Rakin, for the team skill, add a small test set alongside the example library. Use the same 10 prospect briefs after each change: a few strong signals, a few weak ones, and at least one with no usable evidence. Keep these separate from the examples the model sees. Have it return the source-backed observation, a buyer-problem hypothesis, the email, and a send/hold decision. Score whether every factual claim is supported, whether the offer follows from the observation, and whether the question is easy to answer. A new hire supports ‘you’re hiring’; it doesn’t establish that their pipeline is broken. That gives you a way to test whether a new example actually improves the skill rather than just changing its voice. In an agentic outbound workflow, keep this check separate from sending and stop follow-ups when a reply arrives. Our full ten-case checklist and a worked example: autonomousgtm.ai/blog/test-cold-email-skill-before-sending
Dhiraj, put the agent to work before the message: find a current buyer signal, save the source, then draft one question tied to it. Reject drafts that could go to any company. Hiring a first SDR manager, for example, gives you a specific prospecting workflow to ask about. For inbound, turn one recurring buyer question into a useful LinkedIn post with a worked example. When someone responds, carry their exact question into the follow-up. Track qualified conversations and agreed next steps so you can see which topics attract buyers.
