AI Agent Development
AI agents that take real work off your plate: handling support queues, researching leads, onboarding customers and answering internal questions. They connect to the tools you already use.
Who it suits, what it includes, what you own
- Who it suits
- Teams with a queue of repetitive, language-heavy work: support tickets, research briefs, onboarding questions, or the internal questions that get asked again and again. If the answers already exist in your documents or your systems, an agent can do the first pass.
- What it includes
-
- Discovery: finding the tasks that eat your team's time and working out which ones an agent can own.
- The agent itself, built on Claude or ChatGPT, or on a locally hosted model where the data is sensitive.
- Connections to your existing systems through automation tools, with Supabase for data.
- Monitoring, so you can see what the agent is doing and where it hands off to a person.
- Weekly working demos while it's being built.
- How it starts
- A free 30-minute call. You tell me which work is piling up. I tell you whether an agent is the right answer, then send a fixed quote before any work starts. No day rates.
- What you own
- Everything I build. Full source code, prompts, documentation and handover. No vendor lock-in, and no monthly retainer unless you want one. Accounts can sit in your name or mine, whichever suits you; we agree it in the scope.
- Ongoing costs
- AI model usage is metered per call, and the agent needs hosting on AWS or Vercel. I set out the expected running cost in the quote so there are no surprises.
- Limits
- An agent is only as good as the information it can reach and the rules it is given. Anything customer-facing keeps a route to a person. Locally hosted models suit sensitive data but need their own hardware or hosting.
What this looks like in practice
Illustrative scenarios. Which tasks an agent takes on is decided in discovery.
-
Support triage
An agent reads incoming tickets, answers the ones your documentation already covers, and passes the rest to a person with a summary.
-
Research briefs
An agent pulls from the sources you name and returns a brief in a set format for someone to check.
-
Onboarding
An agent walks a new customer through setup, asking what they need and adjusting the steps to match.
-
Internal questions
An agent that answers from your own documents, with the source shown, so the same question isn't asked in Slack every week.
Tools and platforms
I build agents on Claude and ChatGPT, and on locally hosted models where the data is sensitive. They connect to your existing systems through automation tools, with Supabase for data and hosting on AWS or Vercel.
How it works
-
Discovery
I find the tasks eating your team's time and work out which ones an agent can own.
-
Build
You see working demos weekly. No disappearing for months and then unveiling something off-brief.
-
Launch
Your agent goes live with monitoring, so you can see what it is doing and where it hands off.
One example
LucasRAG is my own self-hosted retrieval system: a knowledge graph built on LightRAG over more than 12,000 documents from my journals, project notes and research, queried from Claude through MCP. It's personal and it's live. It uses the same pattern as an agent that answers from your documents with the source shown.
Which queue should an agent take?
Tell me what's slow, disconnected or still only an idea. We'll work out whether I'm the right person to help.
Based in Reading, England.
Working with people in the UK and New Zealand.
Calls from 11am to 11pm UK time, Monday to Friday.