The AI Transformation Brief—October 20, 2026
AI tools are doing harder work, few companies see clear value yet, and each company still answers for what its AI does.
The AI Transformation Brief
// Today’s Signal
This week's AI news is about who answers for the results. Gartner found that only 13% of technology chiefs see significant value from AI, after the average number of AI projects rose 400%. IT Brew Microsoft told companies to treat their own AI tools as possible insider threats and to keep control apart from the supplier. TechCrunch OpenAI reported large gains in rewriting decades-old code, while warning that experts still need to check the work. Data Phoenix In each case the AI does more, and the company still carries the responsibility.
// Top Stories
Gartner surveyed 2,002 chief information officers and other technology leaders from May to July. Only 13% said their company has achieved "significant or transformational value" from AI, even though the average number of AI projects rose 400% between 2024 and 2025. Companies with rules for when an AI trial should stop before wider use were 2.3 times more likely to lead on results, and only 31% have such rules. Companies with plans to redesign roles around AI were 2.4 times more likely to lead. IT Brew
Gartner analyst Alicia Mullery said executive teams are "experiencing FOMO, fear of missing out," and that "volume doesn't necessarily equal value." IT Brew Counting AI projects measures activity. The companies pulling ahead decide what success looks like before they start and agree in advance when to stop. Write those stop rules this quarter, attach a named owner and a target result to every AI project, and redesign the jobs around the work AI now does.
Microsoft added Guardian Agents, a feature that watches the AI tools a company uses, to its Project Perception security program. Its Defender product can now find up to 60 kinds of AI tools employees bring to work, including Claude Code and Codex, limit their access to data, and require a person's approval when one appears to make a risky decision. Chief executive Satya Nadella wrote that companies need to "separate the supply of intelligence from the authority over it." TechCrunch
Microsoft security chief Hayete Gallot said the leading AI systems cannot be trusted because they are unpredictable, whoever makes them. TechCrunch Microsoft also sells AI tools and the checks on them, so read the advice with that interest in mind. The principle holds anyway: the company using the AI owns its actions. Keep the rules, permissions and stop controls with a party separate from your AI supplier.
A supplier's safety promise does not move responsibility. Your company answers for what its AI tools do, so name who sets their permissions, who reviews risky actions and who can shut them off.
OpenAI published new tests for writing code in older and less common programming languages. In a test of moving COBOL programs to C++, Java, Kotlin, Python or Rust without changing their behavior, its newest AI passed 86.6% of cases, up from 5.4% for the previous version, at 61% of the estimated cost. OpenAI said real-world use still requires expert review and testing against each system's business rules, and no outside group has repeated the results. Data Phoenix
Banks, insurers and governments still run core systems on COBOL, and the cost of rewriting them has kept many in place for decades. If these results hold, the cost of moving old systems falls sharply, and the scarce skill becomes knowing the business rules buried inside them. Write those rules down now, test any AI rewrite against your own records before trusting it, and keep the people who understand the old systems.
// Shelly Palmer Pulse
Palmer's October 11 post argues that AI assistants grow more useful as they keep more context, and that "remembering something does not confer permission to act on it." He says people need ways to inspect, correct and limit what an assistant can do. Shelly Palmer Microsoft's new controls apply the same idea inside companies: what AI can do and what it may do are separate decisions.
// What It Means For Your Business
AI now does more real work, while value and responsibility stay with your company.
Before approving new AI projects, set the result each must deliver, the point at which it stops, and the executive who answers for it.
Buyers are separating the companies that supply AI from the tools that watch and limit it.
Security firms and cloud providers will compete to hold that control, and AI suppliers lose the claim that their promises are enough.
Companies leading on AI results redesign roles as AI takes on tasks.
Map which jobs change first, starting with work on older systems, and keep the people who know the business rules those systems hold.
List every AI tool employees use, including coding tools they brought themselves.
Limit each tool's access to the data its task needs, and require a person's approval before risky actions.
// The Take
The shift this week is that AI is taking on harder work, from security decisions to rewriting decades-old code, while proof of value stays rare. The assumption it breaks is that more AI activity produces more results. The decision it forces is which projects to stop, who controls what AI may do, and who checks its work.
How many of your AI projects would survive if each had to name its result and its stopping point today?
Make AI accountable. Measure the results. Redesign the business.
Read On
The AI Transformation Brief—October 19, 2026
The limits on AI this week came from power lines, proof of identity and public trust, and money alone will not remove them.
Read the brief →The AI Transformation Brief—October 18, 2026
AI tools now do much of a top AI lab's work, while a breach, federal test rules and China's draft limits show who controls them.
Read the brief →The AI Transformation Brief—October 17, 2026
Platforms and AI labs are writing the rules for AI that acts on your business, from training rights to account access to federal law.
Read the brief →