Welcome to another edition of Talsco Weekly

In This Issue

• IBM i Briefs: 🎤 IBM i's new chief architect takes the stage. 🗺️ Your IBM i guide to TechXchange week. 📈 Power's quiet growth streak keeps going.
• AI: 🔇 The AI model that won't say a word.
• Development: 🛠️ The real reason IBM i shops don't change their code.
• Database: 🗄️ Your AI plans start in the database.
• ERP: 🏭 Infor's new AI agents are open to IBM i shops.
• Hiring: 🎯 Before you hire a junior, check your bench.
• Leadership: 🎛️ Everyone's about to be a manager.
• Learning: 🎓 The AI skill hiding behind every prompt.
• Modernization: 🔄 Your "leave it alone" list needs an expiration date.
• Open Source: 🧩 The open-source fight over AI's real bill.
• Security: 🔒 AI found a 10-year-old hole in 70% of the web.
• Trends: 🎼 The RPG Developer who stopped opening the IDE.

IBM i Briefs

🎤 IBM i's new chief architect takes the stage

Fall TRs, and a new face presenting them.

The fall IBM i Technology Refreshes are expected next week, likely 7.6 TR3 and 7.5 TR9.

The timing fits. COMMON's NAViGATE and IBM TechXchange run together in Atlanta, Oct. 26 to 29, and IBM has a habit of announcing TRs around its biggest user conferences. Nothing's official yet.

The bigger signal may a Nov. 4 Guided Tour on new security, performance, and modernization enhancements that features Kris Whitney, who took over as IBM i Chief Architect after Steve Will retired.

Pass the Baton: Steve Will was the face of IBM i strategy for years. His successor steps up with a bench of senior leads already beside him.

That's the succession move every IBM i shop needs. Name the next leader before the veteran leaves, and put them in front of people early.

Want the new team's view firsthand? Register for the Nov. 4 Guided Tour.

🗺️ Your IBM i guide to TechXchange week

One pass, two conferences, starting Monday.

If you're heading to Atlanta Monday, here's the shape of the week.

NAViGATE runs inside IBM TechXchange, Oct. 26 to 29. One $1,295 pass through COMMON covers both, $304 less than IBM's rate, with no membership required.

IBM i gets its own hub with curated tracks: Db2, security, high availability, RPG, PHP, Python, and AI.

Sessions worth circling include what's new in Db2 for i 7.6 and the latest TRs, Bob and security, MCP with IBM i, and Migrate While Active.

Monday opens with hands-on labs and certification testing.

Browse the IBM i and Power sessions.

📈 Power's quiet growth streak keeps going

Three quarters, three double-digit jumps.

Three quarters into 2026, the part of IBM that sells Power keeps growing fast.

Distributed Infrastructure, IBM's bucket for Power and storage, rose 17% in the first quarter, 37% in the second, its best on record, and 31% in the third, reported this week.

IBM's filings credit demand for Power11 and growth in both Power and storage. Through nine months, each is up double digits.

The rest of the quarter looked solid too. Revenue hit $17.1 billion, up 5%, led by software.

The IBM i Angle: Not every new Power11 box runs IBM i. Some run AIX, some Linux, many run a mix. But plenty run IBM i, and every one of them needs someone who knows the platform.

Growth on the hardware side shows up on the hiring side a few quarters later.

AI

🔇  The AI model that won't say a word

Jev decides. It doesn't write.

Every AI tool you've tried writes text. Jev doesn't.

It's a new kind of model from TypeSafe AI, now in early access, and IBM's technology channel walked through how it works. You hand it a situation and a fixed set of answers. It picks one and tells you how sure it is.

Which team gets this ticket? Is this order urgent? Is this AI answer safe to send?

The price is the headline. Input runs about four cents per million tokens, and output is free. Independent researchers ran 346,009 test requests for under $10.

Our Take: Most of what an IBM i shop runs every night isn't writing. It's deciding. Route the exception. Flag the order. Hold the shipment.

Paying a big LLM to make those calls one word at a time gets expensive fast. A model that just decides, and says when it's unsure, fits that work better.

The confidence score is the real story. How do you let AI make a call without trusting it blindly? Set a threshold. Above it, the system acts. Below it, a person reviews.

That's the guardrail. And someone who knows your business rules still has to set it.

Development

🛠️ The real reason IBM i shops don’t change their code

It’s not stubbornness. It’s not knowing what breaks.

“We’ve always done it that way.”

COMMON President Peg Tuttle has heard it for 30 years, and says it’s sometimes a fair answer.

Decades-old IBM i applications still run the business well. The documentation never kept pace.

The hard part isn’t the change. It’s knowing what it breaks.

New AI tools can map program calls, files, and dependencies, capturing what lived only in people’s heads and helping newer developers ramp up faster.

Her takeaway: projects once too risky deserve another look.

Pass the Baton: A map shows what connects. It doesn’t show why.

Why that file clears every Tuesday, or that program skips one customer. That lives with the people who built it, and many are nearing retirement.

“We’ve always done it that way” is often the only documentation there is.

Capture the why while they’re still here. Pair them with a newer developer and the map, and write it down.

Cheapest modernization project you’ll ever run.

Curious what your system already knows about its own dependencies?

Database

🗄️ Your AI plans start in the database

And DDS has been frozen for 20 years.

Plenty of shops want AI-ready IBM i applications. Db2 for i expert Birgitta Hauser says that road starts somewhere unglamorous: your file definitions.

DDS hasn't gotten a new feature since V5R3. Every new database capability now arrives in SQL.

Her verdict: anyone after future-proof, AI-ready apps "must move to SQL DDL."

The good news is the first step is small. Convert a physical file and change nothing else, and native I/O programs don't need a recompile.

The bad news hides in your data.

Reality Check: DDS never checked numeric data on write. SQL does. Convert a file with one bad packed field and the conversion fails at that row.

How many bad packed fields are hiding in your files right now? IBM's free SYSTOOLS scan will tell you. Run it first.

Then fix the programs that wrote them, because after the move those writes fail instead of slipping through.

Same lesson as this week's AI blocks: rules the system enforces beat rules people remember.

ERP

🏭 Infor's new AI agents are open to IBM i shops

LX, XA, and System21 users aren't left out.

Infor launched a new release of its Velocity Suite this month: AI agents, generative AI, process mining, and automation built for specific industries. Dozens of prebuilt agents ship with it.

The question IBM i shops always ask: does it work for us? Infor says yes.

Customers running LX, XA, or System21 on-prem can use the agents by running Multi-Tenant Infor OS alongside the suite. The agents themselves run in Infor's cloud.

Infor says some IBM i customers, including an XA shop, are already on board.

What to Watch: Agents that execute workflows, not just suggest them, need someone who knows the business process behind the green screens.

That's your ERP power user, and the developer who understands order-to-cash. Before you turn agents loose on your ERP, decide who signs off on what they do.

Hiring

🎯 Before you hire a junior, check your bench

Who's going to teach them?

Every block in this issue points the same direction. AI writes more of the code. People decide whether it's right.

That changes who you hire first.

A junior developer trained on AI tools can ramp fast. But only if someone senior is there to review the work, explain the business, and say no when the agent is wrong.

Large shops have that bench. Many small shops don't. For them, the first hire isn't a junior. It's a senior who can mentor, or a fractional expert who can set the guardrails before anyone else joins.

Not sure which hire comes first? Let's map it out. Book a call.

Leadership

🎛️ Everyone's about to be a manager

Even if nobody reports to you.

A product chief's team runs on AI agents now, and he doesn't call it using AI. He calls it managing it.

Each person trains an agent like a junior hire, sets expectations, reviews plans, and owns the output. A human signs off before anything moves forward.

The engineers who build these agents say the same thing: keep it simple, set guardrails, and build in human checkpoints.

The results are hard to ignore. Prototypes in a day instead of weeks. Release velocity doubled.

His conclusion: in any role, "everyone becomes a manager of agents."

Which means managing well just became everyone's job.

Our Take: So manage them the way you'd want to be managed. Clear guidance. Real guardrails. Then let them work.

Think about the manager who reads over your shoulder all day. How productive were you? Annoying, right?

Nobody does their best work for a micromanager. Your agents won't either.

Even this team micromanaged early, until the agents earned trust. That's onboarding, not leadership. Stay there and you've bought speed you never get to use.

Could you brief an agent on your IBM i system today? If that answer lives in one senior developer's head, start writing it down. It trains your agents and your next hire.

Learning

🎓 The AI skill hiding behind every prompt

Learn how LLMs answer, one step at a time.

Earlier in this issue, Jev's pricing came down to tokens. Here's where those tokens come from.

LLM Inference.- IBM

In about seven minutes, Grace Ableidinger, a developer advocate at Red Hat, follows a single prompt through an LLM:

• Your text becomes tokens.
• The request lands on a GPU, batched with others.
• Prefill reads your whole prompt at once and fills a cache.
• Decode writes the answer one token at a time.
• Two numbers measure speed: time to first token and time between tokens.

New to AI: Watch it once, and AI pricing, speed, and limits start making sense.

For IBM i veterans: Batching and response time are concepts you already manage. Different hardware, same tradeoffs.

Prefer to read it? Here's IBM's written walkthrough.

Modernization

🔄 Your "leave it alone" list needs an expiration date

The quadrant where modernization goes to die.

Every application portfolio has a list of programs nobody wants to touch. They work. They're old. Leave them alone.

In Gartner's TIME model, that's Tolerate, one of four dispositions alongside Invest, Migrate, and Eliminate. A new walkthrough argues it's the most misunderstood box: "Tolerate is a timed hold, not a permanent state."

Every Tolerated application should carry a review date, typically 12 to 18 months out, plus triggers that force an early look. An unpatchable vulnerability. A vendor end-of-life. A key dependency retiring.

The Talent Angle: On IBM i, there's one trigger the framework misses: the retirement of the person who knows the code.

When that developer leaves, every Tolerated program becomes a mystery overnight.

Put a review date on the people, not just the programs.

Open Source

🧩 The open-source fight over AI's real bill

llm-d 1.0 takes aim at inference costs.

Training gets the headlines. Inference gets the bill.

This week, IBM, Red Hat, and Google, the lead maintainers of the open-source llm-d project, released version 1.0. It sits on top of inference engines like vLLM and handles the routing and operations needed to run large models on your own infrastructure.

Remember the KV cache from this issue's Learning video? llm-d routes each request to the server that already holds its cached context.

In one demo, IBM and Red Hat ran 3,000 coding agents at five to 10 times lower cost than a hosted service.

The IBM i Angle: None of this runs on IBM i. It runs on GPU clusters.

But it's the same choice IBM i shops have always made: own the stack or rent it. As AI costs climb, keeping inference in-house starts to sound a lot like keeping your data on Power.

Security

🔒 AI found a 10-year-old hole in 70% of the web

PHP2Shell, explained in under a minute.

So what happened, actually?

PHP powers roughly 70% of websites, including WordPress, Drupal, and Joomla. Researchers using AI found two critical flaws in PHP itself that let an attacker run code on a server without logging in.

The code had been sitting there for more than a decade.

The bugs were found this summer and kept quiet while PHP built a fix. Patches shipped Wednesday. The NSA's advice: patch, or take it offline.

What changed was AI. One researcher did in a week what used to take an elite team months.

Why It Matters: For years, old code got a pass because it was hard to pick apart. That protection is gone. AI can read a decade of code in days, for defenders and attackers alike.

For security teams, the patch window just got shorter. For developers, the job tilts toward testing and review, because AI will find what you missed.

And yes, this reaches IBM i. Plenty of shops run PHP on the box.

Is the PHP on your IBM i patched? IBM i builds are out for supported versions. Anything 7.2 or older needs an upgrade.

Then check your vendors. If your inbox looks like ours this week, it's full of notices from software providers confirming they've patched. Running something built on PHP and haven't heard from that vendor? Reach out and ask.

Trends

🎼 The RPG developer who stopped opening the IDE

Watch where IBM i development is headed.

For forty years, writing RPG meant living in an editor. SEU, then RDi, now VS Code.

The tools changed. The job didn’t. Until now.

I talked with Profound Logic last week. Their platform, CoderFlow, takes the IBM i developer out of the IDE.

So what is it, actually?

A developer describes the change. Several AI agents build it in parallel, then compile, test, and fix until it works.

A judge agent scores every result. The best one waits for the developer to approve.

Profound’s video puts it plainly: “The developer does not disappear, they become the orchestrator.” You design the change, direct the agents, and sign off.

Profound has often been early on IBM i. They told me they rebuilt their own development team around agents, moving every developer into an orchestrator role. Watch the video.

But here is the bigger signal: in shops adopting AI, talk of leaving IBM i is fading.

That’s the best of both worlds: keep the platform and decades of business logic, and add the agents on top. No starting over.

And it isn’t just IBM i. Salesforce developers, mainframe teams, and developers on their own desktops describe the same shift: hand off the task, then direct and review.

Different platforms. Same job change.

Our Take: There’s a rule of thumb making the rounds in AI circles.

Call it 10-80-10.

• The first 10% is human: knowing the business and designing the change.
• The middle 80% is the agents: building, testing, retrying.
• The last 10% is human again: verifying the result and owning it in production.

The person sits at both ends. Not in the middle, typing.

Developers writing about agent-built code land in the same place: check the plan before, check the result after. It’s also why Profound built its agents to stop and wait for the developer.

That isn’t the end of the IBM i developer. It’s a new job description, and whoever knows the business, the system, and the architecture just got more valuable. See the new orchestrator at work.

This has happened before. The spreadsheet didn’t end accounting. It ended hand-tallied ledgers, and the profession grew.

But the shift is real. Developers who only want to code, and shops that won’t adopt AI, are both carrying risk.

So, for your team: what does a developer actually do when the agent writes the code?

Making the jump to orchestrator? Let’s talk about your next move.

Hiring for it? Let’s talk about the role.

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!