FAQ
Frequently asked questions
Straight answers on what autonomous agents are, what HorizonStack agents can own, how compliance and escalation work, and how outcome-based pricing is structured.
What is an autonomous AI agent?
- An autonomous AI agent is a software system that pursues a goal across many steps and channels without a person driving each action. It reads the current state of a job, decides the next move, acts, and observes the result — repeating until the outcome is reached or it escalates to a human. HorizonStack agents own regulated, multi-step work end to end.
How is HorizonStack different from a chatbot?
- A chatbot answers a question inside a single session and forgets it when the session ends. A HorizonStack agent owns a job that runs for days or weeks, across voice, SMS, email, and your systems, with durable memory of everything that happened. It is accountable for an outcome, not for a conversation.
How long can a single job run?
- A single job can run for weeks. Our longest deployment pattern runs about thirty-four days end to end — an employee onboarding that spans background checks, equipment, payroll, and approvals. The agent holds durable memory of the whole job across every gap, so elapsed time does not cause it to lose the thread.
What channels can agents work across?
- Agents work across voice, SMS, email, web portals, and your internal APIs within a single job. The agent chooses the channel each step needs — a phone call to a payer, an email to a broker, a write to a system of record — and carries the same memory of the job across all of them.
Can agents make and receive phone calls?
- Yes. HorizonStack agents place and take phone calls as part of their work — navigating phone trees, waiting on hold, and speaking with a person on the other end. Every call is transcribed and fed back into the job as the record of what happened, so the outcome is auditable.
How does the agent decide what to do next?
- A policy engine decides the next action from the accumulated state of the job. Judgment is separated from execution: the engine chooses the move using rules you can inspect, and the same facts always produce the same decision. That determinism is what makes the agent auditable in regulated work.
What happens when an agent can’t resolve something?
- It escalates to a person, with the full history attached, so the human resumes the work rather than restarting it. Escalation triggers are explicit and configured per workflow — for example, a denial that turns on medical necessity, or a credit above the agent’s approval threshold. A stalled job is never left silently unmoved.
How do you prevent an agent from doing something wrong?
- Three controls. Decisions run through an inspectable policy engine rather than free-form model output. High-stakes actions — binding coverage, granting privileged access, posting large credits — require human approval. And every action is logged with its trigger and reasoning, so a compliance team can replay exactly why the agent did what it did.
Is call recording compliant?
- Calls are recorded and transcribed to create the auditable record regulated work requires, and recording is configured to honor the consent and disclosure rules that apply in each jurisdiction and channel. Recordings and transcripts are retained under the same data controls as the rest of a customer’s job data.
What systems does HorizonStack integrate with?
- Agents work across the systems each workflow touches — agency management systems, billing and revenue-cycle platforms, EHRs, HRIS and payroll, ITSM tools, identity providers, CRMs, and carrier or vendor portals — plus voice, SMS, and email. Where a system exposes an API the agent uses it; where it does not, the agent works the portal or the phone.
How is HorizonStack priced?
- HorizonStack is priced on outcomes, not on usage. You are billed for jobs that close — a bound renewal, a reprocessed claim, a resolved dispute — rather than for minutes, messages, or model calls. If a job does not reach its outcome, there is nothing to invoice for it.
What does outcome-based pricing mean in practice?
- It means the unit you pay for is a completed job, defined up front for each workflow. A renewal is billable when coverage is bound; a denial when the claim reprocesses. Activity that never reaches the outcome is our cost, not yours, which keeps our incentives pointed at closing the work.
How long does a deployment take?
- A deployment starts with one workflow and a clear definition of what closing it means. Timelines depend on how many systems the agent must reach and what access it needs, but the model is deliberately narrow at first — one pattern working end to end — before expanding to adjacent workflows.
Who is HorizonStack for?
- HorizonStack is for teams that own regulated, multi-step work that drags because it requires many touches over days or weeks — insurance renewals, claim appeals, IT access requests, employee onboarding, billing disputes. It is built for operations leaders who want an outcome owned, not another tool their team has to operate.
What industries do you work in?
- Our live deployment patterns span commercial insurance, healthcare revenue cycle, enterprise IT, people operations, and customer operations. The common thread is not the industry but the shape of the work: long-running, multi-channel jobs with compliance requirements and a clear real-world outcome that defines done.
How is customer data handled and retained?
- Customer data — including call recordings, transcripts, and job records — is scoped to the customer it belongs to and retained under configurable retention controls. Access is limited and auditable, and the same controls apply across every channel an agent touches. Specific retention windows and data-handling terms are set per deployment.
Have a question about your workflow specifically?
Book a demo