According to Michelle Smith, Vicki Reyzelman, senior solutions engineer at Akamai Technologies, demonstrated on a recent episode of Zero to Agent in 30 Minutes how she created an agent for her Chat About AI YouTube channel. With over 25 years in technology and a software engineering background, Reyzelman wanted an agent that would monitor channel performance, identify bottlenecks, and recommend actions to grow subscribers and improve click-through rates.
The step-by-step process
-
Define the goal: Reyzelman began by clearly stating the problem the agent should solve — which videos perform well, where the channel underperforms, and what actions could help more viewers discover and subscribe.
-
Review available data: She examined the data sources before designing the workflow. YouTube Studio exports include metrics such as impressions and click-through rates, while the YouTube Data API supplies information about channels, subscriptions, comments, and activity — but not every metric available in YouTube Studio.
-
Choose how the agent will receive data: Rather than connecting the agent directly to her account, Reyzelman downloaded analytics as CSV files. This approach gave her more control during testing; she noted that an authenticated API connection could be added later after testing permissions and guardrails.
-
Write the skills file: The skills file defines the agent’s mission, capabilities, data sources, rules, and expected output. Reyzelman instructed the agent to audit channel metrics, identify bottlenecks, compare performance over time, and recommend actions tied to subscriber growth and click-through rate improvements.
-
Add guardrails and acceptance criteria: She specified how the agent should handle data and what to avoid. Instructions included using only the supplied numbers, ignoring bot activity, reporting silent failures, and staying within approved systems and data sources. These constraints help prevent unsupported conclusions and actions outside the agent’s role.
-
Build and test the agent: Reyzelman used the console’s quick-start process to create the agent, add its instructions and sample analytics, select an environment, and start a session. Her first test encountered input problems, so she stopped the session, adjusted the data, and ran it again.
-
Monitor cost and performance: Observability tools showed token consumption, errors, active time, and deployments. Reyzelman explained that more capable models may require less context but typically cost more to run. Teams can balance model capability against token cost by improving instructions, choosing less expensive models for simpler tasks, or splitting a workflow among specialized agents.
Lessons and recommendations
Reyzelman recommends revisiting the skills file as new requirements arise. Clearer instructions, stronger guardrails, and regular testing help the agent remain useful over time. Her process underscores a practical software-engineering lesson: faster implementation does not eliminate the need for clear requirements, reliable data, security controls, and testing.
Coming next
The next episode will air on August 12, 2026, when AI and machine learning leader Ofer Mendelevitch will explain how to design multi-agent systems that can keep work moving without constant human involvement. He will use Jam to build a team of agents that can plan, divide work, execute tasks, review progress, escalate problems, and repeat the process. He will demonstrate how an architect agent can take a project brief, recruit developer agents, assign work, and guide the project toward a clear definition of done.
