance gets implemented technically.
You've been making infrastructure decisions for decades. This domain is not fundamentally different — it's the same CapEx versus OpEx calculus applied to the agentic layer. What's new is the specific question the agentic economy introduces: do I own the compute my agents need, or do I access it?
Most IBM i shops are on-premises and the agentic transition starts there. Power11 with Spyre is IBM's hardware answer — an on-board AI accelerator that removes the dependency on external GPU for inference workloads. The infrastructure for agentic operation on-prem is more accessible than it's often presented.
Cloud: PowerVS is a maturing option that keeps IBM i workloads inside IBM's infrastructure. It is not the only path — IBM-certified cloud partners and managed service providers offer alternatives. The cloud decision follows use case clarity; it doesn't precede it.
This is the domain that separates IBM i shops from every other organization navigating the agentic transition. Thirty years of business logic embedded in Db2, in RPG programs that have run without interruption since before most current developers entered the workforce, in service programs and stored procedures that encode decisions your organization has long since forgotten it made. No outside architecture firm understands this. No AI vendor has solved for it. The IBM i practitioner is the only person in the room who can read it.
The Coasean Singularity — the economic threshold at which transacting through an agent becomes cheaper than through a human employee — is no longer theoretical. When agents reach into an IBM i environment, the integration architecture is what makes the Coasean math work or fails to. APIs, MCP connections, Db2 access patterns, the service program layer — this is the integration challenge nobody from the AI world understands. It is the only domain where platform expertise is a structural competitive advantage.
The HAI 2026 AI Index documented 362 AI incidents in 2025 — up 55% year over year. The 88% agent pilot failure rate is not a technology failure rate. It is a governance failure rate. The technology worked. The governance architecture didn't.
IBM i environments already have the foundation. 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? These are extension questions, not starting-from-zero questions. That's an advantage most organizations in this transition don't have.
The EU AI Act becomes enforceable in August 2026. For any IBM i shop with European operations, this is not abstract. Agent identity, audit trails, human-in-loop design, access controls, SOX exposure when an agent touches financial logic — these are the questions that move governance from compliance concept to operational requirement.
The six domains exist in every organization navigating the agentic transition. What IBM i practitioners need is the layer underneath — how each domain intersects with the platform specifically.
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. The real question isn't which model to standardize on — it's whether you've made the choice deliberately or by default.
IBM Bob runs inside VS Code through IBM Code for i. It covers RPG, COBOL, CL, SQL, and Python — Python support exists but its practical depth in IBM i contexts is still developing. Many shops will run Bob for IBM i work alongside Copilot or Cursor for everything else. The multi-tool posture is normal, not a compromise.
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. 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's building them on your behalf.
Most IBM i shops are on-prem and the transition starts there. Power11 with Spyre is the on-prem hardware story — AI inference at the system level. For shops evaluating cloud, PowerVS is a maturing option; IBM-certified partners offer alternatives. The cloud decision isn't primarily about cost — it's about sovereignty. Where does your business logic live, who controls it, and what's the access model when agents operate against it at scale?
This is where the IBM i practitioner's knowledge floor becomes the competitive differentiator. 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. The integration architecture is what makes or breaks the agentic transition for IBM i shops. Mapepire is part of that architecture. Natural language interfaces to Db2 are part of that architecture. Understanding the access layer your agents need before you build it is work only you can do.
IBM i environments already have the foundation. 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. These are extension questions, not starting-from-zero questions. That's an advantage most organizations in this transition don't have.
The practitioner doesn't have time to master all six domains simultaneously. The question isn't whether all of them matter — they do — but what order to move through them given where your organization actually is.
This is the domain where your expertise is structural and where the transition will succeed or fail. Before you evaluate any agent, any framework, any vendor — understand your own access architecture. What does it take for an agent to reach your IBM i data? What business logic lives in programs versus in Db2? Where are the seams? This isn't a technology question first. It's a knowledge inventory question. The practitioner who has done this work controls every subsequent decision.
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. The tacit knowledge that makes an IBM i practitioner's prompts better than a cloud developer's prompts — the ability to describe business logic in the terms the platform actually uses — is the differentiator. California Management Review named it directly in March 2026: in a world where data is abundant and models are open, tacit knowledge is the next competitive moat. Build that fluency now.
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 the end of 2026 is in a fundamentally different position than the one that evaluates it in 2027.
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. For now, the vocabulary is the priority.
PowerVS is the right call for most shops evaluating cloud — but the timing of that move should follow use case clarity, not precede it. Don't build the infrastructure before you know what the agent needs to do. Understand your on-prem capability first, identify the gap, then make the capital decision.
Governance feels like overhead until it's the difference between a production deployment and a compliance incident. The EU AI Act deadline in August 2026 creates a forcing function for shops with European exposure. For everyone else, treat 362 annual incidents as the forecast. The organizations that built the governance layer before scaling are not in that dataset.
There is a version of the agentic transition that happens to IBM i practitioners. Vendors arrive with frameworks your organization didn't choose, for problems it didn't define, on timelines it didn't set. The technology moves, the organization follows, and the practitioner who built their career on deep platform knowledge watches from the side while the decisions get made by people who don't know what a service program is.
There is another version. The practitioner who reads this map and sequences their attention deliberately — who does the integration inventory before the vendor arrives, who builds model fluency before the RFP lands, who asks what protocol does your agent use to connect to IBM i before signing the contract — that practitioner is not reacting to the transition. They're navigating it.
Thirty years of connecting business logic to technology has given you the pattern recognition this moment requires. The 49-point gap between AI adoption and production operation exists because most organizations have neither the domain depth nor the platform foundation. You have both.
The map is visible now. The domains are named, the connections are drawn, the sequencing framework is here.
Finding your signal starts here.