If you spend enough time reading AI newsletters, watching YouTube demos, or scrolling through AI Twitter, it can start to feel like everyone is already living in the future.
Autonomous agents are running entire workflows. Custom copilots are embedded everywhere. Teams are chaining tools, querying vector databases, routing model calls, and deploying AI systems that look like they arrived three years early.
Some of that is real.
But I do not think it is as mainstream as our favorite media suggests.
An adoption ladder for AI in business could be summarized this way:
AI awareness → AI assistance → AI workflows → supervised agents → autonomous systems
On the far left: the aware, curious, cautious, and unsure. On the far right: the courageous early adopters, AI-native teams, and people whose job is to push new tools until they break.
A lot of what we see online lives on the far right edge of that curve. That work matters. It shows what is becoming possible. But it can also distort the picture.
Inside many ordinary businesses, especially small and midsize ones, the AI story looks different.
Some are using ChatGPT manually. Some are experimenting with Microsoft Copilot, Claude Cowork, or similar AI work assistants. Some are watching from the sidelines. Others are wondering whether AI can be used safely at all.
And many are still operating in a world where the bigger problem is not "Which agent framework should we use?" but "Where is the data, who owns this process, and why does this task take twenty steps?" Those are not always obvious questions for the electrician trying to manage the day-to-day of a growing business.
That gap between the extreme ends of the adoption ladder in SMBs is what interests me.
Not the gap between today's model and next month's version, but the gap between what AI can already do and what real businesses are actually ready to use.
That is the now what.
The Internet Shows the Demo. Businesses Need the System.
A demo is supposed to be impressive. That is its job. A good demo hides the ugly parts. It uses clean inputs, follows the happy path, and shows the moment where the model does something that would have seemed impossible a few years ago.
And to be clear, that moment is still remarkable. The models have gotten good. Very good.
But a business does not run on a demo.
A business runs on edge cases, interruptions, exceptions, half-documented procedures, missing fields, old habits, weird software, overloaded staff, legacy spreadsheets, ambiguous responsibility, and customers who do not behave like test data.
That is where AI implementation gets real.
The hard part is rarely getting the model to answer. Ask it anything. It will answer confidently, however wrong its answer may be.
The hard part is making sure the answer is grounded, useful, safe, affordable, maintainable, and connected to work that actually matters.
Most Businesses Are Earlier Than the Headlines Suggest
I do not think small businesses are sleeping on AI because they are foolish or behind.
Many are hesitating because the landscape is confusing, fast-moving, and risky. That seems reasonable.
If technologists have a hard enough time keeping up with the flood of model releases, product updates, frameworks, benchmarks, integrations, and competing claims, why would we expect a medical practice, local manufacturer, accounting firm, construction company, or small agency to sort it all out alone?
They have businesses to run. And they are asking practical questions.
Can we use this safely? Where would it actually help? What data would it need? Who checks the output? What happens when it is wrong? How much will it cost? Will the staff use it? Will this create more work than it saves?
Those are not naive questions. They are the questions that determine whether AI becomes useful.
Before AI Can Improve the Workflow, There Has to Be a Workflow
"If a man knows not to which port he sails, no wind is favorable."
The Stoic philosopher Seneca wrote that to express the simple idea that direction has to come before movement. If you don’t know the correct direction or destination, nothing will help you get there.
That idea applies cleanly here.
One of the least glamorous parts of AI implementation is also one of the most important: getting the business ready.
That means looking at the work before looking at the model. Getting the business ready means asking the right questions before choosing a direction.
Where does the process start? Who touches it? Where does information live? Which steps are necessary? Which steps are historical accidents? Where do things get stuck? What context does a person use that is not written down anywhere?
If a process is chaotic, AI may not fix it. It may just help the chaos get deeper and move faster.
If the data is scattered, inconsistent, inaccessible, or poorly understood, then "add AI" is not a strategy. It’s barely even a hope.
Often, the best first AI-implementation project is not an AI project.
It is cleaning up the workflow. Standardizing the inputs. Removing unnecessary steps. Deciding who owns the process. Making the data usable. Creating the conditions where automation or AI can help instead of adding another layer of confusion.
That may not make for a viral demo. But it is often the real work.
The Real Work Starts After the Model Works
Once the model can do the impressive thing, a different set of questions begins.
Is this the right model for the job? Do we need the most expensive model, or would a smaller, cheaper one work? Is the context well constructed? Are we grounding the output in reliable information? What are the edge cases? How do we test this? Who is accountable for the result? What should be automated, and what should stay human?
And maybe most importantly: how do we make this boring enough that people actually keep using it?
Businesses need systems that survive contact with normal work, not AI that is magical for a week.
That is a different discipline than demo building. It lives in the details: workflow, trust, cost, handoff, adoption, maintenance, and measurement.
And I suspect it is where much of the useful AI work of the next several years will happen.
The Now What
This is the first issue of The Now What.
The premise is simple: AI got good. Time to make it useful.
Each week, I will write about the practical work of getting AI out of the demo and into real businesses, especially small and midsize ones.
Not as a breathless roundup of every model release.
Certainly not as a claim that every business needs agents, vector databases, copilots, or automation everywhere.
And definitely not as a place where "AI" is assumed to be the answer before the problem is understood.
This will be a field report from the implementation side: what works, what does not, what is overbuilt, what is underrated, and what has to be true before AI can actually help.
Because the models working was not the end of the story.
It was the beginning of the next one.