Building an Agentic Business: Automation by Design, Accountability in Practice

A business should not need someone to carry the same information from one screen to another simply because that is how the process developed. Benefits work contains plenty of those moments: a request arrives, supporting information lives elsewhere, and someone has to connect the pieces before the real question can be addressed.

By an agentic operating model, I mean giving AI a defined objective and authority to choose and carry out steps within an agreed scope. An agent can work through a sequence, respond to what it finds and complete an authorized task without a person prompting every move. That is a different level of capability from generating an answer or a draft.

At SSGI, the capability for fully autonomous execution of defined operational workflows is already in place. I could enable that operating mode today. The decision now is how much authority to delegate in each part of the business.

That decision belongs to leadership. I am interested in the distance between a request and a useful result: how much unnecessary handling we can remove, where judgment changes the outcome, and who is responsible when the situation falls outside the expected path.

SSGI Benefits is an independent, family-owned brokerage serving employers of different sizes. SSGI One is designed to support businesses as they grow—from a team of a few employees to a national organization with 25,000 employees and beyond. It connects benefits administration and employee service in one portal, helping teams coordinate requests, documents and follow-through across locations and divisions. Each enterprise deployment requires configuration, integration review and capacity validation. My ambition is to give those employers coordinated service with direct accountability to the people running the firm.

SSGI One is our signed-in environment for agency staff, employers and employees. It brings together group benefits administration with workflows for individual coverage health reimbursement arrangements, or ICHRAs. Enrollment, new-hire and life-event requests, waivers and allowance workflows sit alongside requests, documents, status and replies.

Inside SSGI One, replies to customer requests and advice currently require human approval. Fixed-template messages may be sent automatically. Our public Ask SSGI assistant and automated telephone receptionist answer automatically, without individual review of each response. They provide general information and reception, not personalized insurance advice.

Those are different operating choices for different kinds of work. A routine informational answer, a fixed notice and advice about someone’s benefits do not carry the same responsibility. An employer evaluating our approach should be able to understand that distinction without having to learn how the underlying software works.

For a national employer, the challenge is often coordination. A new hire at one location, a family life event at another and a waiver elsewhere can reach the same service team while requiring different information and follow-through. HR needs a coherent way to handle that volume without making employees navigate the organization themselves.

Consider a life-event request as an example of how an agentic workflow could be designed. The agent could organize the submitted information, identify missing details, prepare the next operational step and stop when a decision needs human authority. The design should specify which communications can proceed automatically and which questions need review.

The point of that example is completion. An impressive answer has limited value if the request still sits unresolved. A useful workflow has to connect the employee’s need to the person or system authorized to take the next step.

This is how I think employers should evaluate agentic systems: start with an actual piece of work. Identify the delays, duplicated effort and judgment involved. Then decide what can be delegated, what must be reviewed and how the employer will know the work has reached a satisfactory conclusion.

Every delegated workflow should have a clear purpose, a defined scope of information it can use and a point at which it must stop or ask for help. An objective such as “handle benefits” is too broad to tell anyone what authority has actually been granted.

A useful mandate identifies the task and its limits. It can cover an entire routine workflow, including permitted decisions along the way, or a narrower action such as sending an approved fixed message. The extent of the delegation should be intentional and understandable to the people affected.

Our commitment for carrier connections is explicit: staff approval must come before any carrier API transaction that writes or changes information. The technical ability to execute an action does not replace the approvals or agreements governing it. Enabling an autonomous operating mode does not dissolve those obligations.

Before expanding delegation, I want to see how a workflow handles incomplete information, conflicting instructions and exceptions. Speed on the easiest path tells only part of the story. The more consequential the task, the more important it becomes to understand where the system should stop. That is a business judgment, and I intend to remain responsible for it.

Our public website and SSGI One run on an SSGI-controlled Open Cloud engine with U.S.-based replica nodes. A cloud engine is a dedicated group of computers configured to run an application environment together. Using Internet Computer technology, its operator can choose the nodes, providers, locations and number of replicas that make up the subnet. The official cloud-engine documentation (opens in a new tab) explains those choices.

Here, blockchain is programmable infrastructure. Its application units, called canisters (opens in a new tab), combine executable code with persistent state—the information the software remembers between interactions. When an update changes that state, the subnet’s replicas process it and use consensus to commit the result. Replication can help a service continue when individual nodes fail, provided enough healthy nodes remain for the protocol to operate. In everyday terms, the application has both instructions and a continuing memory.

Now imagine an AI agent working within that environment on an authorized administrative task. It could read a structured request, reason through the permitted next steps, update the work record and pause where a person’s approval is required. On a later interaction, the application could supply that saved state so the authorized workflow can continue. That is the combination I find compelling: AI contributes reasoning and action, while persistent software gives the work a continuing structure.

Application permissions and data handling still have to control access to client information. Our email, document storage and external AI remain separate services. Visitors use the website in an ordinary browser, without cryptocurrency or a wallet.

AI can also help design and write the software itself. I see an opportunity to turn operating knowledge into applications: how a request should move, which information belongs together and where approval is required. As the business learns, those applications can evolve. A founder can keep shaping the tools and processes of the business as its ambitions grow.

I find that possibility extraordinary: a smaller organization can design the software, intelligence and workflows around the business it wants to become. The ambition is to put capable agents to work with an application environment the company can shape, with information that persists and authority that remains deliberate. It expands what I can imagine SSGI becoming.

I expect AI to reduce the human labor required for many administrative functions. Some tasks will disappear, and some roles will combine. A serious discussion of agentic businesses has to include that possibility. Greater capacity does not automatically mean an organization will need the same number of people doing the same work.

It also changes how people learn. Research, routine correspondence, document preparation and basic service requests have often been the work through which new employees develop judgment. Employers that automate those tasks need a deliberate way to build expertise through supervised decisions, meaningful responsibility and exposure to difficult cases.

In When Healthcare Gets Cheaper: AI, Employee Benefits and the Future of Work, I connect that shift to a larger argument. Technology could lower the cost of delivering care while also making stable administrative employment less dependable. Healthcare and benefits therefore need to become more portable and less dependent on holding one particular job.

That is why I view benefits strategy and operating strategy as connected. Employers planning for more automation should also consider how people move between roles, how expertise develops and what support remains available when work changes.

What interests me most is the possibility of an owner directing a much broader operation through agents that can execute defined work. Across industries, that could change what a small team can build and how organizations coordinate across locations and time zones. It is an ambitious business model with implications well beyond insurance.

For SSGI, the purpose is concrete: serve complex employer needs with less administrative friction and a clear connection to the people responsible for the business. I want autonomy to carry more of the operational load while judgment is concentrated where it makes a difference.

The standard I want to be judged by is whether the work gets done, whether exceptions receive attention and whether our clients know who stands behind the service. Having autonomous capability gives me choices. Making those choices responsibly is part of leading the company.

For a large or multistate employer considering this approach, the conversation should begin with the benefits work your organization needs to accomplish. Our national employer services and the SSGI One platform are two practical places to start.