Repetitive administrative work
Requests, forms, reminders, status updates and the copy-paste between tools that fills a working day.
AI automation and custom softwarethat removes repetitive work from your business.
Agentswerk is an AI consulting and software engineering company from Vienna. We find the processes that eat your team's time, then build the automation, the AI agent or the complete application that takes them over: integrated with the systems you already run, tested, monitored, with a person approving where it matters. For small and mid-sized companies in Europe and the US.
Contacting us means a free 30-minute call. No cost, no obligation, an honest answer.
Wohni · PropTech platform we builtClaraven · LegalTech we build on7+ years enterprise automationJava · Python · CI/CD · OpenShiftClaude · GPT · Codex · n8nISTQB · IREB · Tricentis certifiedVienna, Austria · EU data residency
What we automate
The work that should not be done by hand.
We start where the hours go: repetitive, document-heavy, cross-system tasks. Then we build the automation, the agent or the application that takes them over, with a person approving where it matters.
Document processing
Incoming PDFs, emails and scans: read, extract, validate, file and record them, faster and with fewer errors.
Customer inquiries
Classify what comes in, prepare the answer from your own knowledge, escalate the uncertain cases, improve response times.
Disconnected systems
Email, CRM, ERP, spreadsheets, invoicing and calendars talking to each other through proper integrations, not manual re-entry.
Manual data entry
Structured records created from unstructured input, checked before they land in your systems.
Decision-support workflows
Monitoring, scoring, research and preparation steps that repeat every day, so your people decide instead of collecting.
How we deliver
Automation and AI, delivered like enterprise software.
We cover the whole path, from spotting the right process to production monitoring. Automations run inside your systems, agents plan multi-step work with the tools you approve and ask for approval when judgment is required, and everything is tested, traced and owned by you.
Explore our servicesHow an agent works
Inside one task.
Keep scrolling, the five steps orbit the core. Your policies stay in the middle the whole time.
Scroll to rotate
It understands the task
An email, a ticket, a document or a scheduled trigger comes in. The agent interprets it using your records, policies and the current workflow state, and breaks the goal into controlled steps.
It acts, with tools you approved
Looks things up, fills forms, drafts replies, updates systems. Every tool has scoped permissions; consequential actions are validated first. Nothing outside the allow-list can be touched.
It checks its work and asks when unsure
Results are compared against expectations and your rules. Missing data, low confidence or a policy conflict → the agent pauses and routes the case to a person, evidence attached.
It hands over a result you can trust
Structured output that systems and people can rely on, a record, a summary, a decision packet, plus a trace of every input, source and tool call.
And it stops when it should
Your policies and stop controls sit at the centre of the loop, not at the end. Owners can pause, override or switch the agent off at any moment.
Leaders want AI to work by next Tuesday
But three foundations have to be in place first.
This is where we come in. We put the foundations down while building the first agent on top of them, so you see value in weeks, not after a year of "readiness".
01
Rebuild the data foundation
Agents are only as good as what they can see. We connect them to trusted documents, records and APIs with permissions, retrieval and source metadata, so answers are grounded and actions are authorised.
02
Rewire the engineering foundation
Testing, evaluation, CI/CD and observability for AI systems. We have built exactly this inside a regulated Austrian operator: AI-generated tests, automated code review, hash-based traceability.
03
Reimagine the governance foundation
Who may an agent act for, what may it touch, when must a human decide? We encode your policies into approvals, stop controls and audit trails, EU AI Act documentation included.
What we do
Six ways to work with us.
Every one starts the same way: a free 30-minute discovery call, in which we tell you honestly which of these fits, or whether none does. Then a "from" price, because every implementation is different, and something running at the end, not a report.
AI Opportunity Audit
We analyse your workflows and find the processes where AI or automation saves time or cost: feasibility, value estimate, recommended architecture and a prioritised roadmap.
What you get: a process analysis of the workflows that cost your team the most hours, the automation opportunities in them, a technical feasibility check, a value and ROI estimate, a recommended architecture and a prioritised implementation roadmap.
How it runs: a workshop with your team, a short review of your tools and data, then a written report within one week. You can hand the roadmap to anyone, including another vendor.
Best for: "We know we should do something with AI, but where does it pay off?"
Automation Sprint
One focused automation, built and launched in a short cycle: email classification, document processing, lead qualification, CRM updates, API integrations, n8n combined with custom code.
What you get: one focused automation, built, tested and launched within a short cycle, on n8n combined with custom code where the workflow tool ends. Documented and owned by you.
Examples: email classification and processing, CRM automation, document processing, lead qualification, customer inquiry automation, internal workflows, API integrations between the tools you already use.
Best for: repetitive admin, lead handling, document intake, reporting. Typically two to three weeks.
Custom AI Agents
A support, knowledge, research, document or operations agent, integrated into your systems, with validation, permissions, human approval and monitoring where necessary.
What you get: an agent for one job that plans multi-step work, uses the tools you allow, checks its own results and asks a person when unsure. Integrated into your existing systems, with validation, permissions, human approval and monitoring where necessary.
Examples: customer support agent, internal knowledge assistant over your documents, sales research agent, document analysis agent, operations agent, research agent.
Best for: work that needs judgment, your own data, or several systems at once. Typically four to eight weeks.
Custom AI Applications
Complete applications: frontend, backend, APIs, database, authentication, AI functionality and retrieval, integrations, testing, deployment and monitoring. Internal platforms, customer portals, document systems.
What you get: a complete application: frontend, backend, APIs, database, authentication, AI functionality and retrieval over your documents, integrations with external systems, testing, deployment and monitoring. Code, workflows and documentation belong to you.
Examples: internal AI platforms, document processing applications, customer portals, AI-assisted workflow systems, industry-specific business applications.
Best for: a workflow that deserves its own tool and user interface. Typically six to twelve weeks.
AI & Automation Consulting
An experienced technical partner for your own AI initiative, by the hour: architecture, LLM and agent design, retrieval, integrations, evaluation, testing, security, deployment and scaling. The same two engineers who build our projects, for as long as you need them.
What you get: the two founders as your technical partner, by the hour, for as long as you need them: AI architecture, automation strategy, LLM integration, agent design, retrieval, API integrations, evaluation, testing, security, deployment, scaling. Reviews of what your team or another vendor built. Second opinions before you sign.
How it runs: single sessions, a fixed number of hours per month, or a few days embedded with your team. No minimum project, invoiced monthly.
Best for: companies with an AI initiative already running, or an internal team that needs experienced engineers next to it.
Managed AI & Support
After go-live: monitoring, bug fixes, model and prompt updates, API maintenance, usage monitoring, small improvements and a monthly review. Priority support depending on the support plan.
What you get: monitoring, bug fixes, model updates, prompt improvements, API maintenance, minor improvements, usage monitoring and a monthly review. Priority support depending on the support plan.
Why it matters: models, APIs and your business change. A solution nobody maintains degrades within months. With a support plan it keeps working and gets better.
Best for: everything we built for you, and solutions built by others that you want kept alive.
€490 fixed
Book a Process CheckNot ready for an audit? Start with a Process Check.
One hour with one of the founders about one process that costs your team time. Within two business days you get a written recommendation: automate it, build for it, or leave it alone. The fee is credited if you continue with an audit or a sprint.
Examples of what we automate
Five workflows, end to end.
Each one is a chain of steps that today runs through somebody's inbox. We build the chain, and keep a person at the step where judgment is needed.
Property management
- Tenant email arrives
- Request is classified
- Property information retrieved
- Response prepared
- Employee approves
- CRM updated
Professional services
- Document received
- Information extracted
- Data validated
- Structured record created
- Responsible employee notified
Sales
- Prospect found
- Company researched
- Opportunity scored
- Personalised outreach prepared
- CRM updated
Customer support
- Customer request arrives
- Intent identified
- Company knowledge searched
- Answer prepared
- Uncertain cases escalated
Operations
- Incoming information monitored
- Predefined condition detected
- Workflow triggered
- Task or notification created
A person decides at the highlighted steps. Chains without one run fully on their own, inside the rules you set.
Why Agentswerk
We are engineers, not just AI consultants.
Two software engineers with enterprise and product experience. We can advise, and we can build and maintain the complete solution: not only a prompt, not only an n8n workflow, but the integrated system around it.
Software engineeringJava, Python, Quarkus, Spring Boot, PostgreSQL. Applications built to last, not demos.
AI implementationLLMs, agents, retrieval over your documents, structured output, evaluation. Used where it helps, left out where it does not.
Automationn8n and custom code, combined. The fast path for small companies, the robust path for growing ones.
API integrationsEmail, CRM, ERP, invoicing, calendars, e-health and public-sector systems. Solutions that live inside your environment.
Quality assurance & testingTest engineering is where we come from: test suites, edge cases, adversarial inputs, traceability.
Enterprise systems experienceRegulated Austrian environments: security, permissions, audit trails, EU data residency.
Human approval built inAgents ask, show evidence and pause where policy or judgment requires a person.
Production deployment & supportContainers, CI/CD, monitoring, maintenance. We stay after go-live.
Architecture
Modular by design, so nothing has to be rebuilt.
Models, tools, data sources and policies change. We layer agent systems so each part can evolve independently, and so your security team can see exactly where the controls sit.
How we buildExperience layerChat, voice, embedded UI, work queues, review screens
Orchestration layerState, planning, routing, approvals, retries, recovery
Agent layerRoles, prompts, skills, tool permissions, memory rules
Model layerFit-for-purpose models, routing, fallbacks, controlled endpoints
Knowledge & data layerRetrieval, vector and relational stores, source metadata
Integration layerAPIs, events, enterprise apps, repositories, pipelines
Governance & securityIdentity, authorisation, policy, human review, stop controls
Observability & evaluationTraces, quality metrics, cost, latency, drift, feedback
AI that works in action
Results, not roadshows.
What our founders have delivered inside enterprises and in their own products, measured against the manual process.
1 wk → 3 minManual test-data process automatedRegulated public-sector environment, in production
0Daily users of internal tools we builtTesters, project managers
0 yrsEnterprise software & automationJava, Python, CI/CD, regulated systems
0Live products co-built end to endPropTech and LegalTech
Case · Regulated operator
AI-assisted engineering platform for a regulated operator
Designed and built a platform that analyses Java code changes, generates JUnit tests and supports automated code review, integrated with Bitbucket via webhooks, queued generation jobs and hash-based traceability so every AI output is auditable.
Read the lessonsCase · Own product
Data-driven property platform with agentic imports
Co-built Wohni, a two-sided PropTech platform: agency listing imports, matching, market analysis and interactive calculators, the kind of production AI workflow we now build for clients.
Meet the teamHow we work
Five steps from a call to a running solution.
You can stop after any step and keep everything we made. Most clients start with the call and the audit.
Discovery call
A free 30-minute call to understand the business problem. No obligation.
Opportunity audit
We identify the most valuable automation opportunities and the order to tackle them.
Prototype or sprint
A focused, production-ready first solution on your real data.
Custom implementation
Integration into your actual systems, with testing, security and documentation.
Managed support
Monitoring and continuous improvement after go-live.
Engagement sizes
Pricing at a glance.
"From" prices, because every implementation is different. You get a fixed proposal within two business days after the call, and you always know the number before we start.
| Discovery call | free | 30 minutes, always the first step, no obligation: we say what fits and what it would take |
|---|---|---|
| Process Check | €490 fixed | one hour, one process, a written recommendation within two days; credited if you continue |
| AI Opportunity Audit | from €1,500 | process analysis, opportunities, feasibility, value estimate, roadmap |
| Automation Sprint | from €4,000 | one focused automation, built and launched in a short cycle |
| Custom AI Agents | from €7,500 | an agent integrated into your systems, with approval and monitoring |
| Custom AI Applications | from €15,000 | a complete application, frontend to deployment |
| AI & Automation Consulting | from €140/hour | senior engineers by the hour, no minimum project: architecture, strategy and hands-on technical guidance |
| Managed AI & Support | from €750/month | monitoring, fixes, updates, monthly review |
Insights
News and thinking.
Why most AI agents never leave the demoAgentic AI · 8 min read
AI-generated tests are only useful if you can trace themEngineering · 6 min read
The EU AI Act for mid-sized companies, on one pageStrategy · 5 min read
Designing the human-in-the-loop so people actually use itDesign · 7 min read
Featured
From pilot to production: the agent checklist
Twelve questions we ask before any agent goes live, evaluation, permissions, fallbacks, cost, ownership.
Read the checklistPartner ecosystem
Built on the platforms you already run.
We work model-agnostic and cloud-agnostic: the right model for the task, deployed inside your security perimeter.
Anthropic ClaudeOpenAIMicrosoft AzureAWSGoogle CloudOpenShiftBitbucket · JenkinsPostgreSQL · Oracle
Watch an agent work
This is what a good Tuesday morning looks like.
No demo magic, a real run log. The agent receives a request, plans, uses only the tools it is allowed to, checks itself, asks a human when it should, and hands over a result with a trace. Hover the console to slow it down.
Who builds thistools: 3 allowedpolicy: approvals > €5k0.0s