AI Agents for the Steps Rules Can't Handle.
When a step needs to read, decide or reason, I build AI agents and custom code, alongside n8n and API integrations, to handle it.
Book Your Free Discovery CallWhat's Actually Going Wrong.
Rules Break on Exceptions
A fixed if-this-then-that workflow breaks the moment a case doesn't fit the pattern.
Unstructured Input Everywhere
Emails, PDFs and free-text fields hold the data you need, but simple rules can't extract it.
Triage Still Needs a Human
Someone reads every incoming request just to decide where it should go.
Tools That Don't Talk to Each Other
When one tool has no native connector to another, off-the-shelf automation tools stop there.
A Layer Above Simple Workflows.
Most of a process can run on straightforward workflow logic. The steps that can't get an AI agent or custom code, wired into the rest of your automation. For example, an enquiry arrives by email, the agent reads it, replies or routes it to the right person, and flags anything unusual for a human.
What I Build.
AI Agents for Triage &
Response
Reads incoming requests, classifies them, and responds or routes them with context.
Document & Data
Extraction
Pulls structured data out of PDFs, emails and scanned documents and feeds it into your systems.
Custom API Integration
Builds
Where no native connector exists, I build direct API integrations between your tools.
n8n Workflows for
Complex Logic
Multi-branch workflows for when Zapier or Make hit their limits on logic or cost.
How I'd Build This For You.
Identify the Judgement Steps
Where a fixed rule breaks down and reasoning is needed.
Design Guardrails
What the AI decides alone, and what gets a human check before it acts.
Build & Test Against Real Cases
Tuned against your real, messy inputs.
Connect & Monitor
Wired into your existing systems and watched closely after go-live.
Not Every Judgement Call Needs a Person.
Talk through the process on a free call.
No obligation.
Frequently asked questions.
Is this the same as just using ChatGPT?+
No. A chat tool isn't wired into your systems or data. Here, the AI agent is a built, tested part of your actual workflow, with guardrails on what it can decide alone.
How do you prevent the AI from making costly mistakes?+
I scope exactly what it can decide alone, add a human-review step for anything higher-stakes, and test against real, anonymised examples before go-live.
Where does my data go?+
When an automation uses AI, the data for that step is sent to the AI provider through your own account. I only send what the step needs, and I'll go through exactly what that involves, including any data protection requirements, on the discovery call before anything is built.
Do you need our own AI/API keys?+
Typically yes, so the provider bills you directly and usage stays under your control. I help you set these up as part of the build.