AI development process
We run AI and software projects through an AI software development lifecycle of five stages: Discovery, Design, Build, Test and Launch. Each stage has a clear deliverable and a review before the next one starts.
Five stages, from discovery to launch
The same stages appear on the homepage. Here, each one also shows what happens, what you receive and how the stage changes on an AI project. Nothing on this page is a promised duration, a fee or a client result.
- Step 1
Discovery
Clarify the workflow, the users, the data sources and what a useful first release must do.
What happens
We map who uses the system, which steps they repeat, which information they may use and how you will judge the first release.
What you get
A written scope for that release: the workflow, the sources, what is outside it, and the review that has to happen before design starts.
On an AI project
For an AI project we also check who may access the data, whether the use case fits, and whether the job is an agent, a chatbot or a feature inside software you already run.
- Step 2
Design
Map the journey, the architecture and the points where a person reviews the work.
What happens
We sketch the user journey, the system shape and a prototype of the risky parts before the full build.
What you get
The journey, architecture notes and a prototype you can review. Build does not start until that review is agreed.
On an AI project
On an AI project this includes how prompts and retrieval are designed, which tools an agent may call, and where a person must approve an action.
- Step 3
Build
Develop the agreed features and connect the approved systems.
What happens
We build the scoped features and the integrations, with access limited to what the design allowed.
What you get
The first release in a test environment, plus notes on the integrations and the accounts your business will own.
On an AI project
Where the product needs it, the build can include retrieval, agent tools such as LangChain, n8n or an API, and a model chosen for the job. Naming those tools describes capability, not a completed client project.
- Step 4
Test
Check the release against the acceptance criteria before anything goes live.
What happens
We run the agreed checks, including the cases that should stop, and you review the result before launch.
What you get
A record of the checks against the acceptance criteria, and a list of anything still outside this release.
On an AI project
For AI work the checks include an evaluation set, a look at unsafe or unsupported answers, and a discussion of what a run costs to operate. This page does not publish a score or a cost.
- Step 5
Launch
Deploy, hand over and agree how the system is watched after release.
What happens
We deploy to the agreed environment, hand over access and walk through how a later change is requested.
What you get
A live first release, the handover and a note on monitoring. Later tuning is a separate scope.
On an AI project
After launch, an AI system needs monitoring, a way to collect feedback and a decision about what gets tuned next. Tuning is agreed separately. It is not an open-ended promise.
How we build AI agents
An agent build still follows those five stages. The illustration below is one support question moving through a bounded workflow. It is not a client project. The engagement itself is the AI agents and workflow automation service. For a feature inside an application, start from AI software development.
AI customer support agent
One question moves through five steps. The agent searches approved sources, drafts a reply and hands the conversation to a person when the answer is not confident enough.
Customer question
A customer asks a question in chat, email or another agreed channel.
Knowledge base search
The agent searches approved help articles, product notes and policies.
Drafted answer
It drafts a reply from those sources, and the sources stay available for review.
Confidence check
A confidence check decides whether the draft is clear enough to send.
Reply or human handoff
A clear answer is sent, or the conversation is handed to a person.
Read the scope before you compare a price
AI development process FAQs
What is your AI development process?
We use five stages: Discovery, Design, Build, Test and Launch. Each stage ends with a deliverable and a review before the next stage starts. This page describes that process. It is not a timeline or a price.
How do you build AI agents?
We start from one workflow, the approved tools and the actions a person still takes. The build can use LangChain, n8n, Zapier or an API when that tool fits. The AI agents and workflow automation service is that engagement, and the AI agent cost guide explains what changes an estimate. This page does not publish a client result.
What does the AI software development lifecycle include?
Discovery agrees the job and the data. Design sets the journey and the review points. Build makes the scoped features. Test checks the acceptance criteria. Launch deploys and hands over. Later tuning is scoped on its own.
Do these stages include a promised duration or fee?
No. Duration and price follow the workflow, the integrations and the review rules. The AI development cost guide explains how to compare estimates. It is not a price list.
How do you work with teams in the US, UK and Europe?
We are a remote team. Meeting overlap, access and handover are agreed in writing. Working with us describes that collaboration. We do not have a US office.
Discuss your AI development process
Tell us about your business, the goals you want to achieve and your first priorities. We can discuss a practical next step.
Discuss your project