AI agent development and process automation
AI agent development for business processes
Many cases need judgement, such as a claim with a document missing or an invoice that does not match its order. Much of the work on them is still routine. An agent does those steps in your systems and hands the case to a person when a decision is due or it is unsure, so your team decides only where it matters. You see what each handled case costs and how often the agent got it right. The agent runs on your own servers or in your own cloud account.
Good fits, and poor ones
Where agents help
Cases that pass through several systems, involve documents and need some judgement. For example, the intake of claims and service requests, checks of invoices against orders and contracts, onboarding and KYC reviews, and the sorting of customer and IT requests.
Where they don't
Work that is the same every time, which a rule or a script does better. Decisions that the law or your own policy reserve for people. Processes where the agent can't reach the data or systems it would need.
Where a plain workflow does the job, we build that instead.
How much the agent decides
Assists
The agent
Reads, extracts and drafts
A person
Decides every case
Acts with sign-off
The agent
Completes routine steps within set limits
A person
Signs off on set actions, such as payments, and on anything the agent flags
Acts within a scope
The agent
Handles defined cases from start to finish
A person
Checks samples and every case the agent passes on
Every agent starts at the first level, and we review its results with you before widening its scope.
What a project delivers
- A map of the process showing, step by step, whether code, a model or a person does the work
- Agents that can use only the tools they need, with sign-off steps, logs and an audit trail
- A test set of your past cases with known outcomes, run again before every change
- Routing that sends routine steps to open-weight models in your environment where quality allows
- Connections to the systems you already use, through their APIs or MCP
What each case costs
We report one figure per handled case, in euros, that includes the models, the platform, review time and any rework. Next to it you see how many cases ran without a person, how often the agent passed a case on, how often it was wrong and how long cases took.
The figures in every report
Measured on your own past cases before the agent goes live, and again after every change.
Rules and approvals
Agents that touch personal data or employees' work raise questions for your data protection officer, your compliance team and, where you have one, your works council. We prepare what they need to decide: what the agent does, which data it sees, where a person decides and how each decision can be traced. The legal assessment stays with your counsel.
How a first project runs
01
We map one process with you, build a test set from past cases and measure what a case costs today.
02
We build the agent with your team, starting at the lowest level of autonomy that pays off.
03
We hand over the code, test set, runbooks and monitoring, and review the results with you at agreed intervals.
Reviewing an agent you already run
A project can also start from an agent that is already in production. The questions are usually similar. Does it complete cases correctly? Which of its failures end up with your staff? Do cost and quality shift after a model or prompt upgrade? We review the current workflow for one process and build a test set from historical cases you approve.
We score the business outcome of each case separately from its tool traces, because a run in which every API call succeeded can still close a case wrongly. Automated scoring is checked against your experts' judgement on a sample. Where evidence or access is missing, we state what cannot be assessed. This is engineering work. It gives your compliance and legal teams technical evidence and leaves their assessment with them.
- Cases the agent should deny, cases with unclear input and cases where a system returns an error
- Who owns each escalation, and whether monitoring shows the failures that reach people
- Repeatable tests your team can run again after every upgrade
- A prioritised list of fixes and a decision on what to change first
Common questions
Does an AI agent pay off for our process?
It usually does where cases arrive often, follow recognisable patterns and cost your people time on routine steps. We measure hours and errors per case today, test the agent on your past cases and show you the cost per handled case before you decide to build it.
Should our agents run in the cloud or on our own infrastructure?
It depends on the data the agent reads and the systems it acts in. Agents that work on contracts, personnel files or pricing usually run on your own servers or in your own cloud account in an EU region, behind one gateway that controls which model sees which data. The agent itself is the same either way, so you can start in your cloud account and move later.
Are you an AI automation agency?
We are engineers rather than an agency that resells automation tools. We build agentic AI in your environment one process at a time, measure the cost of each handled case and hand over the code, test set and runbooks.
Do you build intelligent document processing?
Yes, when documents are part of a case: reading invoices, claims or contracts, extracting the fields, checking them against your systems and passing anything uncertain to a person. Each extraction is tested on past documents with known answers.
Which platforms do you build on?
Frameworks such as LangGraph, the OpenAI Agents SDK, the Claude Agent SDK, Google's Agent Development Kit or Mastra, or plain code where that is simpler.
Can the agents and models run on our own servers?
Yes. Agents, test sets and open-weight models run on your own hardware. Commercial models are used only through data flows you have approved.
What happens when the process changes?
The test set changes with it, and every change to the agent is checked against it before it goes live.
Will you automate a whole department?
No. We automate one process at a time and measure it before the next one starts.
How long does a first project take?
You get an estimate once we have seen the process and the access to your systems it needs.
Related pages
AI engineering and architecture
Assistants, voice interfaces and agents in the products you sell, with the serving and retrieval underneath.
Custom model training
A small open-weight model trained for one task and compared with your current model on cost and accuracy.
How we work
How a project is scoped, how our engineers join your team and what stays with you at the end.
Related services
AI agent evals
Find out how often an agent gets the job done before it acts in your systems, by checking outcomes, tool calls and how reliably runs repeat.
AI automation
Routine work taken off your teams: processes automated step by step with rules, workflow tools, models and agents, approved by people and measured per case.
AI chatbots
Customers and staff answered faster from your data, with approved actions and a handover to people when it matters.
AI strategy and rollout
AI strategy and rollout that starts with one running, measured system on your own processes and builds the roadmap around its figures.
Data integration for AI
Answers from file shares, SharePoint, ERP and tickets in seconds, for assistants and agents, with permissions carried through and tested.
AI engineers for your team
Extra AI engineering capacity when your team lacks AI skills: experienced engineers who build with you and leave the know-how.
AI evals engineering
Evals built from your own traces show whether each change made your AI product better or worse, with a release gate that stays in your repository.
AI system testing
Before go-live, acceptance tests on your own cases tell you how often your chatbot, assistant or agent is right, in terms your QA team already uses.
Free AI usage review
One page for management on where your AI money goes, whether it pays off and what to change, from exports you choose to share. Free, no system access.
Enterprise AI
One approved AI platform for the whole company: an internal assistant instead of shadow AI, approved models, data that stays in-house and spend per team.
LiteLLM gateway
Budgets and access rules for every team's AI use, through a LiteLLM gateway that is hardened, connected to your identity provider and kept up to date.
AI platform for all teams
One approved route to AI for every team, with budgets, logs and a free choice of model, run on your servers or in your cloud.
System access for agents (MCP)
AI agents that look up and update data in your systems, with narrow permissions, human approval for changes and tests.
On-premise AI
AI on offers, contracts and customer files, with the data staying in your company: own servers or own cloud account, tested before go-live.
Azure OpenAI and Bedrock
GPT and Claude on company data, under the cloud contracts you already have: Azure OpenAI and Amazon Bedrock in your own account, with EU processing and one gateway.
Knowledge assistants (RAG)
Staff find answers in internal documents in seconds, with sources cited, permissions kept and tests on your own questions.
RAG evaluation
An assistant that answers from your documents, tested on retrieval and on answers in each language your users write in, so you know how often it is right and cites the right source.
vLLM on Kubernetes
Open-weight models on your own GPUs, so data stays in house and costs follow capacity you plan, sized from measured load.
From the blog
Agent test sets from past cases
How to pick past business cases for a process agent's test set and keep a held-out part that nobody tunes against.
Human approval in agent workflows
Which agent actions on business records need a person's approval, what the approver must see and how to keep approvals from stalling every case.
Monitoring a live process agent
The signals that show a live process agent getting more expensive or handling cases worse, measured per case and tied to objectives you agree in advance.
Talk to an engineer
Bring one piece of work you would like to improve. The call takes 30 minutes and needs no access to your systems or your data.
Look for the draft in your email program
Send it from there, and an engineer replies in person with times for a 30-minute call. If no draft has opened, write to us at miki@cloudsail.com.