AI agents that run a process
It's not the model that hurts. It's the question of what the system will do on its own at 3 a.m., who answers for it, and what a mistake costs. So the agent we build runs one named process inside boundaries you set: operations with real impact wait for your approval, and every action lands in the log. From €6 000 net, 6–16 weeks to production, with a fixed quote before the contract.
In short
- Price and time
- from €6 000 net · 6–16 weeksa fixed quote before the contract; you pay for the work, not for the number of agents
- Boundaries and trace
- State-changing actions wait for your approvalthe model picks the next step from the case; every action is logged for audit
- Before you pay for the whole thing
- The pilot must hit a written targeton your data, only then do you decide on the full rollout
Problem
An agent without boundaries, escalation and a trace is a risk, not automation. Nobody hands it a real process, rightly so.
Outcome
In 6–16 weeks you get a system that runs the process itself: it watches the sources, does the work in your tools and stops exactly where you draw the line. Operations that change state on your side are drafted and wait for approval, and every decision stays in the log. From €6 000 net, with a fixed quote before signing. First the pilot on real data has to hit the written target; only then do you commit to the full build.
Who is accountable
The same team builds and maintains every deployment.
They sign the build
- Artem LisovtsovFounder, Full-Stack & AI/ML Architect
- Serhii IvanchatenkoHead of Backend & AI Integrations
The same team and the same rules will work in your process. Start with a free scan.
Book a free process scan (30 min)This is how an agent runs one process.
From a signal in the source to the trace it leaves. The green line is the boundary the system does not cross alone.
- 01
Watches the sources
the agent sees a new case in the agreed systems
- 02
Decides within boundaries
it checks the rules and sets the next step inside the task scope
- 03
Works in the tools
it runs the action on minimum permissions, with a trace
Boundary gate
in-bounds actions move on, sensitive ones stop
Done within boundaries
the step is closed, with an auditable log
Exception to a human
an operation with production impact waits for approval
We set the boundary with you during scoping. A named owner on your side takes the escalations.
Security
What stops the agent if someone tries to hijack it at 3 a.m.?
Not an input filter, but boundaries on the output. The agent runs with real permissions, and the most common attack is prompt injection: a hidden instruction in an email or document the agent reads starts steering it. OWASP puts agent goal hijack (most often via prompt injection) at the top of its 2026 risk list for agentic applications. We design so that even a hijacked agent can do nothing dangerous.
OWASP 2026Even a hijacked agent can do nothing dangerous
Least privilege, read-only first.
ToolsA hard "never" list enforced in code; external content is data rather than a command.
BoundariesIrreversible steps wait for human approval.
EscalationEvery action logged and replayable.
TraceThese are four of the same seven criteria that separate an agent from a mock-up. Boundaries are not an add-on to security: they are the security.
Data & compliance
An isolated, agreed environment: data does not leave it without your consent. Operations with production impact wait for human approval, and every action stays in an auditable log. Trace, human oversight and monitoring are modeled on the EU AI Act regime and provided as input to your operator documentation; classifying the system and its compliance stay on your side.
What ships with the agent.
- An agent running one named process from start to finish, within set boundaries and escalating exceptions
- An operator panel: live view, manual approvals, a stop and a log of every decision
- Integrations with your systems on minimum required permissions
- An isolated runtime: data does not leave the agreed environment without your consent
- A hard boundary between data and instructions: content the agent reads from sources is treated as data to assess, never as commands to execute
- Monitoring, alerts and handover documentation for your team
Not for
- Autonomy without boundaries or an owner on the company side
- Build work without a signed scope
Seven criteria every one of our agents meets.
The trust frame
Entry price
from €6 000
Delivery window · 6–16 weeks
You pay for the process the agent runs, not for the number of agents.
You get a fixed quote in the proposal after the free scan. You know the full cost before signing, with no hidden items.
What drives the price
- The number of data sources and integrations
- The scope of autonomy and the complexity of rules and exceptions
- Security, consent and trace-audit requirements
Cost over time
- Implementation buildfrom €6 0006–16 weeks
- Maintenance and oversightpriced individuallymonthly
Cost to run
An agent works in a loop, so it spends 5 to 30 times more tokens per task than a chatbot. What counts is completed-task cost rather than prompt cost. We keep it in check: caching for repeated context, a cheaper model for simple steps, and a hard limit at the escalation gate. You get the range for your process in the proposal, and we meter cost from day one: Cost Per Query, the cost of a single request.
Gartner 2026, via CIO DiveFirst the pilot has to hit a written goal. Only then do you pay for the full build.
What if it doesn't work
You've seen a demo that looked great and died in production.
You're not buying a demo. The pilot on your data has to reach a quality bar written down before we start, or we don't move to the full build.
The price will creep during the project.
You know the fixed quote before the contract, with no hidden items. You pay for the work rather than the number of agents.
We'll get locked into one vendor.
The code, prompts and data stay with you, on your infrastructure if your security rules require it.
What if it turns out we don't need an agent?
If the process doesn't meet the seven criteria, we say so plainly and don't build it. Either way you keep a written summary after the scan.
Free process scan
Start with a free process scan.
- 30 minutes with the engineer who would build it, not a salesperson.
- A review of the processes that cost you the most time and money.
- A written summary: what to automate, in what order, with cost ranges.
No sales deck and no obligations. If automation doesn't make sense, we'll write that too.
€0
30 minutes · written takeaway within 2 business days
Agent questions
Does my process actually need an agent?
Not necessarily, and that's good news. If an automation or an app can handle the process, we recommend the cheaper form. An agent makes sense when the process needs multi-step decisions and work across several tools at once.How do you keep control of an agent in production?
The same seven criteria the agent meets at launch keep applying in production: every action shows up in the panel and the log, and operations with production impact wait for human approval. After launch the system runs under maintenance.Is our data safe and GDPR-compliant?
Data is processed only in the environment we agree on, in the EU by default, and the scope of access is written down before we start. Where personal data is involved, we sign a data processing agreement (GDPR). The system gets the minimum permissions it needs, and only the assigned engineers have access.How is this different from a plain automation or other vendors?
An automation runs a fixed path; an agent makes decisions across multiple steps and tools at once, which is why it costs more and why we hold it to seven criteria. Our pricing is open from the start: we show the band here and a fixed quote in the proposal after the scan, with the system's boundaries written down explicitly.How much does an AI agent cost?
From €6 000 net. The price depends on the work the system has to do, not on the number of agents: data sources, integrations, autonomy, risk and audit. You get a fixed quote in the proposal after the free scan, and you know the full cost before signing.How long does an agent build take?
Usually 6–16 weeks from contract to production, depending on the number of integrations and approval paths. The timeline is part of the proposal and is not settled mid-build.What if it doesn't work in our process?
At the scan and before launch we set a quality threshold and a baseline, so the result is measured against something concrete. If the process doesn't meet the seven criteria, we say so plainly and don't build the agent; after the scan you keep the written takeaway whatever you decide.
You leave with a plan, not a sales pitch.
- 30 minutes with the engineer who would build it, not a salesperson.
- A review of the processes that cost you the most time and money.
- A written summary: what to automate, in what order, with cost ranges.
No sales deck and no obligations. If automation doesn't make sense, we'll write that too.