What shapes an AI project in the United States
American AI regulation is a moving patchwork. The safest way through it is to start from what the agent does and to whom, since most of the law attaches to the action and not to the technology.
| Topic | What it means for your agent | Decide early |
|---|---|---|
| Federal position | No general federal AI statute was in force when this page was reviewed. The Federal Trade Commission applies its rules on unfair and deceptive practices to AI products and to the claims made about them | What you will say the agent can do, and the proof that it does |
| State laws | States are legislating on their own. Colorado replaced its first AI act in 2026 with a narrower law on automated decisions due to start in 2027, California has adopted rules on automated decision-making technology, and New York City requires bias audits of automated hiring tools | Which states your customers, applicants or employees are in |
| Bot disclosure | Some states require a bot to say what it is in certain settings, such as sales conversations, and more have proposed similar rules | The agent's opening line, and when it offers a person |
| Calls and texts | Under the Telephone Consumer Protection Act, automated calls and texts need prior consent, and the communications regulator has ruled that AI-generated voices count as artificial voices. Several states also require every party's consent before a call is recorded | How consent is captured, and what the agent announces on a call |
| Regulated decisions | Hiring, lending, housing and insurance decisions carry anti-discrimination and notice duties whether a person or a model makes them. A lender must still be able to state the reasons for a refusal | Where the agent recommends and a person decides |
| Health and other sensitive data | Protected health information may go only to vendors that have signed a business associate agreement, and a model provider is such a vendor | Which provider terms cover your data, and what is masked before it is sent |
Agent workflows we build for American businesses
These four are where most companies start. The AI agent solutions catalogue covers further cases by department and industry.
Support agent with human handoff
For service teams that answer the same questions all day.
- Customer writes by chat or email
- Agent says it is automated
- Answer drawn from your help content
- Account action taken if permitted
- Handoff to staff with a summary
- Outcome logged
Lead qualification
For sales teams that lose the inquiries arriving after hours.
- Inquiry comes in from the website
- Agent asks qualifying questions
- CRM record created
- Meeting booked with the right rep
- Summary sent ahead of the call
Voice agent for scheduling
For practices and service businesses whose phones ring more often than staff can answer.
- Caller hears that the assistant is automated
- Recording notice given where required
- Appointment found, booked or moved
- Confirmation texted with consent
- Complex calls passed to staff
Accounts payable intake
For finance teams keying supplier invoices into an ERP.
- Invoice arrives by email
- Fields extracted
- Matched to the purchase order
- Mismatches queued for review
- Approved invoice posted
What we build
Conversational agents
Chat, email and voice agents grounded in your content and connected to the systems they need. The voice agent page shows one in detail.
Workflow automation
Agents combined with conventional automation, so predictable steps run on rules and only judgment calls go to a model. See business process automation.
Guardrails and evaluation
Test sets built from your real cases and measured before launch and after every change, with limits on what the agent may do without approval.
Records for compliance
Consent, disclosures, prompts, outputs and approvals stored so that a decision can be explained later to a customer, an auditor or a regulator.
Private knowledge
Retrieval from your documents with access rules respected. Our comparison of RAG and fine-tuning explains when each fits.
Integration
Connections to CRM, help desk, telephony, ERP and calendar systems through their APIs.
What we need from you
- A workflow with clear inputs and a measurable result
- Sample conversations, documents or tickets with identifying details removed
- Your counsel's view on consent, disclosure and any regulated decision
- Access to the systems the agent will read and update
- A subject expert to review outputs during the pilot
Ownership and licensing terms are defined in the project agreement.
A readiness checklist for an AI pilot in the United States
Before the pilot starts
0 of 8 checkedUnsure which process to pick? Our guide on identifying business workflows suitable for AI gives a scoring method.
What to do next
Tell us which task you want off your team's desk, how often it happens and what a mistake would cost. We will reply with the kind of automation that suits it and the shape of a first pilot. Start a conversation or book a 30-minute call.
Common questions
Is there a federal AI law in the United States?
Not a general one, as of this page's last review. Federal agencies apply existing law to AI, including consumer protection, civil rights and financial rules. States have moved on their own with laws on automated decisions, chatbot disclosure and hiring tools, and those laws are still being revised. A project should start from where the affected people live and what the agent decides, then check current law with counsel.
Can an AI voice agent call our customers?
Yes, with the right consent. Federal telephone law treats an AI-generated voice as an artificial voice, so calls generally need the prior consent of the person called, and marketing calls need it in writing. Do-not-call rules apply. Several states also require all parties to agree before a call is recorded. Answering inbound calls raises fewer issues than placing outbound ones, which is where most projects begin.
Do we have to tell people they are talking to an AI?
In some states and settings the law requires it, and the list is growing. Beyond those laws, the Federal Trade Commission treats misleading people about what they are dealing with as a deceptive practice. We build agents that identify themselves at the start, offer a route to a person and do not claim to be human when asked.
Can we use customer data with a commercial AI model?
Usually, if the provider's terms and your own privacy promises allow it. The business terms of the major providers generally state that customer inputs are not used to train their models, and some offer zero retention. Health data requires a business associate agreement with the provider. We also reduce what is sent in the first place, by masking identifiers and retrieving only the records a task needs.
Where to go from here
Related serviceAI AgentsAgents that work inside your systems, with approval where it matters