AI Agents & Agentic Workflows
I design agentic systems when a workflow genuinely benefits from adaptive reasoning or tool selection, while keeping permissions, state, irreversible actions, and recovery under explicit system control.
Janna Wong · AI Automation & Systems Architect · Remote · Global · async-first
Where I help
The implementation starts with the operating problem and its failure modes, then moves into tools and architecture.
Agent prototypes with broad permissions and unclear operating boundaries
Workflows that need adaptive tool selection or multi-step reasoning
AI systems that cannot explain what action happened or why
Teams using agents where a simpler deterministic workflow would be safer
Deliverables
Technology
I do not force every system into the same platform. The goal is the lightest architecture that remains reliable, observable, documented, and maintainable.
FAQ
Direct answers written to be useful to both technical buyers and search/retrieval systems.
Use agentic behavior when the system needs to interpret context and choose among tools or steps dynamically. If the path is known in advance, deterministic orchestration is usually easier to test and operate.
I keep the tool surface narrow, apply authorization outside the model, require approval for sensitive actions, validate tool inputs, and log execution so behavior can be reviewed.
Yes, but the integration should expose only the capabilities and data the agent actually needs rather than giving unrestricted access to a broad business system.
Related expertise
Send the current process, tools, constraints, and desired outcome. I'll review the context asynchronously and identify the most practical technical next step.