Pricing guide

What does an AI agent cost in the US in 2026?

In the US in 2026, a working AI agent for one business workflow is quoted anywhere from about $15,000 for a pilot-sized build to $60,000–$200,000 for a production agent wired into your systems, at domestic agency bill rates reported at $125–$240 an hour. The same scope built by a studio working from India at fixed fees runs $4,500 for the pilot and $8,000–$42,000 for one workflow in production, which are our published bands. Running costs are the same wherever it is built: usually $30–$600 a month in model usage and hosting at small-business volumes. Quotes under $5,000 are a chatbot or a prompt wrapper, not an agent that acts in your systems.

What drives the price up and down, what you pay every month after launch, the costs that are not in the quote, what changes when the builder is in another time zone, and how to pay less without getting less. Our bands are published; the US market figures are derived from reported agency bill rates and agencies’ own published ranges, and are marked as such.

Side by side

What you are buyingTypical US agency quote (USD, reported)Our fixed-fee band (USD)Timeline
A chatbot or prompt wrapper on a website$2,000–$15,000We do not sell this on its own1–2 weeks
Fixed-price pilot: one workflow, two systems, approval step$15,000–$40,000 (100–160 hours at reported bill rates)$4,5002–3 weeks
One workflow in production, integrated$60,000–$200,000$8,000–$42,0004–8 weeks
Agentic workflow across several systems$150,000–$500,000$15,000–$60,0006–12 weeks
Agent platform or AI product feature$250,000 upwards$60,000–$150,0003–6 months
The US column is derived from agency bill rates reported at $125–$240 an hour in 2026 and from ranges US agencies publish for single-agent and multi-agent builds. They are observations from parties with an interest, not a survey. Our bands are fixed fees against a scoped brief, billed in USD.

Which to pick, and when

Under $5,000

A chatbot, a prompt wrapper or a Zapier flow with an AI step. Useful things, sometimes. Not an agent that reads, decides and writes to your systems with an approval step.

  • You need FAQs answered or a form routed
  • There is nothing to book, change or reconcile

$4,500 to $40,000

The pilot band, depending on who builds it. One workflow, two systems, real data, a measured result and a handover. The cheapest honest way to find out whether an agent is worth building properly.

  • You have one workflow that clearly costs hours
  • You want proof before a bigger spend

$8,000 to $200,000

Production builds. The spread is mostly who builds it and where; the rest is integrations, autonomy and data readiness, below. Most small-business agents land in the lower half of whichever column you buy from.

  • The workflow runs daily at volume
  • Several systems, a console, monitoring and a retainer are needed

What moves the price

  • Number of systems the agent reads and writes. Each integration is work, and systems without APIs (older practice software, legacy accounting packages, PDFs) cost more than those with.
  • Autonomy. An agent that drafts for a person is cheaper than one that acts with approval, which is cheaper than one that acts alone within rules, because each step up needs more guardrails, logging and testing.
  • Data readiness. Clean records in an API-accessible system are quick. Scans, photos, six bank formats and a spreadsheet maintained by whoever edited it last are not.
  • Channels. Web chat is simple; WhatsApp adds template approval and opt-in handling; SMS adds carrier registration; voice adds a great deal.
  • Volume. Thousands of runs a day need rate handling and caching; a handful a day does not.
  • The console. Every production agent needs a place for a person to see what it did and correct it; how much of one depends on who uses it.
  • Who builds it, and where. The same scope at a domestic agency bill rate and at an Indian studio’s fixed fee is the widest gap in the table, wider than any of the technical drivers.

What you pay every month after launch

Model usage is the main running cost and it scales with how many runs the agent does and how much context each run carries. Hosting on Supabase and Vercel is tens of dollars a month. WhatsApp templates are billed by Meta per message; SMS by the carrier. A retainer for changes, model upgrades and monitoring is optional; the tiers are on the retainer page.

VolumeModel usage and hosting (USD / month)
Under 500 runs a month$30–$120
500–5,000 runs a month$120–$600
5,000+ runs a month$600–$2,500
Billed to your own accounts so you see every dollar. Messaging carrier fees are extra. These are the same bands the cost estimator uses.

The costs that are not in the quote

  • Messaging channel setup. In the US the first channel is often SMS or email rather than WhatsApp: A2P 10DLC registration for business texting takes days and has its own carrier fees; WhatsApp needs business verification and template approval either way.
  • Your team’s time: someone has to own the rulebook and spend an hour or two a week on the log. Builds that skip this fail quietly.
  • Data access. If reaching your records needs a six-week IT ticket, that is six weeks before the agent can do anything.
  • Compliance review where it applies: HIPAA for anything touching patient data, TCPA consent for outbound texting, state privacy laws for consumer data. Scope it in week one, not after go-live.
  • Rework caused by an unscoped brief. The cheapest fix is a written scope in week one, which is what a pilot buys.

What changes when the builder is in India

The price, mostly. The model costs the same, the hosting costs the same, and the hours your own team puts in are the same; what differs is the rate the build is billed at, which is why the two columns in the table are so far apart for the same scope. The things to check are the ones you would check with any vendor, plus three: the overlap in working hours (ours is your morning on the East Coast and early morning on the West; the scope call and the weekly review sit there), where the data lives (your own cloud and model accounts, in your region; nothing of ours in the loop), and the contract (IP assigned, fixed fee, a written handover, thirty days of support).

Payment is by card or bank transfer in USD. There is no sales tax or GST on the invoice from us; whether your state taxes custom software services is a question for your accountant.

How to pay less without getting less

  • Start with one workflow, not three. The second is cheaper once the first is running.
  • Keep the agent drafting for a person until the log earns it more autonomy. Guardrails cost less than audits.
  • Put a shared sheet between the agent and a system with no API for the pilot. Integrate after you know it is worth it.
  • Use the channel you already have. The inbox or the number customers already use beats a new app.
  • Fix the definitions before the automation. An agent on top of disagreeing data produces wrong answers faster.
  • Ask for a fixed fee against a written scope, and for the code, prompts and accounts to be yours. Then the second vendor, if there ever is one, starts from something.
  • Buy the hours where they are cheapest and keep the things that cannot be outsourced, the rulebook and the weekly review, in your own building.

Decide in 4 questions

Answer them in order. Most people are done by the third.

  1. 1

    Is there something to book, change, send or reconcile, not just answer?

    Yes: You are pricing an agent. Continue.

    No: A chatbot or a Zap. Under $5,000.

  2. 2

    Can you name the one workflow that costs the most hours?

    Yes: Price a pilot: $4,500 with us, $15,000–$40,000 domestically. Continue to see where production lands.

    No: Run the readiness quiz first; it ranks your candidates.

  3. 3

    Two systems or fewer, with APIs or exports?

    Yes: Production is likely the lower half of $8,000–$42,000.

    No: Expect the upper half, or the agentic band if the agent must plan across systems.

  4. 4

    Does the agent act, or draft for a person?

    Yes: Acting adds guardrails, logging and a console: budget the agentic band.

    No: Drafting keeps you in the single-workflow band.

FAQ

Common questions

Why do US quotes for "an AI agent" range from $3,000 to $300,000?

Because the phrase covers everything from a prompt on a website to a system that reads documents, decides and writes to five systems with an approval queue, and because a domestic agency bills at $125–$240 an hour while a studio in India bills at a fraction of that. Ask what the agent is allowed to do, which systems it touches, what happens when it is wrong, and who is doing the hours.

Is it safe to have an Indian studio build an agent on US customer data?

The data never needs to leave your accounts: the agent runs in your cloud and your model account, in your region, and we work inside them. The contract assigns IP and the handover is written so nothing depends on us. For HIPAA-covered data, a business associate agreement with us and with the model vendor is part of scoping; say so on the call.

How much does the AI model itself cost per month?

At small-business volumes, $30–$600 a month, billed to your own account so you see it. A conversation of a dozen turns with a few tool calls costs a few cents; a document classification costs less. It is rarely the number that decides the project.

Do you charge in USD?

Yes. Clients billed in USD pay $4,500 for the pilot and the bands on this page for builds, by card or bank transfer, with no sales tax or GST on the invoice. The bands are what we quote, not a conversion.

What is the cheapest way to find out if an agent would work for us?

The readiness quiz and the cost estimator on this site are free and take five minutes between them. After that, a scope call is free, and the pilot is the cheapest honest proof: one workflow, two to three weeks, a fixed fee, a measured result.

How much does an AI agent cost in the US?

From a US agency, a pilot-sized build is quoted at roughly $15,000–$40,000 and a production agent at $60,000–$200,000, derived from reported bill rates of $125–$240 an hour. From a studio working from India at fixed fees, $4,500 for the pilot and $8,000–$42,000 for one workflow in production. Running costs are $30–$600 a month at small-business volume wherever it is built.

How much do AI agencies charge per hour in the US?

Reported 2026 agency bill rates for mid-to-senior AI engineers run $125–$240 an hour, and senior contractors $130–$200; consultant day rates are reported from $1,000 to several thousand. These come from staffing vendors and talent marketplaces with an interest in the numbers, so treat them as ranges and get a fixed quote against a written scope.

Is it cheaper to outsource AI development to India?

For the same scope, substantially: the gap between the two columns in the table above is wider than any technical price driver. The model, the hosting and your own team’s hours cost the same. What to check is the same as with any vendor, plus working-hour overlap, where the data lives (your accounts, your region) and a contract that assigns IP and includes a written handover.