A Dash of AI

Created: September 30th, 2026 Published: September 30th, 2026

In a more mature era of AI enablement, the question is no longer whether to adopt—it's how much. For most enterprises.

The Renaissance of AI has matured into a state of more broadly understood terms and acceptance of the capabilities upon which our reliance has formed. Implementations of endless shapes are now underway throughout businesses of all size and purpose.

With seemingly countless hours of contemplation invested in the understanding of middleware, model capabilities, efficient use, infrastructure requirements and security considerations (a mouthful I know) all thought necessary to professionally advise, I now find myself seeking notice of impactful innovation. The constant sprint to keep up with training methodology, modalities, hardware advancements, trial cases and enterprise use cases has now morphed into a state of mastering all this enablement at our disposal.

What once seemed as an endless sprint to maintain understanding of the constant introduction...

to training methodology, modalities, hardware advances, and new use cases has morphed into something else: actually mastering the tools we have.

From my perspective, Open Claw was the last swell of the set, introducing asynchronous execution, workspace functionality and the agent.md file—a notion of the agent skills or bundled, measured and refined prompts that enable specific capabilities of cross-platform model integration.

This gave way to an evolution embraced industry wide, touching all but few of the offerings in the consumer and mid-market space. Anthropic, X and OpenAI followed suit in an instant, making use of the skill capability and vastly expanded connectivity, the end result an unbounded integration harness presented in a more recognizable format, empowered both with proprietary and competitor models, now, all but defines the consumer AI experience.

This was the stepping off point for me, in conjunction with an objective to thoroughly understand virtual compute "cloud" as it's known, I found myself equal parts confident and exhausted with the endless progression, unsure of when it would prove wise to build.

Time and again, aware of the potential for leaps in productivity, I found myself intrigued to dive in, lease a server, download some multiple trillion parameter model, license a framework in preparation to build Jarvis, my conscious intelligence engine (conceptualized like many others, since first watching Ironman) equipped with all modes of interface, success in all areas would be effortless. Fortunately, I held off.

An ounce (or year) of patience spared some decision makers from years or millions of sunk investments.

Past experience had taught me the pace of innovation would lead to a more favorable economy for the technology and just as sure a greater understanding of the costs, along with the benefits, and risks along with rewards. An ounce (or year) of patience spared some decision makers from years or millions of sunk investments.

Much the same as the flat screen monitor, the cell phone, the CD player and walkman, in time, the supply will meet the demand, one thought, but is the time lost, the gap by which you trail your competitors position?

A more complete understanding proved favorable for all but a few in this latest generation of technology, while the benefits are more widely realized every day, the risks, shortcomings and surprise expenses, of less than sufficient understanding have negated times over, the benefit realized by more than one brave executive who deployed their investment in those early moments.

I was surprised to catch an AI enabled tool, synthesizing a response it was not capable of providing.

On a recent build (albeit more of my time has been spent in understanding than with engagement) I was surprised to catch an AI enabled tool, synthesizing a response it was not capable of providing. To think we had gone through all of this, the schema, the intended use, the framework, the hosting... Then to realize the tool was producing confident-sounding nonsense from the very technology I have paid to leverage. It is called Hallucination, and it is one of several notable shortcomings which must be understood. Fortunately, the model proved effective, after some modification to the framework.

Cost is another consideration, a more simple understanding at the surface, however one which can be impacted by a number of factors in the deployment. The size and type of the model, the prompts, the compute resources, even the approach utilized by the developer ultimately impact the cost and quite possibly will prove to be the determinant of economic benefit or involuntary education.

I have enjoyed a multitude of exploratory trials with a variety of generative models; and believe with certainty as many others, empowering our designers, researchers, innovators and thought leaders with chat or voice driven inference of the flagship or frontier variety will drive our creativity forward, vastly expand and increase the speed of our innovation and lead humanity as a whole, and business as a unit, to places never thought possible.

I use tools such as Claude CoWork, Grok Build and OpenClaw, empowered with the latest model offerings on a daily basis and my capabilities have expanded beyond my wildest expectations. However, this is one mind in mostly creative, educational and exploratory engagements, when it comes to predictable cost, reliability and consistent structured output, a more narrow and specific deployment of "AI" capability will likely prove successful.

Most enterprises are built on the understanding of predictability, repetitive processes which can be anticipated with the objective of incremental linear growth.

While there are circumstances where a relatively large parameter set will be necessary, such as anticipation of external economic forces or optimization of supply chain variables, enabling operational capabilities will not.

Now in a more mature phase of enablement, the efficiency model or transformer model is better understood, this a capability that enables only a fraction of parameters in a given model, the part necessary for the job at hand, vastly reduces compute cost. This is a great option for an organization that is unwilling to invest in retraining or distillation of specific model capabilities.

Another deployment to consider is a model of a previous generation, deployed in a modernized harness, often referred to as "orchestration" model, its cost is often pennies on the dollar and their limited scope is intended at times for a singular purpose.

I've come to believe most executives require a specific level of cost predictability and operational certainty, they are not in a rush to risk an outcome.

In this case, a traditional, logic based approach which is entirely predictable, can be expanded to open an entire new realm of transparency or capability with a singular enablement of specific AI functionality.

In these cases and I believe there are many, just a dash of AI will do.

— Mike

---