The opportunity often sits between systems: finding context, preparing an answer, reviewing it, and moving it into the tool where work happens. I build workflows that connect those steps, combining agent capabilities with explicit rules and human judgment.
Understand the workflow before automating it
We identify the inputs, decisions, handoffs, and outputs that matter. The workflow distinguishes preparation from approval and defines what should happen when information is missing. This gives the team a common understanding of what the system is responsible for.
Connect knowledge to controlled actions
I build integrations around the context the workflow needs and the actions it is allowed to take. That can include MCP tools, APIs, knowledge retrieval, and familiar collaboration channels. Access is scoped to the task, with review before consequential changes to external systems.
Make progress visible and recoverable
A useful workflow needs to handle interruptions, changing context, and unsuccessful tool calls. I design for persisted state, clear review points, and observable execution. We evaluate outputs against representative tasks and use feedback from the people doing the work to improve the system.
What we scope together
Deliverables are agreed around your needs. A typical scope can include:
A working, scoped workflow
The agreed path from input to reviewed output, connected to the systems in scope.
Evaluation and operating controls
Representative tasks, quality checks, permissions, and a clear account of how people intervene.
Code and practical handover
Documentation and walkthroughs covering how the workflow runs, how to change it, and how to handle failures.
Before we begin
Can you work with our existing systems?
Yes. The starting point is your current environment. We assess the available APIs, data access, identity controls, and constraints before agreeing which integrations to build.
Do we need multiple agents?
That depends on the work. A single agent or a deterministic workflow may be sufficient. I use multiple agents when distinct responsibilities justify the extra coordination and operating complexity.
Can people approve actions before they happen?
Yes. Review and approval can be explicit workflow steps, including before an external write. We agree which decisions need human control and implement those boundaries in the system.