- Towards AGI
- Posts
- Only 13% Of CIOs See Real AI Value. Are You One Of Them?
Only 13% Of CIOs See Real AI Value. Are You One Of Them?
here's what the other 87% keep getting wrong
Shen Pandi
October 20, 2026
Today, we're diving into:
Gen AI: Gartner finds only 13% of CIOs are seeing real value from AI
Hot Tea: Two German courts say Google owns every word its AI search writes
OpenAI: Its new models jump from 5% to 87% on COBOL migration tests
Only 13% Of CIOs See Real AI Value. Are You One Of Them?
Two years of pilots, copilots and agent launches should have produced a wave of results by now. According to the people who run enterprise technology, it mostly has not.
A Gartner survey of 2,002 CIOs and technology executives, released yesterday at its IT Symposium/Xpo in Orlando, found that only 13% say their enterprise has realized significant or transformational value from AI so far.
More AI Use Cases Will Not Fix A Value Problem
Find out whether your data, ownership and lineage can support the few AI bets that are actually worth scaling.
Volume Is Not Value
The number of AI use cases has grown over the past two years. The odds that any single one succeeds have not. Gartner puts the chance of an AI use case succeeding at about one in five.
"Volume is not value." Alicia Mullery, VP Analyst, Gartner
Mullery says AI is creating a reckoning for organizations that never defined what value a use case should deliver, or when they would walk away from it.
The Missing Discipline Is Knowing When To Stop
The CIOs Gartner calls vanguards, those who deliver better business outcomes and whose CEOs see high AI value, behave differently in one specific way. They write stopping rules.
Vanguards are 2.3 times more likely to have rules for when a pilot should not go to production. Only 31% of CIOs have them.
Vanguards are 2.6 times more likely to have rules for retiring low-value AI already in production. Only 17% of CIOs have them.
In other words, most enterprises have a process for starting AI projects and no process for ending them.
The Bill Is Becoming Harder To Predict
At the same time, the money question is getting sharper. Gartner says 58% of CIOs face more pressure to deliver AI cost improvements, and 51% expect AI to raise the total cost of ownership of their technology.
The reason is structural. Traditional software costs are largely fixed once you sign the contract. AI costs, in Mullery's words, are "probabilistic and determined at run time."
This Fundamentally Changes How You Fund AI
Most leaders will react by cutting the AI budget or freezing new pilots. That is the wrong reaction. The problem is not that you are spending on AI. It is that you cannot see which spending is producing value.
The better alternative is to measure each workload by what a completed, useful task costs you, and to compare that against the value it creates. Know Your Inference (KYI) is built for exactly this: understanding how every inference affects cost, latency, accuracy and governance, so you can route each workload to the right model and retire the ones that never pay back.
What You Should Do Before Your Next AI Budget Review
Define value up front. No pilot starts without a written outcome and a number it must beat.
Write your exit rules. Agree in advance when a pilot dies and when a production system gets retired.
Make the business co-own the call. Gartner is clear that walking away from low-value work is a team sport, not an IT decision.
Plan for a human-agent workforce. Only 4% of CIOs have plans to redesign roles around agents, and Gartner calls that the true vanguard differentiator.
None of these rules work if your teams cannot trust the data each use case runs on. Building AI that earns its keep starts with trusted enterprise context. DataManagement.AI helps organizations create that foundation by unifying metadata, lineage, governance, business glossaries and knowledge assets into a single intelligence layer, so every AI use case starts from data your business already agrees on.
The enterprises in the 13% will not be the ones that launched the most AI. They will be the ones that knew which AI to stop.
German Courts Just Made Google Own What Its AI Says
For years, search engines have argued they simply point you to what others wrote. Generative AI quietly broke that argument, and two German courts have now said so out loud.
As The Decoder reported on Sunday, courts in Munich and Aachen have both treated Google's AI-generated answers as Google's own statements, not neutral search results. That makes Google directly liable for what they say.
Two Rulings, One Message
The first case came from the Munich I Regional Court in late May. Google's AI Overviews had falsely linked two Munich publishers to fraud schemes they had nothing to do with.
The court classified the output as Google's "own statements." Google announced an appeal, then settled instead, so no higher court has reviewed the reasoning.
"The respondent provides AI-generated search and is therefore also responsible for the answers generated in this context." Aachen Regional Court
The second case, an injunction from the Aachen Regional Court dated September 24, concerned Google's AI Mode wrongly tying a senior physician to criminal proceedings that involved someone else. The court relied on the Munich reasoning and set fines of up to 250,000 euros per violation. The order is not final, and Google can still challenge it.
The Detail That Should Worry You Most
According to the court, the physician had contacted Google in late August and September. In mid-September, Google said it would take no action, and the false texts stayed live.
The legal exposure did not come from the AI making a mistake. It came from the company not having a way to correct it once it was told.
Why This Is Not Just A Google Story
Most leaders will file this under Big Tech legal trouble. That is the wrong reaction. If your chatbot, support agent or AI search summarizes information about customers, suppliers or people, the same logic can land on you.
This is not a model accuracy problem. It is an accountability problem. Courts are starting to treat what your AI generates as something you said.
Treat AI output as published content. Apply the same review standards you would to anything your brand puts its name on.
Build a correction path. Someone harmed by an AI answer needs a fast route to a human who can fix or remove it.
Log what your AI says. You cannot defend or correct an answer you cannot reproduce.
Ground answers in governed sources. AI that draws from vetted, owned data is far less likely to invent claims about real people.
The better alternative is to design accountability into AI products from the start, rather than bolting on a disclaimer after the first complaint arrives.
The companies that scale AI safely will not be the ones whose models never make mistakes. They will be the ones that can find and fix a mistake within hours of being told.
OpenAI's Models Now Speak COBOL. Do They Know Your Business Rules?
Somewhere in your organization, a decades-old COBOL system still settles payments, calculates interest or runs payroll. Everyone wants to modernize it. Very few people still understand it.
On Monday, OpenAI published new research on low-resource programming languages such as COBOL, Fortran and Visual Basic, with two new evaluations, one of them built specifically around COBOL migration.
From Almost Useless To Almost Right
The jump is large. On OpenAI's COBOL migration evaluation, GPT-6.1 Sol reached 86.6% accuracy at low reasoning effort, compared with 5.4% for GPT-5.6 Sol at the same setting. At higher reasoning effort, OpenAI says accuracy reached 97.9%.
A migration only counts as successful if every hidden test passes, and those tests require the new code to match the old system's observable output byte for byte.
On its broader MultiLangBench test across 20 languages, GPT-6 Astra averaged an 87.4% pass rate, with COBOL rising from 10.9% to 72.1%. These are OpenAI's own figures, and OpenAI has published the test suite on GitHub.
The Bit Most People Will Miss
OpenAI itself points to the trap. COBOL's fixed-point arithmetic truncates extra decimal digits unless rounding is requested. A rewrite that rounds instead can run without a single error and still produce different financial results.
The code works. The numbers are wrong.
That is why OpenAI notes that production use still requires expert review and testing against the system's business rules.
The Real Bottleneck Is Not The Code
This fundamentally changes the economics of legacy modernization. Translating code is becoming cheap. Knowing what that code is supposed to do is not.
Most legacy systems carry decades of rules nobody wrote down: a rounding choice made for a regulator, an exception added for one client, a field that means something different in one region. An AI model can translate the syntax perfectly and still miss all of it.
Capture the rules first. Document what each critical program actually decides before anyone rewrites it.
Test against outputs, not opinions. Copy OpenAI's own standard and compare old and new systems on real outputs.
Keep experts in the loop. Use AI to speed up your few remaining COBOL experts, not to replace their judgment.
Know the cost per migration. OpenAI says GPT-6.1 Sol cost 39% less than its predecessor at the low setting. Measure what higher reasoning effort really buys you before defaulting to it.
Modernizing safely starts with trusted enterprise context. DataManagement.AI helps organizations build that foundation by unifying metadata, lineage, governance, business glossaries and knowledge assets into a single intelligence layer, so you can see where legacy data flows, what depends on it and what each field means before an AI model starts rewriting it.
The organizations that modernize first will not be the ones with the best code translator. They will be the ones that understood their own business rules before the translation began.
Journey Towards AGI
Research and advisory firm guiding on the journey to Artificial General Intelligence
Know Your Inference Maximising GenAI impact on performance and Efficiency. | Model Context Protocol Connect with us, and get end-to-end guidance on AI implementation. |
Your opinion matters!
Hope you loved reading our piece of newsletter as much as we had fun writing it.
Share your experience and feedback with us below 'cause we take your critique very critically.
How's your experience? |
Thank you for reading
-Shen & Towards AGI team