Do we really need an AI agent?
Not always. A combination of traditional automation and AI in well-defined steps is often more reliable and less expensive.
SERVICE
AI and process automation
I integrate AI and automation where they remove friction from a concrete process. Before selecting a model or tool, I separate tasks that need reasoning from those better solved with reliable rules. This creates systems that are useful, controllable and sustainable.
Let’s talk about your projectProblems suited to automation
- Your team repeats classification, research, summarisation or content preparation tasks.
- Information is manually moved between tools, spreadsheets, email and business systems.
- You want to add AI capabilities to an existing product without losing control of the output.
Possible solutions
- Workflow analysis and an estimate of where automation creates practical value.
- Internal assistants, company knowledge retrieval and human-approval workflows.
- LLM integrations, API orchestration and automations across existing tools.
- Evaluations, error tracking and fallback rules for uncertain cases.
PROCESS
A pragmatic approach to AI
- 01
Measure the starting point
We define the current process in terms of time, quality and risk so that improvement can be verified.
- 02
Choose the level of autonomy
Not everything needs to become an agent: we distinguish rules, suggestions and actions requiring approval.
- 03
Design controls
Structured inputs, evaluations and fallbacks make behaviour observable and more predictable.
- 04
Fit the real workflow
The solution should enter the existing tools and responsibilities instead of creating a parallel process.
FAQ
Common questions
How do you limit incorrect output?
With selected data, structured outputs, evaluations, confidence thresholds and human approval for sensitive steps.