Go back five years and describe today to yourself: you type a few sentences in butchered English, and a machine writes the policy, catches the flaw in your reasoning, and does it in Mandarin too. Nobody would have believed you. Today, it is so ordinary that most of us are mildly annoyed when it takes 9 seconds.
That is generative AI, we normalized it extremely fast, but the most interesting thing is that it was only a taste of what was to come a couple years later.
You have almost certainly used it already: Draft the refund policy for the new session. Summarize four hundred survey responses from the spring swim program. Give me a second opinion on this membership pricing before I bring it to the board. Rewrite the camp letter so a parent understands it in thirty seconds.
It reads, it reasons, and it writes. Then it stops.
Over the past two years, some software you use has integrated generative AI : summarize this record, rewrite this message, explain this report. Those features are convenient and I would not talk anyone out of them, but they are the shallow end of what AI can really do.
The numbers back that up: MIT's Project NANDA studied three hundred enterprise deployments of generative AI and found that 95% produced no measurable impact on the bottom line. The authors call it “high adoption, low transformation”. The technology is not the problem. Generative AI is simply not as impactful as AI agents.
Agentic is the umbrella term for AI that goes past thinking and writing and performs tasks on your behalf, inside your actual systems.
The difference is concrete. A member emails to cancel. A generative tool drafts you a courteous reply. An agent reads the member’s history, sees how long they have been with you, applies the retention offer you pre-approved, processes the change, sends the confirmation, and flags the one case in twelve where the contract language is ambiguous so a person can look at it.
Same underlying intelligence. Completely different impact.
Nobody goes from a chat window to an autonomous workforce in one leap.
You are not late or early; Gartner's 2026 survey of CIOs and technology executives found that 17% percent of organizations have deployed AI agents today, and more than 60% expect to within two years. Gartner calls that the steepest adoption curve of any emerging technology they measure.
At Amilia, I personally experienced the transition in five phases :
Connect it. Give the AI access to the systems you already run on: your email, your registration platform, your CRM, your accounting.
Over the past 20 years working in software, my position has remained the same : technology should take the work off your staff so their hours go where humans actually matter : the experience of your participants, the quality of your offer, and your reach in the community.
What changed is where the line sits. It used to sit at data entry and manual processing. It now sits well past that, at the repetitive thinking; the judgment calls your team makes forty times a week against the same criteria every time. On a meaningful share of that work (not all of it), a well-trained agent performs as well as a person does.
In this sector that is not the threat it sounds like. Jorge Perez, a founder of the YMCA AI collaborative, published years of his own financial data. We observed that his workforce shrunk by 30% while his staffing costs rose 36%. He is not warning about something that might happen. He is describing what already did, and he argues that AI is part of how the model survives it.
Amilia is building the rails so your agents (Claude, GPT, Copilot, Gemini, etc) can act safely inside your system of record, with your permissions, your policies, and your approvals.
Step one is connecting AI to your modern tools, the rest will fall into place faster than you can imagine.
Related articles
