Agentic Systems and Claude Tooling

Works alongside experienced engineers on Claude-based delivery work, explaining when a fixed workflow suffices versus an agent loop and naming common patterns such as routing, parallelization and orchestrator-workers. Sets up and runs basic terminal sessions, talks through permission modes, and describes what the SDK and MCP add over raw API calls. Connects existing MCP servers, outlines subagent and skill packaging concepts, and flags finite context and typical agent failure modes in discussion.

Builds small, supervised pieces of client agent work: simple chained or routed workflows, a basic SDK agent with system prompts and allowed tools, and a minimal MCP server over stdio or HTTP. Configures project memory files, slash commands and plan mode, reviews diffs before applying them, and keeps sessions alive with compaction and just-in-time retrieval. Creates scoped subagents and a working skill, traces runs, adds smoke tests, and resolves tool-namespace clashes across servers.

Owns delivery of production agent components end to end. Designs single-agent systems with tools, memory, verification, stopping conditions and budgets, matching autonomy to action risk. Automates with hooks and headless pipelines, deploys sandboxed SDK services, and builds MCP servers with OAuth, least-privilege scoping and efficient tool surfaces. Orchestrates subagent fan-out, skill libraries and deterministic code, sets context budgets, outcome evals, injection hardening, guardrails and cost controls.

Acts as technical authority across client agent engagements. Challenges over-engineered designs, surfaces hidden failure modes, and introduces multi-agent orchestrator-worker or specialist-pool structures only where justified. Reviews SDK agents, MCP servers, skills, subagents and context strategies for safety, cost, drift and operability at fleet scale. Drives team adoption through settings policy, permissions and shared commands, coaches engineers, establishes templates, and runs production-readiness reviews and observability.

Shapes how the practice and its clients build agentic systems. Defines the agent architecture playbook, reference designs, orchestration and long-horizon task patterns, and the enterprise MCP orchestration architecture with gateways, registries and governance. Architects fleets of production agents on shared infrastructure, sets org-wide rollout, provider routing and cost controls, establishes reliability and postmortem standards, advises leadership on assistive-to-autonomous roadmaps, and reports productivity impact.