Welcome to another edition of Talsco Weekly

In This Issue

• IBM i Briefs: 📝 The IBM i survey that only works if you take it. 🗺️ Bob maps what breaks. You still own the why.
• AI: 📈 The AI coding chart IBM Bob isn't on (yet).
• Career: 🔀 The RPG skill that isn't RPG.
• Database: 🗄️ The insurance policy that can pay for IBM i AI.
• Hiring: 📊 The IBM i worry that just passed cybersecurity.
• Leadership: 🔁 "AI-native" is the new buzzword. Here's the test.
• Learning: 🧪 Who tests the code AI wrote?
• Modernization: 🔌 AI can find the rule. It can't tell you why. Yet.
• Security: 🔒 What can your IBM i service account actually reach?
• Trends: 🪃 AI is taking the same trip offshoring took.
• Did You Know: 🔤 Apache Pulsar ran on the AS/400 first. Sort of.

IBM i Briefs

📝 The IBM i survey that only works if you take it

The 2027 IBM i Marketplace Survey closes soon.

The survey opened September 15, its 13th year running.

It matters more than most. The big analyst firms stopped tracking Power Systems in any detail years ago, and IBM shares little about its own customer base.

That leaves this survey as one of the few public reads on where IBM i shops actually stand. IT Jungle also uses the raw responses to model the installed base.

The catch: shops on current hardware answer most. Shops on older releases often don't, so the picture skews newer than reality.

Why your answers count: If your company has never taken part in this survey, this is the year to start.

That goes double if IBM i is a small slice of what you run. Those shops are easy to miss, and they belong in the picture.

One AS/400 running the warehouse still counts.

Your answers shape where IBM and the vendors put their effort. They matter to the community more than you might think.

It takes about 20 minutes. Take the survey.

🗺️ Bob maps what breaks. You still own the why.

IBM ships Understand for Premium Package for i.

IBM's Bob Premium Package for i turned three months old this week, and IBM shipped a feature called Understand.

It scans your application libraries once and maps how every object connects: program calls, service program bindings, triggers, foreign keys.

The map lives in a Db2 schema on your system, so Bob answers from facts instead of re-reading source every time.

Ask what breaks if you change a file, and it traces callers and callees both ways.

It also sorts programs by modernization effort and flags code that hasn't run in over a year.

Before you touch CUSTMAST: This is the blast-radius question every IBM i developer asks, now answered in seconds.

What it maps is the what. It won't tell you why a dependency exists.

As our Modernization piece below argues, that part still lives with your people.

AI

📈 The AI coding chart IBM Bob isn't on (yet)

Where would IBM i's own agent land?

Developers are switching AI coding agents faster than most shops can approve one.

A global survey of more than 15,000 professional developers, run May through July and published in August, found 90% use an AI agent at work weekly and 68% daily.

Claude Code leads at about 39%, more than double its January number.

Codex climbed from 3% to 16%.

GitHub Copilot, which started the category in 2023, fell to 21%. Cursor dropped from 18% to 12%.

Leadership in this category is rented, not owned.

Our Take: Bob doesn't appear on the chart. IBM made it generally available April 28, just weeks before the survey opened. The Premium Package for i didn't ship until June 24, in the survey's final stretch. And IBM i developers are a rounding error in any global sample.

If it were plotted, our bet is it lands near the bottom, under the 6% tier. That's fine. Bob isn't competing for the general developer.

It's competing for the RPG developer, where the competition is often nothing at all.

A lot has shipped since, including the Understand feature in Briefs above.

The more telling number: VS Code just passed RDi in the 2026 IBM i Marketplace Survey. Bob is a fork of VS Code. The doorway is already open.

Career

🔀 The RPG skill that isn't RPG

Git is moving from nice-to-have to expected.

For many IBM i shops, change control has meant source members, libraries, and a promotion tool.

A senior IBM i engineer lays out how he runs RPG, CL, and SQL through Git instead, end to end.

Every change starts as a requirement, gets its own branch, and passes a pull request review before it merges.

The build has to be dependency-aware, since one copybook or service program change can ripple into multiple objects.

And every object in production traces back to an exact commit.

His sharpest warning: reverting a commit does not always undo Db2 changes or data already deployed.

From the Field: I've talked with IBM i modernization experts, and they keep making the same point. Understanding Git is essential.

For an RPG developer, that turns Git from a side skill into a core one. It's how modern IBM i teams review, build, and ship.

If your experience is source members and libraries, start now. Branch, commit, pull request, merge. Practice on your own code.

Adding modern tooling to your RPG skill set? Let's talk about where it opens doors. Book a call.

Database

🗄️ The insurance policy that can pay for IBM i AI

Your HA copy is already an AI asset.

AI on IBM i usually means one of three things: AI that runs on the platform, AI that reads its data, or AI that helps people work on it.

Mixing them up stalls projects.

Here's where it gets practical. If you already fund an HA or DR environment, that replica is a current copy of production with no transaction load.

Point read-only AI work there, and the insurance policy starts earning.

That turns a line item into what the authors call "infrastructure that both produces business value today and strengthens recovery readiness for tomorrow."

Last week, we said it: on IBM i, the data is the application.

The catch is that much of that data speaks in six-character field names and CYYMMDD dates, with meaning that lives in someone's head.

Build a layer of SQL views with plain-English names first. It serves AI, reporting, and the next developer you hire.

Then check replication lag, and keep AI load clear of your recovery targets.

Hiring

📊 The IBM i worry that just passed cybersecurity

Skills is now the top IBM i concern.

For the first time in a decade, cybersecurity is not the top worry in the 2026 IBM i Marketplace Survey.

Skills is. 69% of respondents named finding people who can work on the platform as a top concern.

Modernization came in third, at 62%.

Steve Will, then IBM i's chief architect, made the counterpoint on the results webinar. Teaching IBM i to an experienced developer, he said, is relatively easy.

So the shortage isn't only a supply problem. It's also a question of who you're willing to train.

Most shops still write the req for the senior hire who already knows everything. That person is scarce, and getting scarcer.

The other path is a developer who knows Java, Python, or SQL, learning the platform on the job while your veterans are still there to teach. That's the Hire and Train model, and it's one we offer.

Is your next IBM i hire someone you find, or someone you build? Let's sort out which.

Leadership

🔁 "AI-native" is the new buzzword. Here's the test.

What to ask before AI touches your data.

Kore.ai made Autoloop generally available on October 7. It's an engine that builds, tests, diagnoses, and fixes AI agents against business goals, before and after they go live.

It tracks task completion, accuracy, safety, customer experience, and cost.

A feature called StateTrace records each agent's full execution path, every handoff and tool call.

Kore.ai says it runs Autoloop on its own agents, which produce about 6,500 code commits a month.

The company calls its platform AI-native: built with AI from the ground up, at every layer, rather than bolted onto an existing product.

Kore.ai isn't part of the IBM i stack. It's here for the bigger point it makes about what AI-native really means.

Cut through the pitch: You're going to hear "AI-native" in every vendor pitch this year.

Ask three questions. Was AI built in, or bolted on? Can you trace every decision the agent made? Are the rules enforced outside the model, where it can't talk its way around them?

If your team can't answer those for anything touching your IBM i data, start there.

Learning

🧪 Who tests the code AI wrote?

Code for i Fridays tackles it this afternoon.

When AI writes the code, how do you prove the business behavior your users depend on still works?

That's today's Code for i Fridays topic. IBM Champion and Redbook author Pascal Polverini demos automated IBM i testing with ReplicTest, VS Code, and IBM Bob over MCP, with time for questions.

It's free, hosted by Seiden Group, at 2:30 pm ET. Sign up on the page for the invite.

New to testing on IBM i? This is a practical place to start.

Veterans: if Bob is touching your RPG, this is how you catch what it breaks.

Modernization

🔌 AI can find the rule. It can't tell you why. Yet.

The context gap stalling AI projects.

Most executives couldn't define enterprise context, a Constellation Research analyst argues, but they know when it's missing.

His example: pipeline is down 12%. A good operator instantly asks which accounts matter, what counts as qualified, and who approves the next move.

AI can find and summarize the data. It can't yet supply that context cheaply, consistently, or accurately enough to move most pilots into production.

Once AI moves from recommending to acting, it needs business meaning, current state, policy, and decision rights, not just trusted data.

What AI can't see: On IBM i, the hardest context isn't in the database. It's the why.

Why that flag exists. Why that customer gets the exception. Why the nightly job runs before the invoice batch.

Sometimes the why isn't even business logic. It's just how one person has done it for 40 years.

AI can read the code and the data. It can't recover a reason nobody ever wrote down. It can only guess, and a wrong guess on a live process could stop production.

That's where the human in the loop earns their keep. Capture the why while the person who knows it is still in the building.

Security

🔒 What can your IBM i service account actually reach?

The quiet risk in your API layer.

Most IBM i shops now expose RPG programs to web apps, partners, and soon AI agents.

The usual setup routes every call through one shared service account with far more authority than the job needs.

ZCOMX, from Zanden Telekom, takes the opposite approach.

It's REST middleware that runs as a native ILE program, with no JVM or Node layer.

One config file declares every program, table, IFS directory, and IP address it can touch. Everything else is closed by default.

Each call is logged to a queryable audit trail, tied to the real user when individual authentication is configured.

Why It Matters: Ask your team one question. If someone hijacks the account your middleware runs under, what can it reach?

If the answer takes a meeting to figure out, that's the finding.

The same question is coming for AI. ZCOMX also speaks MCP, so programs exposed as REST endpoints become callable by AI agents under the same config and audit trail.

Whatever tool you use, the boundary should be a document you can read and sign, not a guess.

Trends

🪃 AI is taking the same trip offshoring took

AI is heading back inside the walls.

A June survey of 1,500 enterprise architects and infrastructure leads found 66% moved AI workloads off public cloud in the past year, back to private cloud or on-prem.

84% said AI has pushed their infrastructure costs up.

Governance hurt more. 95% delayed or canceled AI projects over governance, compliance, or regulatory complexity.

Over half did it six times or more.

Asked which environment is hardest to govern for AI, only 9% picked on-prem, the lowest of any option.

It isn't a cloud exit. Hybrid still leads for inference, and spending plans point every direction.

Déjà vu: We've seen this cycle before.

Development went offshore to save money, then came back onshore when companies wanted the work inside their own walls.

Infrastructure went to the cloud. Plenty of it came back on-prem once the bills and the control questions showed up.

Now AI is making the same trip. The first wave ran on frontier models in someone else's data center.

The next wave runs on the company's own data, on its own hardware. That's when enterprise adoption really takes hold.

For many IBM i shops, that's familiar ground. The data, the security model, and the audit trail never left the building.

The question isn't whether AI comes to your data. It's whether your team is ready to run it there.

Did You Know

🔤 Apache Pulsar ran on the AS/400 first. Sort of.

Two chip names, one open source project.

Apache Pulsar is an open source messaging and streaming platform, first built at Yahoo.

Long before that, both words were AS/400 processor names.

Apache was the 64-bit PowerPC chip IBM put in AS/400s in 1997.

Pulsar followed around 2000, borrowing design elements from Apache and built on IBM's new copper-wiring process.

A 450 MHz Pulsar powered machines like the AS/400 Model 270. One configuration listed at $48,401 and supported about 200 users.

No relation to the software. The AS/400 just got the names first.

Join Sign up for Talsco Weekly to get the latest news, insight and job openings for the IBM i professional.

Contact us If you are an RPG programmer looking to explore opportunities or a client who is looking for a talented IBM i professional, please contact us. We look forward to assisting you.

Share

Do you know of someone who could benefit from Talsco Weekly? If so, please use the social media buttons to spread the word. Thank you!