Talsco Weekly: IBM i: Keep It In the Building

Welcome to another edition of Talsco Weekly

IBM i Brief: 🎟️ COMMON is bringing IBM i to a bigger stage. 📉 Deferred is not the same as dead.
AI: 🪨 Granite is the AI model you can run yourself.
Development: ⚙️ The AI workload does not have to leave the building.
Hiring: 🧩 Hiring IBM i devs? Skip the syntax trap.
Learning: 🧊 The model is becoming the chip.
Modernization: 🧰 IBM made you wait. Here is why.
Open Source: 📊 Nobody surveys the parts that already work.
Security: 🛡️ When the guardrails blocked the good guys.

IBM i Brief

🎟️ COMMON is bringing IBM i to a bigger stage

NAViGATE at IBM TechXchange pairs COMMON’s IBM i and Power content with full access to the broader conference: AI, hybrid cloud, Red Hat, and open source. Familiar speakers, the same community, wider reach. COMMON President Peg Tuttle describes it as a much bigger playground. Registration runs through COMMON at special pricing, and membership is not required.

Our Take

We have sat in a lot of rooms. Regional user groups. COMMON events. NAViGATE more than once. PowerUp has been the anchor event for years, and it earned that spot. But this is different. Putting IBM i content inside IBM’s largest technical conference changes who walks past the room. It puts RPG developers in the same hallway as the AI and hybrid cloud crowd. That visibility is hard to buy and harder to manufacture. This could flip the script on where the IBM i community shows up.

Your career is your business. If your employer covers the ticket, take it. If they do not, look at the number differently. Treat it as an investment in you. The relationships you build, the skills you sharpen, and the visibility you walk out with are not an expense.

📉 Deferred is not the same as dead

IBM missed Q2 badly. Revenue landed at $17.2 billion, the stock fell 25% in a single day, and mainframe sales dropped 42%. Wall Street read it as decline. IBM read it as delay. Customers absorbing 15% to 30% AI-driven hardware price hikes pushed mainframe orders out a quarter. Some have already placed them.

IBM reports no clients leaving the platform.

Our Take: A bad quarter changes stock prices. It does not change who runs the systems. The customers who pushed their orders to next quarter are still running the same workloads, on the same platform, with the same aging teams.

Nothing in this report touches IBM i. Nothing in it reduces demand for people who know RPG, Db2, and how a real production environment behaves.

Wall Street is pricing a quarter. Succession planning is priced in decades.

AI

🪨 Granite is the AI model you can run yourself

IBM’s Granite 4.1 is an open model family in 3B, 8B, and 30B sizes, all under Apache 2.0. It does tool calling, code generation, instruction following, and math, with real gains over 4.0. Run it on Ollama, vLLM, or LM Studio, on your own hardware. Cryptographically signed and ISO certified. Enterprise AI without shipping your data to someone else’s cloud.

Why It Matters: Granite is exactly the kind of model the Hugging Face breach argued for. One you own and run yourself (see Security below). But the real job for IBM i developers is fluency. Open weights, frontier models, agentic tools. They are not interchangeable, and knowing the role each plays is becoming a core skill.

Development

⚙️ The AI workload does not have to leave the building

IBM’s Power S1112 arrives July 24. One socket, Power11, built for compact on-premises deployment, and able to run AI inference locally.

IBM claims twice the core performance of the Power S914 and three times the S814, with up to 69% greater energy efficiency than the S914.

Entry-tier hardware that runs inference on site changes what a small dev team can prototype.

Hiring

🧩 Hiring IBM i devs? Skip the syntax trap

You don’t need to be technical to run a great interview.

Open the whiteboard. Ask them to outline your department’s core functions.

Share an actual problem and have them diagram their approach.

Ask for a rollout plan across six months.

Then ask what they’d add to your bottom line.

Their answers reveal fit, not just skill.

Learning

🧊 The model is becoming the chip

For years, AI ran on general-purpose chips. Load a model, shuttle its data in and out of memory, compute, repeat. Google’s reported Frozen v2 flips that. It etches Gemini’s architecture into the silicon itself. Fewer steps. Faster answers. Up to ten times more efficient.

As we all learn AI, here is the shift to watch: compute is climbing down from software into hardware.

Frontier, open, or on-chip? Where IBM i fits.

All this AI talk blurs together. Here is a clean way to hold it. Three layers, three different jobs.

Frontier models. Strongest at general reasoning and code. Licensed and cloud-hosted. IBM’s Bob IDE runs on Anthropic’s Claude. You use these. You don’t host them.
Open models. IBM’s Granite family is smaller, open, and strong on RPG. Small enough to run yourself, where your data lives.
AI-ready hardware. Power11 adds on-chip acceleration plus the Spyre Accelerator, inference silicon built to run open models on-prem, inside your own walls.

The key word for Power11 is inference, not training.

It serves models, it does not build them. Reach for a frontier model when you need raw power and the cloud is fine. Run Granite on Spyre when the data has to stay home, pulling straight from Db2 on the same box.

Both worlds coexist. Knowing which to use and when is the new skill.

Modernization

🧰 IBM made you wait. Here is why.

IBM i 7.6 TR2 and 7.5 TR8 were announced July 15 and hit general availability that Friday.

Technology Refreshes add operating system support for new hardware, so IBM held them until the entry-level Power S1112 was ready. The result is a steady, incremental release.

Bring Your Own Key arrives with 7.6 TR2 only.

Open Source

📊 Nobody surveys the parts that already work

The LibrePower 2026 open source survey for IBM Power has closed, and its question set is the interesting part.

Package management via yum and ACS. Python, Node.js, and Git adoption. ppc64le container availability.

Vendor trust across IBM, Red Hat, SUSE, and independent maintainers.

An independent group built a benchmark for the Power ecosystem and made it entirely about open source.

Read that as a signal.

Security

🛡️ When the guardrails blocked the good guys

Hugging Face got breached by an autonomous AI agent. The strange part came next. Western frontier models refused to assist forensics, unable to tell an attacker from an incident responder. So the team turned to an open-weight model they could run themselves. The lesson lands hard: have a capable model vetted and ready on your own infrastructure before an incident, not during one.

Our Take: The new security frontier looks nothing like exit points and object authorities. Agentic attacks and guardrail lockout aren’t problems most IBM i shops plan for yet. That cautious posture means you get to watch before it lands on your doorstep. When you’re ready to think through AI-era data governance, we can point you in the right direction.

As Hugging Face put it:

The practical lesson for defenders: have a capable model you can run on your own infrastructure vetted and ready before an incident, both to avoid guardrail lockout and to keep attacker data and credentials from leaving your environment.

Join

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.

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!