An AI or tech services agency has two features that shape its formation choices more than any other agency type: its product is intellectual property, and it touches client data. Both show up in contracts, in tax choices and in what clients ask before signing. Here is how to set one up properly in the US.
LLC or C-Corp?
For a bootstrapped services agency billing clients from day one, an LLC with default pass-through taxation is usually the right start - simple, flexible, no double taxation. The Delaware C-Corp enters the picture when you plan to raise institutional venture capital or build a product on top of the services; some investors may request a Delaware C-Corp with standard documents - confirm their requirements before choosing. Converting later is possible but has tax and paperwork costs, so decide based on your realistic two-year plan, not on what a blog says startups do.
IP assignment is the whole business
Clients buying AI and tech work are buying deliverables they must own: models, code, prompts, pipelines, documentation. Your master services agreement should state that deliverables are assigned to the client on payment, and your contractor and employee agreements must assign their work to the agency first - the chain of title only works if every link assigns forward. Without that chain, you cannot honestly give the client what they paid for, and sophisticated clients now ask to see your subcontractor terms before signing.
Contractors and classification
AI agencies scale with specialist contractors - ML engineers, data annotators, prompt designers. Federal and state tests for employee-vs-contractor status look at control, integration and economic dependence; a full-time contractor working your hours on your equipment is misclassification risk. Use written statements of work, pay for deliverables where possible, and check your state's rules - some states apply stricter ABC tests than the federal standard.
Data terms clients now demand
Enterprise and mid-market clients increasingly send security questionnaires and data-processing addenda before the first invoice. Be ready to answer: where client data is processed and stored, which subprocessors (cloud providers, model APIs) touch it, retention and deletion practices, and whether client data trains any model. Saying "we do not train on client data" and having the contract say it too is becoming table stakes for AI services work. Also note that several US states have comprehensive privacy laws (California's CCPA is the best known) that can apply depending on whose data you handle.
Practical setup steps
- Choose LLC vs C-Corp based on funding plans; form in your home state unless you are raising institutional VC.
- Get your free EIN from the IRS and open a business bank account.
- Sign IP-assignment agreements with every founder, employee and contractor before work starts.
- Prepare a services agreement with deliverable IP transfer on payment, confidentiality, and a data-handling exhibit.
- Write down your answers on data location, subprocessors and model-training policy - clients will ask.
- Price insurance: professional liability / tech E&O is commonly requested for AI and software work.
This article is general educational information, not legal or tax advice. Rules vary by state; confirm them with a licensed professional and the official sources below.
Sources
- IRS: Employer Identification Number
- US Small Business Administration: business structures
- California Attorney General: CCPA
Forming from abroad?
FoundStart's free bilingual guides cover US company formation for international founders - start at the country guides.
FoundStart