Skip to content
nullbot ← Back to site
◍EN
Français English Español Português (Brasil) Português (Portugal) 简体中文 繁體中文 日本語 한국어 Deutsch Nederlands العربية

What it really costs

How much an AI agent costs

What nullbot isConnect your tools to AI agentsBuild your own mobile appBuild your own websiteWhat an employee really costsWhat a website costsWhat a mobile app costsWhat an SEO agency costsHow much an AI agent costsRunning a business with no marketing budgetAutonomous AI agentAI agent for small businessAI agent for freelancersAI agent for foundersAutomate your business with AIAgentic systemAI agents in businessSalesAI agents for customer supportAI agents for financeCRM and AI agentsAI agents for social mediaSEO and GEO visibilitySee your agents at workGovernance and budgets

The advertised price of an AI agent is rarely the one you will pay. On top of it come a model that consumes with every step of reasoning, tools to connect and, above all, the time spent checking its work. The right question is not what it costs in month one, but what it costs in month six, when nobody is thinking about it any more. Here is how to work it out, line by line.

Table of contents

  1. 1. The four lines on the bill
  2. 2. Three ways to pay for the model
  3. 3. The cost you do not see coming
  4. 4. Custom-built agent or platform
  5. 5. What month six costs
  6. 6. nullbot pricing
  7. 7. Frequently asked questions

1. The four lines on the bill

A quote or a pricing page often shows only one of these lines. All four exist from day one, though.

  • The tool: a subscription to a platform, or development ordered from a contractor.
  • The AI model: every read, every draft, every decision the agent makes goes through a model, and that passage is paid for one way or another.
  • Integrations: an agent without access to email, the calendar, the CRM or invoicing does nothing useful. Each connection has to be set up, then maintained.
  • Oversight: the human time spent reviewing, approving and correcting. It is often the heaviest line at the start, and the one you need to see come down.

An offer that looks cheap because it ignores the last two lines is not cheap: it moves the cost onto your hours.

2. Three ways to pay for the model

The same task can cost very different amounts depending on how the model is paid for. None is better in absolute terms; each has its limit.

  • An AI subscription you already have: the spend is fixed and known in advance. In return, the provider sets usage limits; once they are reached, the agent stops until the next renewal.
  • An API key: the provider bills by consumption. Nothing stops by itself, and that is exactly the risk: without a cap set on your side, the bill follows the agent's activity, good or bad.
  • A local model: no per-use bill, and data never leaves the machine. In exchange, the computer has to be powerful enough, and a local model is less comfortable with long or delicate tasks.

3. The cost you do not see coming

Traditional software costs the same whether the task succeeds or not. An agent that cannot finish starts again, rephrases, rereads, and consumes on every pass. A poorly defined mission can therefore cost more than a successful one, with nothing to flag it.

Three safeguards are enough to contain this risk: a spending cap that stops the agent instead of merely warning you afterwards, a mission whose result can be checked without argument, and a record of what each task actually consumed. These are the rules set out in AI agent governance.

4. Custom-built agent or platform

Having an agent built means paying for construction, then for maintenance the initial quote rarely mentions. The services the agent is connected to change their interfaces, models evolve, authorisations expire: every change becomes a job to pay for. Before you sign, ask for the annual maintenance cost on top of the build cost.

A platform is paid by subscription and can be cancelled. The vendor maintains the integrations for all its customers at once, which moves the maintenance burden out of your budget. The useful comparison is therefore about the nature of the spend: an investment followed by billed upkeep, or a monthly charge that stops the day the tool is no longer useful. No universal figure settles the question, and any number claiming to should be treated with suspicion.

5. What month six costs

  • Week 1: you review everything. The real cost is mostly your time, and that is normal: this is when the rules get corrected.
  • Month 3: the rules are proven, approval only covers irreversible actions, and model consumption is read task by task instead of guessed.
  • Year 2: what matters is that the agent still holds up. The integrations have kept pace with changes to the services, the history is kept, the budget has not drifted.

An agent that costs little but has to be rebuilt every quarter is expensive. Going live is only day one; the price that matters is the price of a team still at work two years later.

6. nullbot pricing

nullbot is paid by monthly subscription, with no commitment. Solo, at €24.90 a month, covers one company and five agents, with nullbot AI included. Multi Pro, at €199.90, opens several companies, unlimited agents and a free choice of model: local, an existing AI subscription or an API key. Finance, at €299, adds the trading room. Multi Pro and Finance start with a 14-day trial, and paying yearly lowers prices by about 20%.

Spending caps and rules are configured for the team, and consumption stays visible. See the plans in detail.

7. Frequently asked questions

Can an AI agent be free?

The software can be, and a local model charges nothing per use. That still leaves the machine that runs it, the integrations to maintain and your oversight time. Free to install does not mean free to run.

Why does an agent's cost vary from one month to the next?

Because model consumption depends on how many tasks there are and how hard they are. A spending cap turns that variation into a limit you know in advance.

Does an AI agent cost less than an employee?

The two do not compare line by line: an agent takes over repetitive, checkable tasks, not a whole job. The method for costing an employee is set out in what an employee really costs, and no saving can be promised without knowing what you delegate.

How can you estimate the cost before committing?

Hand over a single mission for one month, with a low cap. Note what each task consumed and the time you spent reviewing, then widen only what holds.

Going further

Read next: governance and budget control, what an autonomous AI agent keeps up over time and AI agents for small businesses.

AI NewsSecurityLegal NoticePrivacyCookiesCGUCGVDPAManage my cookies

© 2026 MARA LABS — nullbot. All rights reserved.Simplified joint-stock company (SAS) with share capital of €100 · 104 321 104