Agent governance from a CTO · free with your pilot
Build on what you own. One lens for the IBM i you know and the agents you don't.
The agent asks. IBM i decides. OneLens4i is agent governance for IBM i shops putting AI agents to work: seven moves to secure, govern and see your agents and your IBM i as one system, using controls IBM i already has. The OneLens4i white paper is free to download. A paid 90-day pilot applies it to one workflow, and the Governance view software comes with it at no additional cost.
Free in your workspace, with your email and no password. The technical companion, for architects and security, is coming soon.
Free to download
Read it
The 13-page white paper, free to download in your workspace, with the technical companion coming soon. The Governance view is there too, simulated, with a working model of what a client would see.
Free with your pilot
Run it
Run a pilot, and OneLens4i comes with it at no additional cost, including the Governance view: the software that shows your agents and your IBM i in one view. Built to run in your own cloud, with no license fee.
The Governance view, executive role · Beta · concept · simulated data · Silver Lake Financial Group is fictitious · See it running on our own agents ↓
Vendor-neutral · Updated November 2026 · Published by Pegasus4i, an IBM Business Partner
Why now · the IBM i community's own numbers
The tools are already in. The governance isn't.
In the 2026 COMMON × Fortra survey of the IBM i community, nearly everyone already uses AI, but almost no one governs it. The community has the same gap as everyone else, and a short window to close it first.
93%use at least one AI tool today
24%already run AI in production
44%have no security or governance framework
7%have a comprehensive, active one
The community saidSecurity is the #1 barrier (24%), and security and governance is the lowest-scoring of nine readiness areas
OneLens4i's answerSecure. Every agent is a user profile, and the baseline is assessed and fixed before production use.
The community said44% have no governance framework
OneLens4i's answerGoverned. The four gates, on controls IBM i already has.
The community saidUnclear ROI is the #2 barrier (23%)
OneLens4i's answerMeasured. Every agent has a business case, measured on the decision.
COMMON × Fortra, The State of AI in the IBM i Community, 2026 Community Survey, 147 responses. Full results at signal4i.ai/common-ai-results. Outside IBM i it looks the same: 23% of organizations are scaling an AI agent anywhere in the business (McKinsey, The State of AI in 2025, 1,993 respondents, Nov 2025).
The seven moves · for everyone at the table
Seven moves. Each one tied to how you decide, spend and earn.
CEOOne objective, an owner for every agent, and a decision at day 90 on the evidence.
CFOA measured baseline, targets signed up front, and the exit cost of every tool priced.
CTOIBM i stays the system of record; gates first, autonomy second.
OperatorYour know-how becomes the agent's playbook, and you review the exceptions.
Move 1 of 7 · Build on what you have
The agent asks. IBM i decides.
In one lineYour IBM i already has the controls agents need. Use them.
Your IBM i already holds your source, your data and the way your company does business, and it already has the controls agents need. Don't copy it to a new platform or rewrite working code. Agents call your programs; they never absorb them.
Follow one requestAn agent asks to raise a customer's credit limit.
1
Who is asking?Its own user profileThe agent signs on as itself, never as a person or a shared profile, with no special authorities.
2
May it touch this?Object authorityIts profile is authorized to the customer file and the credit program, and nothing else.
3
Is this way in allowed?Exit programThe exit program on the interface it uses checks the request against an approved list.
4
What do the rules say?Your RPG programThe agent calls your existing credit program. Your business rules decide, not the agent.
5
Can it be undone?Commitment controlThe change runs as one transaction, so it commits completely or rolls back.
6
Is it on the record?The journalThe before and after are journaled, tied to the agent's profile and the request.
IBM i decides. The limit changes within policy, on the record, and can be rolled back. Over policy, it goes to a person.
An illustration of the controls OneLens4i asks you to use, on any IBM i. Not a recording of a live system.See all seven: what agents need, and what IBM i already has
What agents needWhat IBM i already has
An identity for every agentUser profiles, with special authorities you control
Narrow doors into the dataStored procedures, object authority, adopted authority and exit points
Rules an agent can't talk its way aroundDb2 for i constraints and triggers
A way to undoJournaling, and commitment control for transactions
A record of who did whatThe audit journal (QAUDJRN)
Separation of dutiesObject ownership and authority
VisibilityIBM i SQL services
Object-level security answers who is this user. The agentic question is one layer up: who is this agent, acting for whom, and where does it have to stop?
In one lineKeep what's yours on your platform, and treat every tool as a part you can swap.
Models and agent tools turn over in months. Keep what holds your business (your logic, your data, your rules and your record) on your own platform, and treat the tools as parts you can swap. Every tool answers one question: what would it cost to change it next year?
Every agent, whether you build, adopt or buy it, passes the same four gates before it reaches IBM i.
The portability test
Anything you adopt or buy has to pass all four. A part you can replace is a choice; one you can't is a dependency.
OPEN
Its source or its specification is public.
STANDARD
It speaks SQL, MCP, REST or another open standard, not a private format.
YOURS STAYS YOURS
Your data, business rules and logic stay in Db2 and your own source, not inside the tool.
REPLACEABLE
You know how you would remove it, and roughly what that would cost.
Take the testThink of one tool you're weighing. Answer the four questions.
OpenIs its source or its specification public?
StandardDoes it speak SQL, MCP, REST or another open standard?
Yours stays yoursDo your data, rules and logic stay in your Db2 and your own source?
ReplaceableDo you know how you would remove it, and roughly what it would cost?
Your answerAnswer all four.
If a model at half the price passed your tests tomorrow, how long would it take to switch? The answer should be days.
No agent gets more autonomy than its gates can prove.
In one lineAutonomy is earned a level at a time, and every level needs evidence.
Every agent has a named owner and an autonomy level, and four gates ask for more evidence as the level rises. A gate that fails at an agent's level stops the work, whether the agent was built, adopted or bought. Each gate uses controls IBM i already has.
GATE 1
Data stays inside agreed boundaries
Each interface an agent uses has its permitted destinations, and the controls that enforce them.
Exit programs and access controls on the interfaces agents use · an approved list of destinations
GATE 2
Named accountability
Every action belongs to a scoped identity, never a shared master key.
Object security · adopted authority · no all-powerful profiles
GATE 3
Rollback before autonomy
Changes inside a commitment-controlled transaction can roll back. Committed work and actions outside IBM i need a tested recovery plan first.
Journaling on the files you choose · commitment control for transactions · compensating entries for decisions
GATE 4
Decisions trace to a person
Every decision follows an approved policy or a person's approval, with a named accountable owner.
User profiles · the audit journal, reviewed
Four autonomy levels
Each level adds one thing the agent may do, and asks more of the gates.
1 · Read only
Reads defined data. Its output goes only to the person who asked.
G1G2G3G4
Gates 1 and 2: a read-only profile scoped to named files, exit points on every way in, and every request logged
2 · Recommend
Drafts and recommends. A person reviews the output and takes the action.
G1G2G3G4
Plus Gate 4: each recommendation logged with the person who acted on it, and quality tested against known answers
3 · Act with sign-off
Writes data or sends messages, after a named person approves each action.
G1G2G3G4
Plus Gate 3: journaling with before images on every file it can change, and each approval recorded in the audit trail
4 · Act within limits
Acts within set limits. People review exceptions and outcomes, not each action.
G1G2G3G4
All four in full: tested rollback, a circuit breaker that stops the agent and disables its profile when a limit is crossed, and a named owner for its behavior
Tap a level to see the gates it needs.
The work stops. Nothing changes on IBM i, anything in flight rolls back, and the request goes to the agent's named owner. The agent doesn't move up a level until the gate passes.
People decide. Agents do the work. IBM i enforces it.
Agents run beside your IBM i, not on it. Secure both worlds.
In one lineEvery agent gets its own identity, its own limits and one way in.
Agents run on Linux: in a partition on the same Power server, on a host next to it, or in a cloud. They reach IBM i only through one governed doorway. That gives you two worlds to secure: the IBM i you know, and an agent runtime you may not.
AgentsPlan and askRun beside IBM i, on Linux or in the cloud
→
Governed gatewayChecks the gatesOne doorway in; nothing goes around it
→
IBM iPrograms and Db2The system of record decides and commits
IdentityEach agent acts as its own IBM i profile, end to end.
ApprovalAn approved policy or a person commits each decision, with a named owner.
EvidenceRequirement: one trace ID from the request to the IBM i journal.
Watch the door
Agent containerAgentGoverned doorwayDoorwayIBM i
Own profile · short-lived keyOwn profile
✓ Through the door, within its limits, on the record.
Shared profile · stored passwordShared login
✕ Turned back at the door: no shared profiles, no keys in files.
An illustration of the rules in Move 4. Not a recording of a live system.
ITS OWN PROFILE
To IBM i, every agent is a user profile: its own, never shared, with no special authorities such as *ALLOBJ.
ITS OWN CONTAINER
Limits on CPU, memory, time and network, so a misbehaving agent's damage stops there. Every version can be restored exactly.
NO KEYS IN FILES
Short-lived credentials from a secrets store, scoped to one agent, and one path to models with cost limits per agent.
BASELINE FIRST
Default passwords, *PUBLIC access and open servers are assessed and fixed before any agent goes into production.
Architect or on the security team? The technical companion goes deeper: the IBM i controls, the gateway and the evidence, as they run on our own IBM i. It's coming soon, free in the same workspace as the white paper.
One view of both worlds, from the request to the record.
The Governance view is the software that comes with a pilot. It shows every agent, its level and status, what each run did and cost, and an evidence record chained entry to entry. One view serves everyone: decisions and value for executives, behavior for IT, evidence for audit. It's running today on our own agents and on our own IBM i, where it reads agent activity read-only and traces a call from the agent to the IBM i job and the Db2 record. No client runs it yet. None of our agents acts on its own without a stop that has been tested and recorded.
Running todayBetaLive Governance view · Pegasus's own agents and IBM i, Nov 8 to 13 · Silver Lake runs are test traffic on a fictitious lender
One request, end to endFrom the agent to the IBM i job and the Db2 record, on our own IBM i.The overviewEvery agent, its level and status, and whether the record holds.RunsCounts and cost per run. Never what was said.The evidence recordEvery change chained to the one before; checked on every load.DriftDecisions watched against their baseline, counts only. Nov 8: rated stable on test traffic. The first alert fired Nov 12, after a deliberate change to the offers.
The role viewsBeta · conceptSimulated data · Silver Lake demo tenant, fictitious
Decisions, people and value across every workflow, and what needs a decision this week. Full view ↗
Every agent call joined to the IBM i job that ran it, with the incident and its action at the top. Full view ↗
Owners, profiles, stops at the crossing and open findings, ready before the examiner asks. Full view ↗
Walk through it in your free workspace → Follow one number to the records, one decision's trace and the action. Beta concept views of the Governance view, on simulated data. The IBM i side runs today on our own IBM i; no client runs it yet. Silver Lake Financial Group and Alder & Pine are fictitious companies.
Watch it catch a problem
Six scenarios, from an out-of-scope tool call to data trying to leave. Pick one, or let the tour run.
Governance viewBeta · conceptSample data · Silver Lake demo tenant
Scenario: Normal day. The animated floor shows on wider screens; the lens below follows each scenario.
Normal day
All clearAll layers
All layers are healthy.
AgentsOK–
MCP & toolsOK–
SecurityOK–
PeopleOK–
IBM iOK–
Follow one request–
One view of both worlds: agent identity at the edge, every MCP tool call and its scope, keys that never leave the vault, data stopped at the egress guard, decisions that wait for a named person, and IBM i load beside the agent traffic that drives it. One correlation id runs from the edge to the record in Db2 for i. Beta concept on simulated data. On our own IBM i, the trace from the agent to the IBM i job and the Db2 record runs today, on Silver Lake test traffic. No client runs it yet.
Agree the targets before you spend, and measure them the same way at the end. This is the shape of the calculation for one workflow; your figures come from your own baseline.
Payback in monthsF = 12 × D ÷ (C − E)One-time cost, divided by what the agent saves each year after oversight. Your figures, from your own baseline.
ACost per decision todayStaff time at a loaded rate, plus errors and rework
−
BCost per decision with the agentRun cost, plus human review and exceptions
×
Decisions per yearFrom your own baseline
=
CAnnual benefitWhat the workflow saves each year
DImplementation costOne-time cost to build and launch
÷(
CAnnual benefitFrom the line above
−
EAnnual oversight costReview, monitoring and upkeep per year
)× 12 =
FPayback in monthsThe number you sign before the pilot
Count the people and the mistakes. Review time, exceptions and the agent's own errors all belong in B.
Quality must hold. Read overrides, errors and business outcomes together; fewer overrides only count if quality holds.
One workflow is one workflow. Don't multiply a pilot result across the business; prove the next one.
We don't publish typical savings. Your numbers come from your baseline, and the assessment helps you measure it. Targets are signed before the parallel run starts, never after. Run your own numbers in Agent ROI → It uses the same worksheet, works only from your own inputs, and is free in your workspace.
Ready to put numbers on your first workflow? Start with the white paper. The Show the math worksheet is in the same workspace.
I spent a decade inside a regulated IBM i company and served as CTO of a billion-dollar consumer finance business. In a company like that, governance isn't a side job. It is the job. Over forty years on this platform, I've watched tools carry IBM i shops forward and then keep the business inside someone else's format. Agents now hold how a business decides. There's no license fee, because governance you can't take with you isn't governance.
Reggie Britt · Founder and CTO, Pegasus4i · Forty years on IBM i
Where to start
Three steps, at your pace.
1 · FreeDownload the white paperThe white paper is free in your workspace, with the Governance view beside it and the technical companion coming soon. The 3-minute Read, eight questions you answer yourself, gives a first look at where you stand.
2 · Paid assessmentCheck your IBM i properlyWe check your IBM i against the four gates and run a security baseline scan, so you know what the Read could only estimate. Scoped with you.
3 · PilotRun one workflow for 90 daysYou sign the measures first. OneLens4i comes with it at no additional cost, Governance view included. Ongoing oversight and support are optional.
One partition, one decision, one agent. Run the Read and the baseline scan, and fix what it finds. Pick one decision and wrap its existing RPG in one stored procedure. Give one agent its own profile, let it call only that procedure from one small Linux host or partition, and watch it with the SQL services and audit journal you already have. Add more agents, automation or products only after the first one has earned it.
The full technical path, step by step, is in OneLens4i in your workspace. Get it free →
Free to download · free with your pilot
Download it free. Run a pilot, and OneLens4i comes with it.
The white paper is free to download in your workspace, with the technical companion coming soon, and you can use them without us. Run a pilot with us, and OneLens4i comes with it at no additional cost, Governance view included: built to run in your own cloud, your data stays yours, and there's no license fee. Ongoing oversight and support are optional; pricing depends on the size of the engagement, scoped with you. Anything we build for you is yours, and if we ever recommend a related platform, we tell you, and it passes the same gates and the same portability test.