DEV Community: Maksim Ilin

How much does it cost to implement an AI agent in 2026

Maksim Ilin — Thu, 10 Sep 2026 19:35:34 +0000

"How much does an AI agent cost" is the question I hear more than any other. The honest answer: from about fifteen hundred euros for a narrow process to hundreds of thousands of dollars a year if you hire a team. The difference is not the model. It is what exactly you are buying.

What drives the cost

Five factors I assess in every request, and a sixth that everyone remembers last.

The process the agent takes over. Replying to standard emails and running a full order cycle (find the customer, check stock, issue an invoice, notify the warehouse) differ in effort several times over. More steps and a higher price per mistake mean a higher cost.

Number of integrations. Every system the agent connects to (CRM, accounting, email, warehouse, payments) is a separate connector, separate access rights and separate ways to break. An agent with one integration and an agent with five are different projects.

Data quality. If message history, documents and reference lists are tidy, the work moves fast. If the product catalog exists in three versions in Excel, the first days go into cleaning it up. That has nothing to do with AI, but without it the agent will make mistakes.

Guardrails and evaluation. A test set of real examples, rules about what the agent may not do, thresholds below which it hands the case to a person, a log of actions. In the KORE1 survey for 2026, employers named building an evaluation harness the top skill of an agentic engineer, and it is a visible share of the budget.

Hosting and the model. Beyond one-off development there are monthly costs: calls to the language model API and the server the agent lives on. More on that below.

And the sixth: maintenance. Models get updated, the CRM changes its API, the business gets a new type of request. An agent nobody looks after degrades within a few months.

What the alternatives cost

Market figures as of September 2026.

Freelance. According to Upwork data from May 2026, AI engineers charge 35 to 60 USD per hour, ML engineers 50 to 200. In my experience an agent for one process takes from a few dozen to a couple of hundred hours, so the range is wide. The main risk is not the rate but that you become the project manager: spec, acceptance, quality control.

In-house engineer. The KORE1 survey puts the base salary of an agentic engineer in the US at 155,000 to 210,000 USD for mid level, 210,000 to 290,000 for senior and 290,000 to 360,000 and above for staff. That is 15 to 20% more than a comparable ML engineer, before taxes, bonuses and equity. For a company with one or two processes to automate, that hire will not pay off. In Europe there are fewer than 0.5 candidates per advanced AI vacancy, according to Interface, so the search alone takes months.

AI governance consultant. Day rates of 800 to 2,000 USD, according to an analysis of EU AI Act job postings. Needed if your product falls under the regulator's requirements. This is not legal advice: a lawyer should confirm the obligations for a specific product.

A scoped project. My format: an AI readiness audit from 450 EUR, AI inside an existing product (RAG, LLM features) from 1,200 EUR, an agent for a business process from 1,500 EUR, an MVP from 1,700 EUR. First working version for a narrow process in two weeks. The price grows with integrations and process complexity, but scope and deadline are fixed before the start.

Rates, salaries and why agentic engineers command a premium are covered in my review of the AI job market in 2026.

Timeline

For a narrow process the first working version appears in two weeks. By working version I mean an agent on real data, with an evaluation set, connected to at least one live system, already used by at least one employee.

Then two to four weeks of shakedown: reviewing the cases where the agent got it wrong, extending the test set, tuning thresholds. Then a decision on whether to extend to neighboring processes.

Projects with three or more integrations and messy data take longer. I prefer to say so during the audit rather than in week three.

Monthly running costs

I will not quote numbers here, because they depend on volume and on the model, and provider pricing changes every few months. What matters is the structure:

Model calls are billed by the amount of text in and out. An agent that reads long documents costs more than one that answers short emails. Long instructions in every request cost money too.
Not every task needs the most expensive model. Classification and field extraction often work on cheap models; a strong model is needed on one or two steps.
Hosting for a single-process agent is usually small, comparable to a SaaS subscription. It gets expensive at thousands of requests per hour.
A spending cap is mandatory. An agent stuck in a loop overnight can eat a month's budget.

I include an estimate of expected usage in the audit, because for some processes it is tokens, not development, that decide whether the agent pays off.

What a cheap agent really costs

The temptation is understandable: take a no-code builder, sketch a prompt, connect email and launch in a couple of days. Sometimes it works. More often the demo impresses, then in production the agent answers wrong in some cases, and nobody knows which, because there is no test set. Someone edits the prompt, something else breaks. Two months later the agent is switched off.

Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027. Some of those are large companies with large budgets, but the mechanism is the same: no evaluation, no data, no process owner.

A cancelled project costs more than the money spent on it. It costs the trust of employees who will greet the next agent with a smirk, and the time in which a competitor did the same thing properly.

A budgeting example

A hypothetical company: a sales team of a few people, inbound requests by email and messenger, a CRM. The task: qualify leads and reply within a minute, including evenings and weekends.

AI readiness audit: from 450 EUR, one or two weeks. Output: a process map, an estimate of the effect, a verdict per process, an estimate of monthly model and hosting costs.
Agent for lead qualification with one or two integrations (email and CRM): from 1,500 EUR, first working version in two weeks. A third integration and complex handoff logic raise the amount; I give the exact figure after the audit.
Monthly: model calls and hosting, scaled to the number of requests, plus maintenance by agreement.

The one-off budget for this scenario starts at roughly 2,000 EUR for the audit and the first version of the agent and grows with the number of systems to connect. For comparison: one month of an in-house agentic engineer in the US at the bottom of the KORE1 range is about 13,000 USD in base salary alone.

All three services with prices are listed on the home page.

If you want a price for your own process, send a short brief: which process, which systems are involved, how many requests per day. I will reply with a range and a timeline.

Originally published at ilinmaks.com.

AI agent or plain automation: where the real ROI is

Maksim Ilin — Thu, 10 Sep 2026 19:07:15 +0000

The market pays more for agents right now than for anything else in applied AI. The skill "agentic AI" grew 280% in US job postings in a year, to roughly 90,000 listings, according to Stanford AI Index 2026. An engineer who builds agents earns 15 to 20% more than a comparable ML engineer, per the KORE1 hiring survey for 2026. And the same market, according to Gartner, will cancel more than 40% of agentic AI projects by the end of 2027.

I don't see a contradiction there. The premium is paid for a rare skill. The cancellations hit projects where an agent was put somewhere it wasn't needed, or where nobody could prove it worked. When someone asks me for an agent, my first question is usually not "which agent" but "do you need an agent at all".

When plain automation wins

There is a whole class of tasks where a language model is simply not needed. The signs:

the input is structured: a form, a spreadsheet, an API, an email that follows a template;
the rules fit on one page and don't change every week;
the answer has to be identical every time, and a mistake is expensive;
the volume is high, but there are few decisions per unit of work.

Moving orders from the CRM into accounting. Payment reminders. Parsing a supplier price list that always comes in the same format. Assembling a weekly report from three sources. All of that is a parser, a rule, or a scenario in any automation platform. It is cheaper, it behaves the same way every time, and it has no token bill.

I would not put an agent on a task you can draw as a flowchart. The agent will be more expensive, slower and less reliable than a plain if-then.

When an agent earns its cost

An agent belongs where the input is messy and there are many small decisions.

Inbound emails to a distributor, where one message contains an order, a complaint and a question about delivery dates. Leads who write in a messenger in free text. First-line support where the questions repeat but the wording is new every time. Invoices, contracts and requests that arrive as PDFs, photos and scans.

What these have in common: the rules are too many and too fuzzy to write down. A person handles it through judgment. An agent reproduces that judgment well enough to take most of the routine off the person and leave them only the ambiguous cases.

The second sign is a chain of actions. The agent doesn't just classify an email. It finds the order in the database, checks stock, drafts a reply and creates a task for the manager. Each step on its own is trivial. The value is in linking them.

How to measure ROI

I use four numbers, and I insist they get measured before launch, not after.

Hours. How much time staff spend on the process today. Not a guess of "about half a day", but a one-week measurement. Multiply by the fully loaded hourly cost.
Response time. How long a customer or a colleague waits. In sales this is conversion, directly: a lead who got a reply in a minute and a lead who waited until morning behave differently.
Error rate. Orders typed in with a mistake, invoices sent to the wrong place, requests that got lost. Errors have a price, and usually nobody has counted it.
Revenue being lost right now. Leads nobody answered. Requests that arrived after 6 pm. Repeat sales nobody got around to.

If the sum of those four lines is smaller than implementation plus a year of maintenance, the project should not happen. That is a normal outcome of an audit, and I say so to clients.

Evaluation first, agent second

In the KORE1 survey, employers named the top skill of an agentic engineer as neither prompts nor frameworks but the ability to build an evaluation harness: a set of test cases the agent is checked against automatically after every change.

So before the agent is written, you need real examples from the process with the correct answers, ideally at least a hundred. Emails and the way your best manager replied to them. Documents and what should have been extracted from each. The agent runs against this set every time the instructions, the model or an integration changes.

Without it you don't know whether the agent got better or worse after an edit. You find out from customers. The projects Gartner counts in the cancelled 40% are very often exactly this: the demo worked, production drifted, and nobody could say why.

Data matters more than the model here. If there are no examples because nobody ever recorded the process, the first days go into collecting them. Boring work, and it decides the result.

Guardrails and handoff to a person

An agent in production has to do three things: never do what it isn't allowed to do, stop when it is unsure, and hand the case to a person in a way that lets the person understand the context in ten seconds.

In practice this is a list of allowed actions (reading an order is fine, changing a price is not), confidence thresholds below which the agent opens a task for an operator instead of answering, and a log of every action. Plus a spending cap, because an agent stuck in a loop can burn a month of tokens overnight. That one is not a scare story.

The operator doesn't disappear. They stop doing the routine and start handling exceptions. Usually it is the same person who did it by hand before, and in the first weeks they are the one who spots where the agent is wrong. That is the most useful feedback in the whole project.

Three steps to decide

Audit the process. One or two weeks: where the input comes from, who does what with it, where time and money leak, whether there is data for evaluation. The output is a list of processes with an estimated effect and a verdict for each: rule, agent, or leave it alone.
Pilot one process. Two weeks, a first working version on real data, with an evaluation set and guardrails. Not five processes at once. One.
Measure. The same four numbers as before the start. If they hold, extend to neighboring processes. If not, stop, and the loss is limited to the pilot.

This order is how my services are built: an AI readiness audit from 450 EUR, an agent for one specific process from 1,500 EUR, first working version in two weeks. Why the market pays a premium for agentic engineers, and what that means if you are the one buying, is in my review of the AI job market in 2026.

If you have a process that eats hours and you are not sure an agent belongs there, send me a short description: what comes in, who handles it, how long it takes. I will reply with what I would do in your place.

Originally published at ilinmaks.com.