Contents
- Buy, configure or build: the three options side by side
- Deciding per agent with the HR Agent Map
- What each option costs, and how we price a build
- What building one takes after launch
- Who runs a custom HR AI agent, and who owns it
- Where a built agent stops after launch
- The legal side of building an HR agent
- Questions to ask before you buy or build
- Where Culturro fits
- Frequently asked questions
The buy-or-build question for HR AI usually gets asked once, for the whole function. Buy a platform or build your own. That wastes money in both directions, because the right answer changes from one agent to the next.
Some agents already sit inside the HR software you pay for, and turning them on costs little. Other agents need your policies, your systems and your approval rules, and no vendor sells those in a box.
So there are three options, not two. You can buy an agent from HR software you already run. You can configure one on a platform your team sets up. Or you can have one built, by a firm or by your own people. This post sorts the 122 agents on the HR Agent Map across those three options and sets out what each option costs, including how we price a build. It also covers what a built agent needs after launch, which is the part most comparisons skip.
One disclosure first. Culturro builds HR agents, so we gain when the answer is “build”. We’ll say where buying wins, and it wins often.
Buy, configure or build: the three options side by side
The table compares what you get, who carries the work after launch and what you hold when the contract ends. Cells marked “our reading” are Culturro’s view, not a published figure.
| Buy (agent in your HR software) | Configure (a platform you set up) | Build (a firm or your own team) | |
|---|---|---|---|
| What you get | An agent the vendor designed for its own product | An agent your team assembles from the platform’s parts | An agent made for your policies, systems and approval rules |
| Who runs it after launch | The vendor runs the model and ships updates. Your team sets access and content (our reading) | Your team keeps the flows and prompts working. The platform vendor runs the engine (our reading) | Your team, or the firm under a support contract. Someone must be named before launch (our reading) |
| Who owns it | The vendor owns the agent. You own your data | You own your setup. The platform owns the engine it runs on | Whoever the contract says. Ask for code, prompts and test sets to transfer to you |
| How the cost is charged | Subscription or usage. Workday sells agents with annual AI Flex Credits charged per action, not per employee | Platform licence or usage, plus your team’s time (our reading) | A project fee, then a separate run cost for hosting, model use and upkeep |
| Time to first use | As soon as it’s switched on and set up (our reading) | Depends on your team’s capacity (our reading) | In a Culturro engagement, a working proof in a few weeks, then typically a few weeks to a couple of months to build it |
| What you hold on the last day | Your exported data. The agent leaves with the contract | Your setup, which only runs on that platform | The code, documents and tests, if the contract transfers them |
The last row matters most for a company of 500 to 5,000 people with no AI team. A bought agent leaves when the subscription ends, while a built agent stays. But it stays useful only if someone on your side can run it.
Deciding per agent with the HR Agent Map
The HR Agent Map lists 122 agents across 12 HR areas, and each one carries a verdict. Of these, 48 are proven and 55 have a strong case. Another 7 are emerging, and 12 are marked “human decides” because the job itself is a decision about a person.
Each agent also names where a person sits in the loop. A person approves the output in 66 agents and reviews it in 25. In 30 agents a person sets the rules the agent follows. The one agent with no gate drafts total rewards statements, which involve no decision about anyone. Culturro builds no system that makes autonomous decisions about people, and every agent on the HR Agent Map follows that rule.
Before you price anything, run each agent you want through four questions:
- Does software you already pay for offer this agent today, and can it read your policies?
- Does the agent need data from more than one system, such as the HRIS or payroll?
- Who is the person in the loop, and are the approval rules your own or the vendor’s default?
- Who changes the agent when your policy changes, and how fast?
If the answer to the first question is yes, buy it. The other three only matter when the answer is no.
| Kind of agent | What to ask | Where it usually lands (our reading) |
|---|---|---|
| Answers employee questions from policy documents | Does our HRIS or service desk already offer this? Can it cite the policy it used? | Buy, or configure if no vendor covers your documents |
| Runs steps inside one system, such as onboarding tasks in the HRIS | Does the vendor’s agent cover our steps, or only theirs? | Buy |
| Works across several systems | Who connects the data, and who fixes it when one system changes? | Configure or build |
| Drafts an output a person approves, such as an offer letter or a pay change | Can we see and change the approval rules? Is every approval logged? | Build when the rules are your own. Buy when the vendor’s rules fit |
| Prepares a case where a human decides | Does the agent only gather and summarise? | Any option, as long as it never decides |
Sorted this way, most agents on the HR Agent Map should be bought or configured, even though we build agents.
Build is the answer for a smaller group of agents that cross several systems or follow your own approval rules. Some of them change every time policy does. If you haven’t picked which agents matter first, start with how to prioritise HR AI use cases before you decide how to source them.
What each option costs, and how we price a build
We found no independent study that compares the cost of buying an HR AI agent with building one. The figures you’ll find online come from companies that sell one side, and they don’t price the same thing. Some quote a single agent, and others quote a whole HR AI platform built in-house. Ask any firm which of the two it’s quoting.
How a Culturro build is priced:
We don’t publish a price, because the fee depends on the agent and your systems. We do publish how the fee works, and it’s built so you never sign for a full build on a guess.
- Strategy takes about a month on one fixed fee, quoted after a first call and scaled to the size of your function. It ends in a written go or no-go on whether anything is worth building. The roadmap and business case are yours either way.
- Proof takes a few weeks per use case, on a fixed fee per use case. It runs in our sandbox on synthetic data, or in your own cloud tenant on data your security team approves. Your team uses it, and we measure accuracy, failure cases and cost per query.
- Build is a fixed fee for the agreed scope, quoted only after the proof. It typically takes a few weeks to a couple of months per use case.
So before you pay for a build, you’ve seen the agent work and you know what each query costs to run.
Run cost, which most quotes leave out:
A built agent keeps costing money after launch. Someone also has to fix it when the HRIS changes a field, and update it when a policy changes.
A Culturro build fee includes a support period after handover. Your own providers bill you for hosting and model usage, so you see the run cost directly. There’s no Culturro platform underneath, so the agent keeps running if you stop working with us. Whoever you hire, ask for the run cost in writing, separate from the build fee.
How a bought agent is charged:
Buying doesn’t always mean a flat price per seat. Workday charges for the work an agent completes, through credits bought once a year. Its Self-Service Agent uses 1 credit to retrieve information and 5 credits to complete a task on its own. Workday’s page doesn’t publish a price per credit. So a bought agent costs more as use grows, and you need a usage estimate before you sign.
For the budget side (what HR teams are setting aside and where it comes from), see budgeting for AI in HR.
What building one takes after launch
A build keeps costing time after launch, as well as money. Every comparison we read weighs build against buy on price and speed, yet none says who runs the agent on day 90.
Who runs a custom HR AI agent, and who owns it
A built agent needs four roles on your side, even if a firm does the build:
- An HR owner who answers for what the agent says and does. This person signs off changes to its rules, so it needs someone senior enough to say no.
- A technical owner in IT or at the firm under a support contract. This person fixes it when something upstream changes.
- The people in the loop. These are the managers or HR partners who approve or review the agent’s output, as the HR Agent Map sets out for each agent.
- A data owner for the HRIS fields the agent reads. If those fields are wrong, the agent is wrong. Our post on HR data readiness covers what a build needs from your systems.
Employees will ask questions too. Why did the agent tell me this, and who checked it? Where a works council or union is involved, its representatives will ask the same questions. Someone in HR must be able to answer, with logs to back it up.
Ownership is a contract question, so settle it before work starts. You should hold the logs, and the accounts the agent runs on should be in your name. Our guide to HR AI consulting firms covers what you should hold on the last day of any engagement.
Here’s how a Culturro build covers those roles, as one example. In the strategy month we write down what AI will not decide in your function, and you name an owner on your side for every number. At build, we bring the architecture document and data-flow diagram to your security team’s first review. The code and prompts transfer to you at handover, with the test sets, documents and runbooks. We hold no production credentials after that.
You run the system. If your own team would rather build it, the roadmap and proof are written so they can. Those terms are the ones to look for in any firm’s contract, ours included.
Where a built agent stops after launch
The risk for a built agent sits in the months after launch, not on launch day. The HRIS gets an update and a field moves. A policy changes and nobody updates the prompts. The HR owner moves roles and nobody replaces them. We cover the common stopping points in why HR AI pilots stall.
Measurement is the other gap. A survey by Remote, a vendor, found that 91% of People leaders said AI had helped their business goals but only 62% had data that showed it. A built agent should ship with the measure it will be judged on. Our builds agree that number before anything ships, and your team also gets a usage and outcome report it can pull without us.
The legal side of building an HR agent
This section matters most if you employ people in the EU.
The EU AI Act treats AI used in employment as a high-risk use. Since the Digital Omnibus amendment, the rules for these uses apply from 2 December 2027, per the European Commission’s AI Act timeline.
The Act gives different duties to the “provider” of a system and the “deployer” who uses it. When you buy an agent, the vendor is usually the provider and you’re the deployer. Deployers still have duties under Article 26, such as human oversight and informing workers’ representatives. When you have an agent built and put it into service under your own name, you may become the provider, a role with heavier duties. The definitions sit in Article 3 and Article 25 of the Act. Ask your counsel which role you hold before a build starts, not after.
Questions to ask before you buy or build
Take these to any vendor or firm.
What you’ll own
- What do we hold if we end the contract tomorrow?
- Do the code, prompts and test cases transfer to us at handover?
- Which accounts does the agent run on, and whose name are they in?
Who runs it
- How fast does someone fix the agent when our HRIS changes?
- What is the yearly run cost, apart from the build fee or licence?
- How does the price change as usage grows?
How it treats people
- Where does a person approve or review the output? Can we change that rule?
- Can we see why the agent gave a particular answer?
- Is every action logged? Can we show the log to an employee or their representative?
What the numbers rest on
- Where does your cost or timeline figure come from: your own projects, or an estimate?
- Can we speak to a client who has run your agent for more than six months?
A vendor that can’t explain how its agent reaches an output is a risk you take on. Ask us the same questions. Our answers on ownership, access and security are written out on our services page.
Where Culturro fits
Culturro is an HR AI strategy consultancy first and a builder second, and we work with companies of 500 to 5,000 people. We start with which agents to source and how. Our roadmap will say to buy many agents on the HR Agent Map from software you already run.
Where a build makes sense, we build it inside your environment and under your security team’s controls. You own it and you run it. A person still makes each decision about hiring, pay or promotion, and we design every agent so that stays true.
Do you have a list of agents and no clear view on which to buy, configure or build? Book a 30-minute call. Bring the list and we’ll tell you which ones you don’t need us for.
Frequently asked questions
HR AI software or custom AI agents: which one should we choose?
Choose per agent, not for the whole HR function. Buy an agent when software you already run offers it and it can read your policies. Build when the agent crosses several systems or follows your own approval rules, or when it changes whenever policy does. The HR Agent Map gives a verdict and the person’s role for each of 122 agents.
How much does a custom HR AI agent cost?
We found no independent figure, and the numbers published online come from companies that sell one side. Culturro doesn’t publish a price. Each step has its own fixed fee, and the build fee is quoted only after a proof has measured accuracy and cost per query. Hosting and model usage are billed to you by your own providers.
How long does it take to build an HR AI agent?
It depends on whether you’re building one agent or a whole platform. A Culturro proof takes a few weeks per use case. The build typically takes a few weeks to a couple of months per use case, scoped after the proof. When you compare timelines, check whether a quote covers one agent or a whole platform.
Who owns a custom HR AI agent?
Whoever the contract says. Agree before work starts that the code, prompts, test cases and logs transfer to you at handover and that the accounts it runs on are in your name.
Is a custom HR agent a high-risk system under the EU AI Act?
AI used in employment is a high-risk use under the Act. Since the Digital Omnibus amendment, those rules apply from 2 December 2027. If you have an agent built and put it into service under your own name, you may become its provider, which carries heavier duties. Ask your counsel.
Can a consultant build an HR agent we run ourselves?
Yes, if the contract says so and your team has named owners for the agent after launch. Culturro builds on your own systems and hands over a system you own and run. Book a 30-minute call to talk through which agents are worth building.