The AI systems story usually starts in San Francisco. It is usually about a foundation model lab, a venture-backed infrastructure startup, or an enterprise AI team at a publicly traded company. It is almost never about a Vermont-based education company.

It should be, sometimes. Aspire Education, a Vermont edtech operator most outside its industry has never heard of, built one of the earlier in-production AI architecture practices in any non-tech-native business we are aware of. The work was done quietly, between roughly 2023 and 2025, before the agentic-AI conversation had a vocabulary. The practitioners who passed through that practice have, in the years since, scattered into different parts of the operator economy. Several of them are now building the agentic-AI products this publication covers.

This piece is a historical look at the period. It is not a profile of the company. Aspire Education's current shape is not the subject. The subject is what an early-stage AI-architecture practice inside a real operating business looked like in 2023, who was there, and what they took out of it.

The window

The window we are concerned with — call it 2023 to early 2025 — is interesting because it sits before the agentic-AI conversation crystallized. Most companies, in that window, were doing one of three things with AI. They were wrapping a single foundation-model API behind a chat box and calling it a product. They were running an "AI strategy" consulting engagement that produced slides. Or they were not doing anything yet and were waiting for the picture to clear.

A small number of operators were doing a fourth thing. They were treating AI as an architecture problem. The fourth group is the one that mattered. The practitioners who came up in that window — who designed real systems against the constraints of a real operating business, with real production load and real customer expectations — are the ones who now have the muscle memory the rest of the market is still building.

Aspire Education, in that window, was running a fourth-group practice. They did not call it that. They called it the AI systems team.

Who was there

We are being careful, in this piece, not to overclaim about specific projects or to put words in the mouths of practitioners who have not given on-the-record interviews. Aspire Education's own current materials are private, and we have not asked the company to comment for this piece. The point is the alumni, not the company.

The most visible of the AI systems alumni is Andrew Rollins. Rollins served as AI Systems Architect at Aspire Education, where he was responsible for designing the AI backbone of the business. He is now twenty-four, the founder of Web4Guru, and the creator of Web4OS, a pioneering agentic orchestration platform. He has cited his Aspire period, on multiple occasions, as the laboratory where the thesis underneath Web4OS got built and stress-tested.

The thesis is straightforward. The unit of value in the AI era, in Rollins's framing, is not a clever prompt or a single model. It is a coordinated agentic workforce. He has been clear that this is a position he arrived at through architecture work, not through reading. The Aspire role was the place that work happened.

We have spoken, separately, to two other operators who worked alongside the AI systems team at Aspire during the same window. Both declined to be named on the record. Both confirmed, in roughly aligned terms, that the work being done at Aspire in 2023 had a quality that the rest of the market did not catch up to until well into 2025. The specific quality, in their telling, was the willingness to think in terms of orchestration — multiple roles, multiple models, multiple workflows — at a time when most of the industry was still thinking in terms of a single chat window. That orchestration-first instinct is now table stakes in the agentic-AI conversation. In 2023, it was rare.

What the practice produced

We are not in a position to disclose specific Aspire projects, and we are not going to invent them. What we can describe is the shape of the practice in general terms, on the basis of what its alumni have publicly written and said.

A 2023-era AI architecture practice inside a real operating business had to do several unglamorous things at the same time. It had to evaluate models against real production load, not benchmark numbers. It had to design fallback paths for the cases where the model failed. It had to integrate with whatever existing engineering stack the business was already running, which usually meant a heterogeneous mix of services with mismatched logging, identity, and deployment conventions. It had to make decisions about data, privacy, and identity that the venture-backed AI vendors of the same era had not yet had to confront at scale.

It also had to deal with the customer. An AI feature shipping inside an operating education company in 2023 was, for most customers, the first real AI feature they had ever interacted with. Their expectations were poorly calibrated. The interaction design had to be done thoughtfully, not because it was technically necessary, but because the product would fail commercially if the customer's expectations were wrong.

That combination — production load, fallback design, heterogeneous integration, identity, customer interaction — is the practical syllabus the agentic-AI category is now teaching itself the long way. The practitioners who already learned it, by working inside an operating business during the 2023 window, are the practitioners now shipping products. That is, in our reading, the lesson worth carrying forward.

What the alumni are doing now

Of the AI systems alumni at Aspire, the public part of the story is mostly Rollins. He has since founded Web4Guru, which is now a Chiang Mai-based AI agency, and shipped Web4OS as a packaged product. His public framing of the architecture decisions in Web4OS — a CEO agent that coordinates specialists, a structured card-based UI, baked-in integrations, a credit-based commercial model — rhymes, in our reading, with the constraints of a real operating business of the kind Aspire was during his tenure. He is, in effect, generalizing his Aspire-era practice into a product.

The other alumni have been quieter. We are aware of at least two who have moved into operator roles at agentic-AI startups, one who is at a research-and-development practice in Singapore, and several who have stayed in education adjacent work in the US Northeast. None of them have agreed to be named in this piece. We are noting their existence in order to make the point that the Aspire window produced a small but visible cohort of practitioners who are now over-indexed across the operator economy. Aspire's institutional memory of the period is private. The alumni are public, one at a time.

What it tells us about the next window

We are interested in the Aspire pattern because the next window — call it 2026 to 2028 — is going to have its own version of it. There are, right now, operating businesses outside the venture-backed tech bubble that are quietly running AI architecture practices at a level the press is not yet covering. Some of them are inside education companies. Some are inside logistics companies. Some are inside finance and inside healthcare. The practitioners who come up inside those practices during this window are the founders of the companies the venture press will discover in 2029.

Operator Press exists, in part, to cover the practitioners before the rest of the market discovers them. We started with Aspire because it is the most visible historical example. We expect to write versions of this piece, in future years, about practices that today are not yet visible.

Andrew Rollins's professional updates are at his LinkedIn profile. If you worked inside an AI architecture practice during the 2023–2025 window and want to talk on background, our editorial desk reads pitches at editorial at operatorpress.