Skip to content
For Agencies Talk to Us
Cost guide

How much does an AI agent cost in 2026?

An AI agent pilot on one real task costs $3,000–$31,250 at the direct rate. It is a fixed quote once the scoping sprint (fixed scope, fixed price, with an end date) has set the agent's tools, access limits and steps needing your approval. The sprint and monthly running are priced separately below. Every figure becomes a written proposal.

Request this optionSee the full rate card

Direct rate, in ₹ or $ · Reply within four business hours (typical)

The range

What it costs,
at the direct rate.

A scoping sprint first checks whether an AI agent suits the task at all. Then comes a pilot on one real task. Once the agent has proved it can run alone, it moves to a monthly fee.

  • One-offScoping and evaluation sprint$1,000–$6,250one-time
  • ProjectAgent pilot$3,000–$31,250fixed quote, one-time
  • RetainerProduction run and monitoring$300–$3,750per month

Indicative: every figure is refined into an itemised proposal.INR figures shown excl. GST.

Where the engagement includes it, a Prism workspace holds requests, approvals, tasks and reviews.

What moves the price

Why the price
varies from job to job.

01

How many tools it can call

Every system the agent can reach needs its own access limit. That means a login limited to that step, a rule for what it may read or change, and a test that the rule holds. An agent that looks up a CRM record and drafts a reply needs one. One that checks stock, contacts a supplier and raises a purchase order needs several, each connected and tested. We size the pilot by counting tools.

Biggest factor
02

Whether the task needs an agent at all

An agent is worth building when the task has many steps, several tools and a path that changes with what it finds. Much of what is sold as an agent is one answer from your documents, or one fixed set of steps. Those are cheaper to build as a chatbot or an automated workflow. The sprint checks this first and writes down the cheaper option where there is one. That is the biggest saving on this page.

Often forgotten
03

What it is allowed to change

Reading data is cheap to make safe; changing it is not. An agent that only reads, checks and drafts for a person to send needs limited logins and a log. An agent that sends, spends or changes a live system needs more. It needs an approval step, a limit you set, a way to undo and a record of who approved what. Each such action is priced by the safety checks around it.

Risk
04

The test cases it must pass

Before it runs on its own, the agent runs under watch on cases from the real job. These include traps where the obvious next step is wrong. A tidy task with few ways to fail needs a short test set. A task where a wrong step costs money or a customer needs a long one, scored against how the work is done today. The monthly running repeats the tests whenever a tool changes.

Hidden cost
05

How often the systems underneath change

A live agent stays safe only while its tools behave as they did in testing. The monthly running is sized by how often the agent runs and how often the connected systems change. A stable ERP needs one retainer. A set of online tools that often change in ways that break things needs another. AI model usage sits outside both: the provider bills it, and the proposal shows it passed on at cost.

Monthly cost
06

Where the data has to stay

The AI model is chosen in the sprint on fit, cost and where your data must be stored. Regulated data or a storage rule can mean a model hosted in a set region, or running inside your own cloud. That brings servers to set up and keep updated. The work is a clear line, settled by the sprint's model advice before the pilot is priced.

Scope
The options

Which option
suits your business.

Each option below is one way to buy AI agents. Read what it is and who it suits, then ask for the one that fits. Your proposal itemises it.

One-off

Scoping and evaluation sprint

$1,000–$6,250one-time
What it is

Planning that decides whether you need an agent, then writes the design. It covers the tools, the access for each step, the limits that need approval, the test cases and a model choice. It ends in an itemised pilot scope, or a written case for a cheaper build.

Who it fits

Anyone with a task of many steps that they think could run itself. Also anyone quoted for an agent elsewhere who wants a second opinion. It does not fit a task you already know is one question or one fixed set of steps.

Included
  • The task checked for fit: many steps, several tools and a path that varies
  • The tools, what each one may do and when your approval is needed, agreed in writing
  • Test cases from the real job, including tricky ones
  • Advice on which AI model fits, on cost and on where your data is stored
  • A detailed pilot scope, or a written case for a cheaper option
Not included, quoted separately
  • The pilot build
  • AI model and API usage during testing
Turnaround
1 to 2 weeks
Moves the number
How many systems the agent needs to use. One CRM lookup and six back-office systems are very different sprints.
Request this option →
Project

Agent pilot

$3,000–$31,250fixed quote, one-time
What it is

A working agent on one real task from your business, run under watch. Each tool gets only the access it needs, and anything that can't be undone waits for approval. Its runs are scored against how the work is done today. It is handed over with its full log and runbook, under IP and ownership terms agreed before work starts.

Who it fits

Teams with a task the sprint has confirmed needs an agent, and the systems to connect it to. It does not fit a buyer who wants it to run alone from day one; that has to be earned in testing. Nor does it fit a buyer who wants it to have access to every system; we don't build that.

Included
  • One real task running, with every step recorded
  • Tools connected to your live systems, each with only the access it needs
  • Your approval needed before anything that can't be undone, goes outside your business or passes a set limit
  • Supervised runs scored against your current results; it runs alone only where it has proved itself
  • A full record of actions, a step-by-step guide, admin training and agreed IP and ownership terms
Not included, quoted separately
  • Model and API usage
  • Hosting
  • More tasks or tools, priced once the pilot has proved itself
Turnaround
3 to 6 weeks
Moves the number
Number of tools: for every system the agent can use, we must set its limits, connect it and test it.
Request this option →
Retainer

Production run and monitoring

$300–$3,750per month
What it is

The retainer that keeps a live agent within its limits as its tools and data change. We watch the logs, repeat the tests on every change and adjust the approval limits based on the logs. You get a monthly report against the pilot's starting point.

Who it fits

Any agent that changes live systems or runs many times a day. An agent that only reads, on systems that rarely change, can be run in-house from the runbook.

Included
  • Runs and records monitored, with anything unusual reported to your team
  • Tests re-run whenever a tool, API or data source changes
  • Access and approval limits adjusted based on what the records show
  • Model upgrades tested before they touch the live agent
  • A monthly report against the starting results agreed at the pilot
Not included, quoted separately
  • Model and API usage
  • Hosting
  • New tools or tasks (priced as extensions)
Turnaround
Monthly, 3-month minimum
Moves the number
How often it runs, and how often your systems change. A stable ERP and three online tools that change every quarter are different retainers.
Request this option →
Cheaper or dearer

What makes your quote
lower or higher.

The pilot is quoted once the sprint has written down the tools, approval steps and test cases. So every item below is a line in the proposal, not a later surprise.

Brings the quote down
  • One task with a clear finish line and one system to read from
  • Tools with documented APIs and logins you can limit closely
  • A task that mostly reads: research, checking figures, or drafting for a person to send
  • Results you already measure today, for the tests to score against
  • A team that will run the agent in-house from the runbook
Pushes the quote up
  • Several back-office systems, each with its own access limits to connect and test
  • Actions that involve money: purchase orders, refunds, live-system updates
  • Old systems without APIs, reached through exports or a browser
  • Regulated data, rules on where data is stored, or running in your own cloud
  • So many runs that a rare wrong step stops being rare
What you get

What your fee
pays for.

The fee buys an agent on one task, the limits it works within and proof that it stays within them. We promise no accuracy, hours saved or completion rate. Those are measured against how you work today, after the pilot, not sold before it.

01

A written answer on what you need

The sprint says in writing whether the task needs an agent, a chatbot or an automated workflow. It scopes the cheaper build where there is one.

02

An access plan you approve

Each tool the agent may use, the access it has and the actions that wait for a person. All agreed before anything is built.

03

Approval steps where you set them

Above limits you choose, sending, spending and changing live systems wait for sign-off. The agent suggests and explains; a person decides.

04

Tests against how you work today

Watched runs on real cases, traps included, scored against how the work is done today. It runs on its own only where the scores show it is ready.

05

A full log of every run

What it was asked, what it used, what it decided and who approved it. So a wrong step is easy to see, never hidden.

06

The runbook, the training and the code

How to change the scope, add a tool, read the logs or switch it off. You own the code, the prompts and the data flows.

AgenciesThe ranges above are direct prices. Agency partners buy from the partner price list, about 20% lower on packages.

See the partner price list →
Questions

Frequently
asked.

Because the price depends on things nobody knows until the task is mapped. How many tools must it use? What may it change? Where do the approval steps go? The sprint writes these down and ends in a fixed pilot price. Or it ends in a written case that a chatbot or workflow would do the job for less.

No. Usage is billed by the provider on your own account, in testing and when live. So you can see it, cap it and change models without asking us. The pilot fee buys the engineering: connecting tools, setting access limits and approval steps, testing and handover. The proposal shows usage as its own line, passed on at cost.

The number of tools first, then whether it can change data. An agent that reads one system and drafts for a person sits at the bottom. One that reaches several back-office systems and can spend or change live records sits at the top. Each such action needs an approval step, a limit and a way to undo. Running in your own cloud adds servers too.

No. The handover gives your team the runbook and training under the agreed ownership terms. An agent that only reads, on systems that rarely change, can run in-house. The monthly running is worth its fee where the agent changes live systems or its tools change often. Every change should mean running the tests again.

As an add-on, priced after the pilot has proved itself against how you work today. A new tool needs new access limits, connected and tested. A new task needs new test cases and approval steps. Each is its own line, so you see what it costs and can add it once the first task's results are in.

Because an agent chooses its own path. A chatbot answers one question, and a workflow follows one fixed path. An agent decides each next step. So checks must cover every risky kind of action, wherever it happens: anything that can't be undone, goes outside your business or passes a limit. Designing and testing those checks across every tool is the extra cost.

Related

Other cost guides
and the service page.

How much does an AI chatbot cost in 2026?Cost guide: the options, what changes the price and what the fee buys.How much does a website cost in 2026?Cost guide: the options, what changes the price and what the fee buys.About AI agentsThe service page: what the work is, how it runs and the proof behind it.AI agents on the rate cardThe same ranges beside every other service, in one card.

Rankings on this topic

Scored on public evidence by the PixelCrayons Index. PixelCrayons is scored like every firm: ranked where it has a case study, listed where it has not.

Last updated