AI Agents & Integrations

I design and build AI agents around real work: research, content, support, internal operations and customer-facing tools.

How I approach the work

Agents built around a real job.

An agent is more than a prompt. It needs a clear task, relevant context, access to the right tools and a way for people to review its work. I help plan those parts together, then build the first version in the environment where it will be used.

Start a conversation
An AI agent interface calling a custom action.
An agent can use approved tools as part of a defined task.

The working parts

The parts that make an agent usable

Agent design
Define the task, user, inputs, outputs and the decisions the agent can make.
Instructions and skills
Write the operating guidance, task steps and handoffs the agent needs.
Tools and MCP
Connect approved APIs, site search, content systems, calculators or MCP servers.
Context and source material
Choose the files, structured data and live information the agent can use.
Review and permissions
Set limits, approvals and a clear handoff when a person needs to make the call.
Interfaces and harnesses
Put the agent in a website, internal tool or agent harness where the work happens.

What I can build

Ways this work takes shape

Internal agents
Agents for repeatable research, content, support or operations work.
Agent-backed website tools
Focused tools that help a site visitor complete a specific task.
MCP and API integrations
Connections between an agent and the systems it needs to read from or act in.
Content and publishing workflows
Agent steps that support briefs, drafting, review, tagging or publishing.
Prototypes and evaluations
Working versions used to test the task, source material and tool choices before a larger build.
Documentation and handoff
Instructions for the team using, maintaining or extending the work.

From task to working system

How we build it

  1. 1. Map the taskStart with the work itself: who does it, what they need and what a useful result looks like.
  2. 2. Choose the componentsSelect the model, instructions, source material, tools, MCP connections and review points.
  3. 3. Build the first versionCreate the agent or tool around representative tasks and content.
  4. 4. Test the workRun it with real examples, inspect the outputs and adjust the parts that are not reliable enough.
  5. 5. Put it to workLaunch in the agreed environment and document how the workflow should be used and maintained.

Starting points

Common first projects

Content assistant
Drafts from approved source files and brand guidance.
Support agent
Answers product questions from structured source material and selected tools.
Research tool
Collects and organizes information for a defined brief.
Workshop guide
Takes a participant through a structured exercise.

Build an agent for a real task

Send the task, source material and systems involved. I’ll help define a useful first version.

Start a conversation