The World Runs on Tech We Should Have Retired, but Most People Are Selling You Agents

Last week, at some keynote, someone with $300 headphones and a clicker in hand said the word "autonomous" for the umpteenth time. Agents that work on their own. The Autonomous Enterprise. The company that runs itself while you sleep.

That same week I went to install the "new" ABAP Development Tools for Visual Studio Code, which SAP announced with great fanfare. The result? Most use cases aren't usable yet. I had to go back to Eclipse, like I've been doing for years.

And there it is, in miniature, today's topic. The distance between what gets announced on stage and what works in the trenches. PowerPoints, paper, and keynotes can hold up anything. The reality of the industry is something else.

I'm not here to bash SAP. SAP is part of how I make a living. I'm here to look at the data with judgment, which is different. Because no matter how married you are to any technology, you can never stop being critical and figuring out what's hype and what's real.

The world runs on dinosaurs that won't die

Let's start with a truth almost nobody says out loud at innovation conferences, because it doesn't make you look so smart and "disruptive." 87% of all the world's transactions run on mainframes.

That 1960s technology we were supposedly going to retire ages ago. 44 of the 50 largest banks on the planet use them. Most of the big retailers. If you used a credit card today, took cash out of an ATM, or booked a flight, you touched a mainframe without knowing it.

And here's the good part. Gartner projects that by 2030, 75% of the providers selling "mainframe exit" services today will pivot their business model or simply disappear. The reason? For most large companies, the volume and interconnected complexity of that data make a complete migration a physical and financial impossibility.

It's not that migrating is expensive. It's that it's impossible. The technology we "should have retired" 30 years ago will still be there in 2030, holding up the world economy on its shoulders. Hold on to this idea, because you're going to need it.

The SAP mirror: the exodus that never happens

Now let's come down from the world to our own SAP backyard. ECC mainstream support ends in December 2027. No more extensions, SAP says, this time for real. The official narrative is that everyone is racing toward S/4HANA and the bright future of the cloud.

But the numbers tell another story. By the end of 2024, only 39% of ECC customers had migrated to S/4HANA. Roughly 14,000 out of 35,000. In fifteen years since S/4HANA launched, SAP convinced fewer than half of its own customers.

More than 60% are still on ECC with no clear path to S/4HANA. And according to DSAG, the German-speaking SAP user group, the industrial heart of the installed base, 54% still run ECC or Business Suite somewhere, and almost half of those remaining plan to migrate as late as 2030, not 2027.

Gartner is even blunter: it projects that more than a third of ECC customers, around 13,000, will still be on the old system in 2030.

And it's worth asking why. Simply because the blockers are always the same, and they're not technical: lack of skills, parallel transformation programs, limited budgets, and complexity accumulated over decades of customization. Companies that call their own environments "fragile, complex, and hard to change," and who are entirely right not to want to touch them if they work.

It's the same pattern as the mainframe. Old technology that works doesn't get replaced just because. No matter how many PowerPoints show 200 agents doing everything "on their own," it's hard to justify moving toward the autonomous enterprise, just like that. It gets replaced when the pain of staying exceeds the pain of migrating. And for a lot of people, that math still doesn't add up.

What your SAP colleagues are actually buying

Here's the part that interested me most, because it comes from SAP users themselves, not from an outside analyst or a competitor. ASUG, the Americas' SAP user group, surveyed 142 members together with Microsoft and Intel. The title of the main finding says it all:

"AI adoption falls short of intent."

Translation: everyone experiments, few deploy.

The sharpest data point in the report is where people fall off. The biggest drop-off happens right at the moment of going from pilot to operational rollout. It's easy to do a demo. It's incredibly hard to put it into real production. Only 51% of the most aggressive adopters, companies over 5 billion dollars, report active deployment. Among the most cautious, that number drops to 15%.

And pay attention to the data point that collides head-on with the Autonomous Enterprise pitch: fully autonomous capabilities are, according to the report itself, "an emerging, near-term phase." In other words, they haven't arrived yet. Only 20% of organizations prioritize autonomous agents. The vast majority want AI that augments human decision-making, not that replaces it.

Let me repeat that, because it's important. While the stage sells agents that work on their own, real SAP customers, asked directly, say they want a copilot, not autopilot. They are not buying what's being sold at the keynote.

And the most revealing part of all: the barriers to scaling AI are not technology or budget. They are always governance, security, and operational readiness. That is, the problem isn't that AI doesn't exist or can't be paid for. The problem is that putting it into real production in an enterprise environment, with compliance and sensitive data, is far harder than a three-minute demo suggests.

The ABAP Tools for Visual Studio Code anecdote isn't an anecdote

Back to my VS Code that didn't work. That's not bad luck on my part or an isolated bug. It's the symptom of a pattern. SAP, like so many large companies, is under pressure to show innovation every quarter and always ends up announcing products before they're mature.

Joule for Consultants inventing ABAP classes that don't exist. Tools that look flawless on stage and on your machine send you back to Eclipse. The Autonomous Enterprise presented as the present when, according to its own users, it's "an emerging, near-term phase."

It's not bad faith. It's the mechanics of hype. The roadmap gets sold in the present tense because the future tense doesn't fill an auditorium or move a stock that's down. But between the announcement and the maturity of the product there's a valley, and that valley is the one you cross, at your job, with your deadlines.

The Operator's judgment: not everyone needs an agent

Here's where I want to be clear, because it's easy to misread me. I'm not saying AI is useless, far from it. I use it at work, I use it every day, I even defend it. What I don't want to say is "do I jump on AI or do I become obsolete?" That's the question of someone who let themselves be scared by marketing. The real question is another: which part of my business, my projects, (myself) actually benefits from this, and which part works perfectly as it is?

There are companies with ECC processes that run like clockwork. Stable, understood, proven over twenty years. What's the real incentive to put autonomous agents into them? Unless you're not afraid to ask the question. And often, none, at least for now. Sometimes the same result is achieved in a simpler, cheaper, and more maintainable way than an agent burning tokens.

And there's the other elephant in the room that nobody on the selling side mentions: cost. There are teams burning their entire annual AI budget in a single quarter, ask Uber if you don't believe me. When you do the fine math, in many cases the tokens cost more than the person who used to do that task. That's why Altman and Amodei drastically changed their "jobpocalypse" discourse, and now things aren't quite as they'd been saying, that AI would wipe out everything by year-end. The economics at the scale of the hype simply don't work for every business model.

That's not being anti-AI. It's doing the math before signing the check. It's exactly what someone with judgment does and a hype buyer doesn't.

The question for this week

The world runs on 1960s mainframes that are impossible to replace. Most SAP customers are still on ECC because migrating hurts more than staying. And those who do adopt AI, by their own account, want a copilot, not an autonomous agent, and they stall on governance, not technology.

Meanwhile, the stage keeps selling the autonomous future in the present tense.

I'm not telling you to ignore AI. I'm telling you to separate the signal from the noise. That your next architecture decision shouldn't come out of a keynote, but out of your own analysis: your real volume, your real cost, your real process.

Because in the end, the one who understands the fundamentals decides what to adopt and what not to. The one who doesn't buys what they're sold and discovers the valley between the announcement and reality once they've already signed.

The PowerPoint can hold up anything. Your system in production can't.

Pablo / En Producción

References

Hong, Euny. "The world still runs on mainframes." 2026. IBM data.
ASUG, Research Report, 2026.
Gartner, via CIO.com. "Nearly half of SAP ECC customers may stick with legacy ERP beyond 2027." 2025.
Kyndryl. "Changing lanes: A faster, smoother path from ECC to S/4HANA." January 2026.
DSAG / SAP Q1 2026 results, via IgniteSAP.
Basis Technologies. "The True State of S/4HANA 2025." November 2025.