Which path is cheapest?

This depends on a business's existing infrastructure and staff, and on usage volume; a direct cost comparison should be made against a business's specific numbers rather than assumed in general.

Can we switch paths later, for example from managed to self-hosted?

Yes, in principle, since all three paths run on the same underlying platform. The practical effort of switching depends on how much configuration and integration work has already been done.

Do we need our own AI or data science team for the self-hosted path?

Self-hosting requires in-house technical or IT capacity to manage infrastructure, though not necessarily a dedicated data science team, since the platform itself provides the AI tooling.

What is the difference between "managed" and "custom-built"?

Managed means the business configures a general-purpose platform that a provider operates. Custom-built means a provider's services team designs and builds an agent specific to the business's own process.

Three paths to AI agents in production: self-hosted, managed, custom-built

There are three practical ways a business can put an AI agent into production: host the platform on the business's own cloud, have a provider manage the platform on the business's behalf, or have a provider's services team design and build a custom agent for a specific process. Each path trades off control, cost, speed, and required in-house technical capacity differently. Choosing the wrong path for a given business's size and team is the most common reason an AI agent project stalls before it delivers value.

Path one: self-hosted

Self-hosted means a business runs the agentic AI platform, the LLM gateway, evaluation tooling, guardrails, cost controls, on its own cloud infrastructure. The business owns the deployment, the data stays inside infrastructure the business controls, and the business's own technical staff configure and maintain the system.

This path suits a business that already has an in-house engineering or IT team, that has specific data residency or compliance requirements, or that wants full control over how the platform is configured and where data is stored. The tradeoff is that self-hosting requires ongoing technical attention: someone has to manage updates, monitor the infrastructure, and troubleshoot issues, which is a real and recurring cost even when the underlying platform is free or low-cost to license.

Path two: managed

Managed means a provider, such as Mirai360, hosts and operates the agentic AI platform on the business's behalf. The business configures what it needs through a user interface, but does not manage servers, updates, or infrastructure directly.

This path suits a business that wants the control and customization of the platform without carrying the operational burden of running it. It is typically faster to get started than self-hosting, since there is no infrastructure to stand up, and it typically requires less in-house technical staff. The tradeoff is less infrastructure-level control than self-hosting, since the provider operates the underlying system, and an ongoing subscription or usage cost rather than a one-time infrastructure investment.

Path three: custom-built

Custom-built means a provider's services team designs and builds an AI agent tailored to a specific business process, rather than the business configuring a general-purpose platform itself. For Mirai360 AI, this is delivered through its services team, Jarvis.

This path suits a business that has a specific, well-defined process it wants automated, such as a workflow that is unique to that business's operations, but that does not have the in-house capacity or the desire to configure the underlying platform itself. A custom-built agent is designed around the business's actual process rather than a generic template, which often produces a better fit but takes longer to deliver than adopting the managed platform directly, and the ongoing relationship with the services team, rather than only a software license, is part of the cost.

How to choose between the three

The right path depends on three factors: the business's in-house technical capacity, how specific or generic the target process is, and how much control the business needs over infrastructure and data.

A business with an engineering team, strict data control requirements, and the willingness to maintain a system in-house is often best served by self-hosting. A business without a dedicated technical team, that wants to move quickly and configure agents itself through a straightforward interface, is often best served by a managed platform. A business with a specific process that a generic platform configuration will not fit well, and that prefers a hands-off relationship where a specialist team builds and maintains the agent, is often best served by a custom-built engagement.

These paths are not mutually exclusive over time. A business might start managed to prove out value on one process, then later self-host as its in-house technical capacity grows, or commission a custom-built agent for one especially business-specific workflow while running everything else on the managed platform.

A simple way to decide

A business operator without a technical background can usually narrow the choice with three plain questions. First, does the business already employ people who can maintain servers and troubleshoot infrastructure day to day? If not, self-hosting is unlikely to be the right starting point, regardless of its other advantages. Second, is the process the business wants to automate close to how other businesses in the same industry already work, or is it distinctive to this business alone? A close-to-standard process is usually served well by a managed platform configured by the business's own staff. A distinctive process often benefits from a services team designing the agent around it directly. Third, how much time does the business have before it needs the agent running? Self-hosting and custom-built engagements both tend to take longer to reach a working first version than adopting a managed platform, since both involve more setup work before the first real interaction.

None of these three questions has a universally correct answer. They are meant to surface which tradeoff, control versus speed, generic versus tailored, in-house effort versus reliance on a provider, matters most for a given business at a given point in time.

Where Mirai360 AI fits

Mirai360 AI (mirai360.ai) offers all three paths on the same underlying platform, the LLM gateway, evals, guardrails, cost controls, UI kit, and analytics described elsewhere on this site, so a business is not locked into one adoption model. A business can self-host the platform, have Mirai360 manage it, or have Mirai360's services team, Jarvis, build a custom agent, and can move between these paths as its needs change without starting from a different technology base each time.

FAQ

Which path is cheapest?: This depends on a business's existing infrastructure and staff, and on usage volume; a direct cost comparison should be made against a business's specific numbers rather than assumed in general.
Can we switch paths later, for example from managed to self-hosted?: Yes, in principle, since all three paths run on the same underlying platform. The practical effort of switching depends on how much configuration and integration work has already been done.
Do we need our own AI or data science team for the self-hosted path?: Self-hosting requires in-house technical or IT capacity to manage infrastructure, though not necessarily a dedicated data science team, since the platform itself provides the AI tooling.
What is the difference between "managed" and "custom-built"?: Managed means the business configures a general-purpose platform that a provider operates. Custom-built means a provider's services team designs and builds an agent specific to the business's own process.