AI Critiques from the Calls

---

Many have been questioning where the promised ROIC is from the capital-intensive AI spend. I can say with confidence that I have seen it.  This AI is a breakthrough in the sector and accomplishes exactly what the user wants (or maybe it's just me). Some might call it AI Slop, and I would be one of them, but it sure is good!

Unfortunately for investors, it was on the side of a beer can from my favorite local brewery, Spyglass Brewing Company, founded by four retired technology and engineering professionals. And yes, that is the actual name and label! The label is quite funny, with people walking above the ground, with more than two hands, sitting on buildings, and melted-looking benches.

---

An interesting development this quarter has been the pivot by some of these technology companies to challenge the AI purist narrative. Software companies are positioning themselves as a necessary tool to supplement the implementation, management, monitoring, and risk mitigation of AI projects.

Palantir

“In contrast, enterprises that are not using Palantir are seeing their token meters spinning endlessly just to get slop without any correlation to value. This token model may be working for the labs, but it is not working for anyone else. It's breaking corporate budgets without results to justify the expense. And worse, companies are paying to give away their most important secrets, the very basis for their competitive advantage, ultimately contributing to the commoditization of their own businesses as their secrets become the training data embedded in the foundations of all future models.” - Alex Karp

Teradata

"We found that 90% expect to increase their agentic AI investments over the next year, yet nearly two-thirds have seen only small or emerging positive returns to date. In addition, 40% of technology leaders surveyed say more than 40% of their AI pilots have failed to reach production because their infrastructure was not built to support them. We are here to change that. We've set a clear vision for this agentic AI era." - Stephen McMillan

"In fact, 77% of executives reported that 20% or less of their data is sufficiently described for agents to use reliably. " - Stephen McMillan

Pegasystems

"...Across the industry, organizations are rethinking how software is designed, built, operated, and evolved. And for a time, AI providers acted a little bit like drug dealers, offering their products for free or charging $20 a month for what felt like unlimited usage. To many users, the experience is magical. But at the same time, the frontier model providers have been investing literally billions or trillions of dollars building the data centers required to power AI.

And now they're going to need to seek a return on that investment, making a significant shift in the economics of the market. These companies are under pressure to generate meaningful revenue, and what was once available for free or for all-you-can-eat licensing is priced now by token use with the attendant anxiety and ambiguity. And it's not done. More of this is coming. The challenge for enterprises is token consumption is opaque until the bill arrives.

Many of the tokens these models consume are reasoning tokens. They don't show up in input or output, but are used by the model itself as it loops through increasingly complex logic. These reasoning costs can become surprisingly and prohibitively expensive. Now this cost uncertainty is leading many, many organizations to sort of freeze and try to figure out what's going on and take a more deliberate approach to technology investments as they assess the economic environment..." - Alan Trefler

Gartner Inc.

“Toni, so you're correct in that clients are reprioritizing their IT spending to shift more towards AI. LLM is one piece of it, but more towards AI. What I'd say is that doesn't impact us in a negative way directly, because if you look at the amount they spend with us relative to their IT budgets, we're very, very small. I think 0.1% or less for typical clients. And so as they reprioritize -- actually, as I mentioned before, the biggest single driver of demand for us for our clients, including IT, is help with AI. So we're actually helping them figure out how do they reprioritize. And so in that sense, it's actually driving demand for our products as we help them reprioritize. And again, across both within IT and across the business, typically, companies are trying to reprioritize to free up more money for AI, and we're helping them with that.” - Eugene Hall

And these strategic pivots are timely, given the recent news that OpenAI, Anthropic, and Meta LLMs have escaped their 'sandboxes' and hacked companies via the internet. Late last week, Chinese firm Moonshot's, Kimi AI broke out of its cyber testing environment.

This is a real concern. Not because it is the precursor to the 1999 dystopian AI movie classic "The Matrix", but because there are considerable risks that can be caused by implementing any immature and not well-understood new technology too quickly. It does not cause the world to be taken over by sentient AI, but it does create material operational headaches and financial liabilities for companies.

Meta explained in their press release that their failure was due to “A misconfiguration by Irregular, an independent testing company Meta uses, inadvertently allowed one of our models access to the internet during evaluation.” And from OpenAI: "After investigating, we now know that this particular incident was driven by a combination of OpenAI models — including GPT‑5.6 Sol and an even more capable pre-release model, all with reduced cyber refusals for evaluation purposes — while being internally tested on a benchmark of cyber capabilities." And from Anthropic: "After reviewing 141,006 evaluation runs where Claude could have obtained internet access, we identified three incidents in which a model accessed the internet from within or while interacting with the evaluation environment of Irregular, one of our third-party evaluation partners, and then gained unauthorized access to the production infrastructure of three different organizations."

In fact, all three companies, Meta, OpenAI, and Anthropic, experienced the AI containment failures while using the AI testing company Irregular. I have been wondering whether or not AI has escaped from any other testing company environments? Is this idiosyncratic to Irregular or a systemic issue?

But this is good news for companies that can provide guardrails to prevent AI from going 'rogue' and is helping to bring expectations of what AI can achieve back down to earth. Software companies that were considered 'extinct' have been revived by the revelation that AI is not infallible and requires considerable structure, data standardization, and guidance to function. Whether this is good for shareholders of AI companies is one thing; I do not think so, but it is certainly a positive development for business leaders who appeared to have unrealistic expectations for AI within their companies.

Just like the bartender cutting off a patron if they consume too much AI Slop, guardrails are necessary for AI, or else the imbiber will turn from a happy patron to a drunken sot. I fear most have over-imbibed on the AI story, and we need to sober up. And the hangover will not be fun the next day.

---

If you would like to subscribe and read our ECE report, published monthly, please use the link below. We greatly appreciate your support. And thank you all for subscribing!

Home | Hickey Advisors LLC l Client Advisory Services

---

The information provided herein is for general informational purposes only and does not constitute accounting, investment, tax, or legal advice. You should consult with a qualified accountant, financial advisor, tax professional, or attorney before making any decisions based on this information. Information presented was obtained from sources believed to be reliable, but accuracy and completeness, and opinions based on this information are not guaranteed. Under no circumstances is this an offer to sell or a solicitation to buy securities suggested herein. All data and information and opinions expressed, are subject to change without notice.