AI Ethics: Who Is Telling The Truth About Enterprise AI?
As enterprise buyers navigate uncharted AI waters, who can honestly and ethically guide them?
On this blog, we have written before about the importance in the enterprise software industry of virtue, the highest good according to Stoic philosophy.
Those selling enterprise tech always are under tremendous pressure to meet hard financial goals—each quarter, each year. They may fail to behave virtuously, manipulating their offering and approach, failing to be completely transparent. Our earlier post quotes Stoic philosopher Seneca in de Officis on the risk this poses to those representing products in the market:
"A good man sells a house on account of some defects, of which he himself is aware and others ignorant. Perhaps it is unhealthy, and is supposed to be healthy, — it is not generally known that snakes make their appearance in all the bedrooms, — it is built of bad materials, and is in a ruinous condition; but nobody knows this except the owner. I ask, if the seller should have failed to tell these things to the buyer, and should thus have sold his house for a higher price than he could have reasonably expected, whether he would have acted unjustly or unfairly ... Who does not see what sort of an act such concealment is, and what sort of a man he must be who practices it?"
Analysts should not be complicit in these types of shenanigans. Credible reporting, however, suggests that on the topic of artificial intelligence (AI), many have been compromised. And my own first hand experience suggests broad covert influence campaigns are underway. I wrote about my experience on this blog here, and Wisconsin Watch interviewed me here on how I was pressured to engage in covert, paid postings in favor of a local AI data center project.
Just to be perfectly clear … yes, at Rathmann Insights we are transitioning from work done confidentially to support our construction and industrial software clients and towards public-facing analyst work under our own byline.
But we will align with software vendors who have a viable story to tell about market value. Some vendors will want to work with us. We are fine with that. As an independent analyst firm, we don’t have expensive overhead to liquidate that would require us to take whatever work comes our way. We can afford to tell the truth while larger organizations face institutional requirements that will create pressure to help a vendor dissemble, obfuscate or obscure information essential to business decision makers. Some software and technology vendors don’t like working with analysts in general because they may not be confident their market message will hold up to scrutiny.
And no technology needs more scrutiny right now than enterprise AI. Not only is the technology new, so there is no established proforma for financial decisions, but the financial underpinnings of the industry are illusory, which means companies that build their processes around hyperscaled AI solutions may
Let’s Be Honest About Enterprise AI
Over the last several years, I have seen artificial intelligence (AI) emerge as a business tool, starting with deterministic, algorithmic-based functionality like business rules engines, expert systems and machine learning. I had a front row seat to the rise of intelligent chat bots, which were set to impact the call center industry.
The rise of large language models (LLMs) offered on hyperscale data centers for broad business intelligence or even agentic AI is a relatively new development, and this technology is difficult to sell into enterprise settings without what Cicero would call “acts of concealment” and acting “unjustly or unfairly.”
While AI is important, enterprise buyers intent on driving AI into their technology stacks will want to watch out for foibles a vendor may not mention including:
Uncontrolled Token Consumption: How much will the solution cost depends on usage, and as senior management incents the business to leverage AI, many enterprises are finding costs running ahead of budget without a corresponding business benefit.
Lack of Auditability: How did AI arrive at a decision? Will it make the same decision each time it faces a specific problem? An inference model like an LLM will struggle here.
Cascade Failures in Agentic AI: Multi-agent workflows must be designed to weed out errors one agent makes before they have a knock-on effect downstream.
Latency: Automated decisions in a real-time, dynamic construction environment will require more speed than a hyperscaled LLM can deliver, and often must follow hard sequencing requirements—which is also a problem.
Business Risk: Intermediate term, relying on hyperscalers may drive risk in the business as OpenAI, Anthropic and other LLM vendors are not profitable.
There are Better Enterprise AI Options
We will continue to delve into deterministic AI, and on-premise small language models (SLMs), and have covered better business AI options for construction here before.
What we are soliciting right now is use cases and case studies from construction contractors and asset-intensive businesses on these technologies. Reach out to us directly and we will debrief you on what you are doing and how it is performing.





Comments