Blog · 5 min read ·
Custom AI Agent Development: Scope, Cost and Timeline
Direct answer: We set up a custom AI agent for $1,500 (S), $2,900 (M) or $5,500 (L), delivered in 7, 12 or 20 business days. Prices are excl. VAT, and the fees you pay the model or tool provider are not included.
What drives the cost of an AI agent
Cost depends on how much the agent has to do. Three things matter most: how many workflows it covers, how many tools it connects to, and how much checking you want around it.
A three-step workflow with one integration is a small job. Two workflows with six integrations and a monitoring panel are a much bigger one.
The main cost drivers:
- Number of workflows and steps
- Number of tools and integrations
- Monitoring, alerts and approval points
- Handover time
- Monthly management after go-live
Anthropic's article Building effective agents separates two kinds of systems. In a workflow, LLMs and tools follow predefined code paths. In an agent, the LLM directs its own process and tool use. The more freedom a system has, the more testing and approval points it needs, and that is where cost grows.
If you are still deciding where AI fits in your business, start with our AI services page.
Setup tiers: scope, price and delivery
| Tier | Price | Delivery | Scope |
|---|---|---|---|
| S | $1,500 | 7 business days | 1 workflow, up to 3 steps, 1 integration, human approval step, 1-hour handover |
| M | $2,900 | 12 business days | Up to 8 steps, up to 3 integrations, monitoring panel and alerts |
| L | $5,500 | 20 business days | 2 workflows, up to 6 integrations in total, 2-hour handover |
Extras are priced up front: an extra workflow is $1,400, an extra integration $250 and an extra hour $90. If your scope grows during the project, we add these as separate items and tell you before we do the work.
Fixed tiers exist so that you can budget before we start. You see the number of workflows, steps and integrations next to the price, and you can compare it with any other quote.
In practice, S suits one repetitive internal task. M suits a longer process that needs monitoring. L suits two connected workflows.
What setup includes and what it does not
Included
- Workflow design and an instruction and rule set
- Tests on sample data
- A human approval step before the agent takes a sensitive action
- Handover, as listed for your tier
- Monitoring panel and alerts (M tier)
Not included
- Model and tool usage fees. You pay these to the provider, and we give you a monthly estimate in writing at setup.
- Personal or sensitive data without a written data-processing agreement.
- Any promise of a business outcome.
- Workflows, integrations or hours beyond the agreed scope.
AI output can be wrong. That is why the approval step is part of every setup and not an optional extra. Before a project starts, an audit report can show what your current tools and permissions look like.
Example: a seven-day S setup
This is a generic example. The order can change from project to project.
- Day 1: confirm the workflow, its inputs and outputs, and where a person approves.
- Day 2: write the instructions and rule boundaries.
- Day 3: connect the one integration and run the first tests on sample data.
- Day 4: fix wrong or unclear outputs.
- Day 5: test edge cases and failure paths.
- Day 6: you review and we make final changes.
- Day 7: one-hour handover.
This only works if the process is already clear. If the workflow changes mid-project, the scope changes with it. How we work describes our review and handover steps.
Choosing a workflow worth automating
Before building anything, pick the right task. Three criteria help:
- It repeats. It has a clear trigger and a clear output.
- It needs little judgement. Most decisions follow known rules, and a person handles the exceptions.
- Its data is readable. The input is text, forms or tool data the agent can read without constant manual cleanup.
Be honest about the downside too: if a task changes every week or depends on constant judgement, it is a poor first candidate. Choose the boring, predictable one.
A hypothetical shop owner could start with supplier emails: the agent reads each one, pulls out the order number and drafts a confirmation for a person to approve. The trigger, the data and the result are all predictable.
After setup: monthly management
Setup is one cost. Keeping the agent working is another. Once the agent is live, the real world starts to change around it: a supplier sends a new format, a tool updates its interface, or a rule that worked in testing meets an unusual case. Tools change, rules need updates and errors need watching. Monthly management is optional:
- S: $290 per month, 1 workflow
- M: $590 per month, 3 workflows
- L: $1,190 per month, 6 or more workflows
It helps most when your agent depends on outside tools that change, or when nobody on your team has time for it. Model and tool fees are still billed by their providers.
For what happens after launch and what a monthly plan should include, see managed AI agents: what monthly management covers.
FAQ
What does the $1,500 S tier include?
One workflow with up to three steps, one integration, a human approval step, tests on sample data and a one-hour handover. It takes 7 business days.
How long does setup take?
S takes 7 business days, M takes 12 and L takes 20.
Are model and tool fees included?
No. You pay them to the provider. We give you a monthly estimate in writing at setup, but the actual bill depends on your usage.
Do you promise a business result?
No. We test the agent and add an approval step, but results depend on your data and your process.
Can the agent handle personal or sensitive data?
Only under a written data-processing agreement. Without one, we do not process that data.