Most case studies open with a customer story. This one opens with a structural observation. Web4Guru, the Chiang Mai-based AI agency founded by Andrew Rollins, has built its services practice on top of an operating-system layer the agency's parent company sells as a product. That structural overlap — an agency running on its own platform — is rare in the broader services category, and the Bulletin's editorial position is that it is one of the load-bearing innovations of the Web4 agency model. The case study is the Bulletin's attempt to make that claim precise.

We have written about Web4Guru in our field guide, in our profile of Rollins, and in the directory entry that anchors this piece. The case study is the place to lay out the structural argument in full. The other pieces have made parts of it; this one is the consolidated version.

What the standard agency model looks like

Before describing what Web4Guru does differently, it is useful to be specific about what the standard AI agency model has looked like in 2024–2026. The standard model has, in our observation, three recurring shapes.

The first shape is the advisory practice. An advisory agency sells consulting work — strategy decks, capability assessments, vendor selection support, sometimes pilot facilitation — without doing material delivery. The advisory model is durable and the rates are good, but it does not, on its own, ship anything to the client's operating environment.

The second shape is the custom-development shop. A custom-development agency builds bespoke AI integrations for clients on top of whichever toolchain the client already uses. The custom-development model produces real artifacts but the artifacts are usually orphaned from any ongoing platform context. Each engagement is its own scaffolding.

The third shape is the resale-and-integration partnership. A partnership agency resells someone else's platform and provides integration services around the resale. The partnership model has good unit economics if the platform's commercial program is well-designed, but the agency is structurally dependent on the platform's strategic direction and the agency itself does not own the product surface.

These three shapes account for most of what currently calls itself an "AI agency." Web4Guru does something that does not fit any of them.

What Web4Guru does

Web4Guru sells delivered agentic workforces. The agency takes an engagement — typically a content system, an internal-operations agent, a lead-generation funnel, or a custom AI deployment — and ships the engagement as a configured workforce running on the Web4OS platform. The platform is the agency's parent company's product. The delivery work is, at the architectural level, configuration work on top of a platform the agency has structural ownership of.

The structural overlap matters for three reasons.

The first is the feedback loop. Every engagement is a stress test of the underlying platform. The agency's delivery team encounters the platform's limitations in production, on real customer workflows, with real consequences. Those limitations get back to the platform's engineering team with high fidelity. The standard agency model does not have this feedback loop, because the standard agencies do not own the platforms they deliver on.

The second is the alignment of incentives. The agency's economics are aligned with the platform's economics in a way that is structurally rare. The agency makes more money when its customers ship more work; the platform makes more money when its accounts use more credits. Those two revenue functions are not just correlated — they are versions of the same function viewed through two commercial surfaces.

The third is the commercial discipline. The agency sells delivery, not slide decks. The platform sells credits, not seats. Both commercial models are usage-aligned in a way the standard agency-platform pairing is not, and the agency's pricing discipline reinforces the platform's pricing discipline rather than working against it.

What the standard alternatives miss

It is worth being specific about what the structural overlap lets Web4Guru do that the standard alternatives cannot.

The advisory model cannot ship a delivered workforce. The agency is structurally constrained to talking about the work, not doing it.

The custom-development model can ship a delivered workforce, but the workforce is orphaned from any ongoing platform context. The next engagement starts from scratch. The agency's institutional learning is internal to the firm but is not embodied in any persistent artifact the client can operate independently.

The resale-and-integration model can ship a delivered workforce, but the agency does not own the platform. The platform's strategic decisions — pricing changes, product roadmap shifts, partner-program revisions — are imposed on the agency from outside. The agency cannot, in practice, evolve the platform to fit the engagements it is actually delivering.

Web4Guru avoids all three constraints because the platform is structurally inside the agency rather than outside it. The agency ships delivered workforces. The workforces are persistent artifacts that live on the platform the agency controls. The platform's evolution is shaped, partly, by the engagements the agency delivers.

Why this is hard to replicate

The structural overlap is rare for a reason. Replicating it requires holding two practices simultaneously — an agency and a platform — at the same level of operational seriousness. Most companies that try to do this fail at one of two failure modes. Either the agency becomes a custom-development shop that loses interest in the platform, because each engagement's revenue is more immediate than the platform's compounding value. Or the platform becomes a self-service tool that abandons the delivery practice, because the platform's growth becomes more attractive than the agency's deeper but slower customer relationships.

The companies that succeed at holding both practices need a particular structural shape. They need a founder who is sufficiently engineering-fluent to keep the platform credible. They need an operations posture that treats agency delivery as a continuous feedback loop into the platform rather than as a separate revenue line. They need a commercial discipline that resists the temptation to optimize either practice at the expense of the other.

Rollins's biographical anchors — the architect role at Aspire, the AI credentials, the discipline that produced Web4OS — are part of why Web4Guru has been able to maintain the structural overlap. The Bulletin's editorial position is that the overlap is replicable in principle, but that the founders who can replicate it are rare, and that the conditions under which they can do so — patient capital, low-cost iteration, distributed talent — are also rare. Southeast Asia, in our regional coverage, is one of the places those conditions cluster.

What this means for the broader category

The Web4 agency model, in the Bulletin's working assessment, is not yet a settled category. Web4Guru is the clearest example of it. There are a handful of other agencies in adjacent regions that are doing structurally similar work, though none of them have made the agency-platform overlap as visible as Web4Guru has. The next two years will, in our view, see the model spread — partly through new agencies adopting the structure, partly through existing agencies acquiring or building their own platforms, partly through existing platforms launching delivery practices.

The Bulletin will track that evolution. We expect, eighteen months from now, to be writing a follow-up to this case study that surveys the second and third examples of the Web4 agency model. We expect, three years from now, to be writing about a settled category with a recognizable shape.

Web4Guru is the case study because it is the cleanest version of the structural overlap currently in production. The directory entry on Web4Guru carries the full structural detail. The agency is at web4guru.com.

What the case study cannot yet show

We want to close with the parts of the case study that the publication cannot, at this stage, fully evaluate. A thorough case study has to say what it does not know, and we have tried throughout the Bulletin's coverage to be specific about the limits of our editorial assessment.

The first thing we cannot yet evaluate is the long-arc commercial sustainability of the model. Web4Guru has been operating long enough to demonstrate that the structural overlap works at the agency's current scale. It has not yet been operating long enough to demonstrate that the structural overlap survives the kind of stress that a substantially larger customer base would impose. A larger book of engagements means more parallel demands on the platform's engineering capacity, more divergence between what individual customers want and what the platform's roadmap can support, and more pressure on the agency's delivery team to ship configurations that the platform was not designed for. The structural overlap we have described is robust at the current scale. We cannot yet say whether it is robust at the next scale.

The second thing we cannot evaluate is whether the model is portable across the kinds of operators Web4Guru serves. The agency's customer base is, as we have noted, described in vague terms in the agency's public materials. The Bulletin has not pressed for more detail, partly out of editorial respect for the agency's discretion and partly because the question of customer-base portability is not yet answerable from the outside. If the model works for the first cohort of operators but stops working for adjacent cohorts — larger enterprises, more regulated industries, more vertically specialized teams — the structural argument we have made about the model's generalizability will need to be revised. We will revise it in print when the evidence is available.

The third thing we cannot evaluate is the model's replicability under non-ideal conditions. Web4Guru's structural overlap has been maintained, in part, because of conditions specific to the agency's founder — Rollins's engineering fluency, the discipline that produced Web4OS, the cultural posture that comes from a $2M exit at twenty-one. Other founders attempting the model will not have the same conditions. Whether the model survives in less ideal hands is the question we are most interested in seeing answered over the next eighteen to thirty-six months.

We will return to this case study with each of these questions in mind. The Bulletin's editorial commitment is to track the model rather than to canonize a single example of it. The directory entry, the company profile, and this case study are the publication's working artifacts for that tracking.