Enterprise ROI continues to be elusive
Plus, lots of links and strange finds from the World Wide Web.
Original Content
Here’s a summary of a new post on my weblog:
Three years into the generative AI boom, enterprise ROI remains stubbornly hard to find. Survey after survey shows a familiar pattern: widespread experimentation, lots of “in production” claims, and very little impact on revenue, costs, or P&L. The usual explanation is “culture,” but that’s a dead end for IT - culture change belongs to executives, and most employees don’t believe leadership has a real plan anyway. What can IT actually do to help? We know this: AI-assisted coding is already increasing delivery speed. This shifts the real bottleneck to day two AI operations. If AI 2x or 4x’s the number of apps an enterprise must run, existing infrastructure and operating models will fail. That’s exactly what platforms are built to help with.
I’ve got a longer version of that on my weblog today with plenty of links to those ROI frownies.
Here’s some other things I’ve done since last time:
I’m starting to give up on solo roleplaying with AI. At least, think less of and expect less from it.
Yes, but… here’s my talk on solo roleplaying with AI from AI for the Rest of Us, back in October, 2025.
What do platform engineers need to do to prepare for a flood of AI-generated apps? If enterprises let loose “knowledge workers” with code generation, I think we’ll see 10x the amount of little applications out there. This is going to be a flood of Day Two problems for platform engineers, security, people, etc. So, on this week’s Tanzu Catsup, I asked Tony how he’d recommend handling that flood.
This Conversation is Hardened, Software Defined Talk #556 - This week, we discuss the end of Cloud 1.0, AI agents fixing old apps, and Chainguard vs. Docker images. Plus, the mystery of Dutch broth is finally solved.
AI Doesn’t Fix Bad DevOps: Lessons from 15 Years of DORA Data, with Nathen Harvey, Software Defined Interviews, #118.
Why AI Needs Platform Engineering: The “Day 2” Reality, Tanzu Catsup - out 6th episode. I think we got to good chemistry in this one.
Relative to your interests
“Since we started using Application Advisor, we’ve seen a 70% reduction in engineering time tied to upgrades,” says Roberts. “Now we can focus more of our time and resources on the delivery of new value-added features for our customers.”
New U.S. GSA and Broadcom OneGov Agreement to Help Accelerate Federal Agency AI and Security Initiatives - ”Under this new agreement, [US] agencies can purchase select Broadcom software solutions at a 64% discount off the list price. These offers are valid until May 2027 across the portfolio” Some more coverage.
Scaling Messaging with Confidence: VMware vSphere Kubernetes Service and Tanzu RabbitMQ - The messaging stack for private cloud.
two questions on software “sovereignty” - To have full sovereign IT, you need to build the software from the ground up.
Perplexed - How the normies experience AI news.
Executives shrug off AI bubble concerns, move forward with adoption plans - ”Nearly 9 in 10 executives plan to increase AI investments in the coming year and 46% said they would do so even if a potential AI bubble were to burst, the report found. More than two-thirds of C-suite respondents used AI tools daily in their work, an eight-percentage point uptick from a similar survey fielded in March.”
CEOs Say AI Is Making Work More Efficient. Employees Tell a Different Story. -
Developers Lean on AI More, But Report Growing Doubts About Accuracy, Stack Overflow Survey Says - “Stack Overflow said 84% of respondents are using or planning to use AI tools in their development process, up from 76% a year earlier. In the AI section, 47.1% of respondents said they use AI tools daily, and another 17.7% use them weekly. Yet developers’ trust has not kept pace. Stack Overflow’s analysis said trust in AI accuracy fell to 29% from 40% in prior years, and “positive favorability” dropped to 60% from 72% year over year.”
Why Walmart Still Doesn’t Support Apple Pay - “Walmart doesn’t support Apple Pay because they want to control the customer transaction directly, and they’re big enough, and their customers are loyal enough, that they can resist supporting Apple Pay. Netflix doesn’t support TV app integration because they want to control the customer viewing experience directly, and they’re big enough, and their customers are loyal enough, that they can resist supporting Apple’s TV app.” // The #1 in a market doesn’t have to give up customer control to Apple.
The Revenge of QA: How AI Code Generation Is Exposing Decades of Process Debt - “AI isn’t revealing new problems - it’s exposing decades of process debt we’ve been carrying all along.” / One bottleneck after another…
Morally judging famous and semi-famous people - “spending real time and energy morally judging famous and semi-famous people is one of the best and quickest ways to make yourself stupider.”
best books, essays, and poems of 2025 - What a list! Meanwhile, some Conan stories are now public domain.
The rise of ‘micro’ apps: non-developers are writing apps instead of buying them- Another go at LowCode: “It is a new era of app creation that is sometimes called micro apps, personal apps, or fleeting apps because they are intended to be used only by the creator (or the creator plus a select few other people) and only for as long as the creator wants to keep the app. They are not intended for wide distribution or sale.”
Our approach to advertising and expanding access to ChatGPT - Finally ads coming to AI land. Plus a $8/month tier for everyone, not just India.
🤖 Well, that escalated quickly: Zero CVEs, lots of vendors - The container security rush toward hardened, zero-CVE images is a stopgap. Real security requires controlling the software supply chain and building from source. True security at scale comes from trusted supply chains, not endless patch-and-scan loops.
IBM Launches Sovereign Core as AI Sovereignty Moves From Policy to Operations -
Who pays for agentic AI? The enterprise budget problem no vendor will address- A whole lot thinking about the silo problem for enterprise AI. When a new, expensive technology benefits multiple groups, who decided on it and pays for it?
On Running a Startup of Claude Code Agents: What You Get For a Billion Tokens a Month - SaaS businesses not looking good now.
Microsofts Ex-VP of HR Shares How to Avoid the ‘Underperformer’ Label - Be well known around the company, do some personal marketing, “have an impact” on other teams and make sure people know about it. // When there’s blood on the boardroom floor, there is no “I” in team.
Nadella talks AI sovereignty at the World Economic Forum - Fears about rivals putting your national and corporate interests into AI training. I suppose it’s just like fear of TikTok: that it’s a propaganda tool, as well as just a simple data gathering tool, i.e., surveillance.
Deloitte sees enterprises adopting AI without revenue lift - Yes: “about what benefits AI is actually providing today, 66 percent said it’s improving productivity and efficiency.” But: “How that works when only 20 percent report revenue growth is left unanswered.” // And: “Currently, 25 percent of organizations say they’ve shifted 40 percent or more of their AI experiments into live use. That number is expected to reach 54 percent of organizations within the next three to six months.”
Electricity use of AI coding agents - ‘usage at the equivalent of 4,400 “typical queries” to an LLM, for an equivalent of around $15-$20 in daily API token spend. He figures that to be about the same as running a dishwasher once or the daily energy used by a domestic refrigerator.’ // Get rid of that old fridge in the garage to keep energy usage flat.
🤖 Trapped in the hell of social comparison - Explores how modern social media reshapes economic perceptions by amplifying envy of affluent lifestyles. Argues that algorithmic feeds fuel widespread financial dissatisfaction disconnected from actual economic conditions.
🤖 Failed software projects are strategic failures - Most failed software projects collapse due to strategic misalignment, not bad code. Clear objectives, domain knowledge, and strong internal capability prevent waste and ensure meaningful outcomes. // Coté: if your product idea is shit, even the best code will result in failure.
🤖 Agent Psychosis: Are We Going Insane? - A reflection on how AI coding agents trigger addictive “slop loops,” flooding projects with low-quality output, straining maintainers, and creating insular communities that blur innovation with collective delusion. // Coté: I wonder, though, if this anti-AI stuff is all about aesthetics.
Wastebook
“I have no idea what Federighi’s stance is on break-room bananas, but it seems a stretch to think it offers clues to Apple’s strategy on data centers.” Gruber
“I am here to tell you, with all the gentleness I can muster, that if you couldn’t fix your life before Claude Code existed, Claude Code is not going to fix it now.” Westenberg.
David: Why don’t we have a Jack Daniel’s ad? John: Or if not an ad, just a big bottle of it! The Political Gabfest, Jan 15th, 2026.
“I ain’t flew all the way overseas - in a MIDDLE seat - for us to get over here, and fuck up. Don’t mess it up for e’ryone... Greyhound don’t float on water.” Still the best.
Logoff
No need for panic. Alarm, yes. Panic, no. The TACO theory holds. Stand up to Trump and he’ll chicken out. Gruber.
What a weird week.
Regarding the topic of the article, Nicholas, your analysis on AI ROI is very sharp. That day two ops bottleneck is realy where the challenge lies.