John SorianoThere is a version of the AI conversation that is loud everywhere: frontier models, autonomous agents, billion-dollar platform shifts, and promises that entire industries will be reinvented overnight.
That conversation is real, but it is not the most useful starting point for most businesses.
The more useful conversation is smaller and more concrete: what is one thing your team does repeatedly, that takes longer than it should, and that could be done faster or more consistently with a tool available today?
That is the opportunity. Not copying what the most visible startup is doing. Implementation inside your actual business, with your team, your customer expectations, your compliance constraints, and your existing systems.
Why implementation is the real skill
The tools are increasingly available to everyone. The models are capable. The APIs are accessible. What remains scarce is the ability to take a capable tool and make it work inside a real operation.
That implementation skill is not only technical. It is part systems design, part change management, part process mapping, and part judgment. You need to understand how a team actually works, what breaks down under pressure, where data moves, and what kind of review the business needs before an output can be trusted.
The pressure is usually speed: faster support, faster sales follow-up, faster reporting, faster product delivery. Speed matters, but the implementation also has to be especially clear about customer data, access, retention, and human oversight.
The businesses that develop this implementation capability build an advantage that is harder to copy than a prompt. You can copy a workflow template. You cannot easily copy the organizational knowledge required to make it stick.
Where to start
If you are a founder, operator, or team lead trying to figure out where AI fits, start with three steps before buying another tool.
First, spend a week paying attention to repetitive work. Not the big strategic decisions: the small daily tasks that people do without thinking. Data entry, drafting standard replies, categorizing inbound requests, preparing weekly reports, cleaning CRM records, reviewing documents, or moving information between systems.
Second, pick the task that creates the most friction. Not the most interesting one. The one that causes backlog, slows the next team down, or forces a skilled person to do low-leverage work.
Third, ask whether the task is specific enough to hand to a tool. If you can write down the inputs, the desired output, the review criteria, and a few examples of good work, you probably have something workable. If you cannot write that down, you have a process clarity problem that needs to be fixed first.
The practical test
A good first AI use case should pass a simple test:
- The task happens often enough to matter.
- The expected output can be reviewed quickly.
- A mistake is recoverable.
- The data boundary is understood.
- The improvement can be measured in a business metric.
That last point is important. “We used AI this month” is not a result. “Support first-response time dropped,” “sales follow-up became more consistent,” “weekly reporting takes two hours instead of eight,” or “document review no longer blocks the team” is a result.
The opportunity is real. It is just not in being first to adopt whatever tool is trending. It is in becoming serious about making AI useful inside the businesses that already exist.