The properties the rest of the industry is now bolting onto its systems — identity on every object, a tamper-evident record of every change, integrity enforced before a write lands, and business rules proven at scale — IBM i has enforced in the operating system for decades. What changed in 2026 is not the platform. It's that agents can finally reach it.
For years, AI on IBM i was something coming later — promised, discussed, always ahead. That changed in 2026. IBM put a production, IBM-branded agentic development partner into the hands of the RPG installed base — and then, in June, extended it to reach into the platform natively. The waiting is over. The question is no longer whether the platform becomes agent-operable. It's who moves first while the knowledge is still in the room.
Spring 2026 — IBM Bob reaches general availability. An AI-first development partner that works across the full lifecycle — plan, code, test, modernize — with human-in-the-loop approval gates built in, not retrofitted. It reads RPG across its generations — fixed-format through free-format ILE — alongside CL, SQL, DDS, COBOL, Java, and Python, and routes work across a mix of models, including Anthropic's Claude and IBM's own Granite family.
Then, on June 24, 2026, IBM went further with the Bob Premium Package for i — adding native IBM i connectivity through Code for IBM i, so Bob can read and write source, compile programs, and run SQL directly against the system. The tool no longer describes your platform from the outside. It operates inside it, under human approval gates, with a traceable record of every action.
"It has a lot of knowledge. It knows things about IBM i."
On what separates Bob from a generic assistant: it knows things about IBM i — and, in Steve Will's words, how IBM i can be attacked and what you ought to do when writing software for it. IBM calls the IBM i-specific edition "the IBM i special sauce." — Steve Will · IBM i Chief Architect
The trajectory IBM's own architect has been direct about: the platform moves toward a fully agentic development model over the next few years. The shops that begin working with an agentic partner now hold a head start on every shop still waiting for the technology to prove itself on someone else's codebase.
This is the claim beneath the whole argument, so it's worth making concrete. The identity, auditability, integrity, and containment that agent-safety now demands were architectural decisions on IBM i, made below the application, decades ago — not an add-on layer. Read the ledger.
The agentic era's hardest requirements — identity, audit, integrity, and a coherent state to reason over — are not features IBM i is missing. They are the reasons it was built the way it was.
The layer of the agentic stack that holds value is orchestration — where your data, processes, and business rules govern what agents actually do. It can't be bought. It has to be encoded — and on IBM i, it already has been. The five layers, in full →
This is the commercial face of the principle the practice is built on: own your source, own your intelligence. Everything specifiable commoditizes — the model, the tooling, the generic workflow. What doesn't is the source. An organization that keeps its source legible and owned keeps the one thing commoditization can't take.
The orchestration layer isn't absent from IBM i environments. It's been running in production for decades. It just wasn't connected to the agent.
Stated plainly: the advantage is a position you hold, not a product you buy. — Signal4i · The Agentic Stack
The gap was never that the intelligence wasn't there. It was that the platform had no presence an agent could navigate — no way to surface the logic, no way to call it. In 2026, two tools resolve exactly those two problems. One makes the source legible. The other makes it callable.
Bob reads a long-lived RPG or COBOL application and surfaces what's inside it — explains it, documents it, diagrams the business logic — so the rules encoded across decades stop being trapped in code or dependent on a shrinking pool of senior experts.
With the Premium Package for i, Bob reaches into the system natively: read and write source, compile, run SQL — every action under human approval gates and traceable start to finish. Comprehension, delivered by the platform vendor.
Mapepire is an open-source database client for Db2 for i — a lightweight, secure connection over a single port that lets modern tooling reach the database, and the SQL and stored procedures around it, without the old ODBC and JDBC baggage. It's the path from outside the platform to the data and logic inside it.
That's the reachability step the argument turns on: the Db2 access and the SQL-callable logic already on the system can be reached over a clean interface — the foundation an orchestration layer builds on to expose them, under governance, as tools an agent can invoke, without rebuilding the sovereign core.
Comprehension and access are no longer things an IBM i organization has to invent. The vendor ships one; the ecosystem ships the other. What remains is the discipline of sequencing them correctly, under governance, with the sovereign core untouched. That's the work Pegasus4i does.
RPG and Db2 for i stay exactly where they are — the encoded business logic, the object security, the journaled record, the rules proven at scale. The agentic layers reach in on your terms and act under governance. The system your business already trusts doesn't get replaced to meet the agentic era. It gets connected to it — sovereign, and multiplied.
Position in the agentic era is earned at the start, while the knowledge is still in the room. See how the five layers build in order — and where your platform already sits in them.
01 / 03 · The technology axis · next: 02 The agentic stack