In Lesson 1 the job changed shape. It moved from doing the work to directing it. This lesson is about the skill that directing actually requires. It has a name people use loosely, prompting, and the first thing to do is take it seriously as a craft rather than a bag of tricks.
Prompting is not magic words
There was a phase where prompting looked like a hunt for secret phrases. Tell the model it's a world-class copywriter, promise it a tip, paste in the incantation, hope for gold. That era is over, and it was never reliable to begin with. Those tricks worked by accident, and you cannot run a pipeline on accidents.
Prompting is a thinking skill. It is the ability to say clearly what you want, who it is for, and what good looks like. A model is only ever as clear as the instructions you give it. The teams winning with AI are not the ones hoarding clever prompts. They are the ones who think clearly about the message and then write that thinking down.
Your thinking sets the ceiling
Here is the line worth keeping. The quality of your thinking now sets the ceiling on what the agent produces. The model can hit your ceiling. It cannot raise it for you.
That cuts both ways. A vague brief produces vague outreach, and the AI will not save you from sloppy thinking, it will expose it at scale. But the flip side is the encouraging part. The skill is learnable. You are not chasing hidden tricks, you are just getting clearer about what you actually want, and clarity is a muscle anyone can build.
What a good brief contains
If you can brief a strong human writer, you can brief an agent. The same five pieces apply.
- Goal. What you want to happen, stated concretely. Not "write an email," but "book a meeting with a VP of Sales who just visited our pricing page."
- Context. Where most of the quality lives. Who you are, what you sell, who the buyer is, what just happened. The model does not know your situation unless you tell it, and the moment you do, relevance jumps.
- Role. Who the agent should be. A peer reaching out, not a rep working a list. One line, and the whole tone shifts.
- Format. Length, structure, style. Three lines, under eighty words, no emojis. If you do not specify, the model guesses, and it usually guesses long.
- Constraints. The underrated one. Constraints sharpen rather than limit. No buzzwords. No claiming integrations you do not have. No "I hope this finds you well." A good no-list is half the brief.
The natural objection is that this is a lot of effort for one email. It is. But you are not writing one email. You write the brief once and the agent runs it across a thousand prospects. The thinking is a fixed cost, and the output is the part that scales. The brief is the leverage.
And none of it is one and done. Great prompting is a loop: first draft, then shorter, then more direct, then less generic. You are collaborating, not vending.
How this works inside Alta
In Alta, this idea becomes concrete, and it surprises people. When Katie writes a message, she is not reading only the small prompt box. She pulls from a stack of inputs at once: the Pitch, the Rep, your Knowledge base, Global instructions, CRM email examples, enrichment variables, the prospect's data and signals, and the logic of the workflow step.
A few of these are worth understanding precisely, because they decide output quality:
- The Pitch is the master input. It carries your positioning, value props, proof points, and call to action. If the Pitch is vague, every message comes out vague, no matter how clever the step prompt is. Fix the Pitch first.
- Global instructions are your always-on rules. Never use emojis, keep it under eighty words, and so on. They ride along in every prompt behind the scenes, and you can switch them off for a single step when a campaign needs something different.
- The Knowledge base only fires when the Pitch calls for it. Uploading case studies does not force the AI to use them. The Pitch decides what is relevant, so the document supports the message rather than dictating it.
- CRM email examples are reference, not law. They show Katie your preferred style and structure, but if your prompt conflicts with the example, the prompt wins. Guidance, not handcuffs.
- Enrichment variables are your sharpest personalization lever. They pull real-time detail, the company name, a website summary, a recent signal, straight into the message. Name them clearly. A variable called company_pain_point is usable. One called custom_field_1 is a guess.
Acme Corp in practice
Acme is prompting Katie to reach the VP of People at an account that just opened a People Ops role. The Pitch carries the positioning. A Global instruction keeps every message under eighty words. An enrichment variable pulls in the specific job posting. The step prompt says reach out as a peer, mention the hire, and skip the buzzwords.
Because each of those layers is clear, the message lands specific and human, even when Katie is sending it across hundreds of accounts. That is the whole difference between an agent that sounds like you and one that sounds like a robot. It is clear thinking, written down.
Key takeaways
- Prompting is a thinking skill, not a set of magic words. Clarity beats tricks.
- The quality of your thinking sets the ceiling on the agent's output. The model hits your ceiling, it does not raise it.
- A good brief has five parts: goal, context, role, format, constraints. Context carries most of the quality, and constraints sharpen the result.
- The brief is written once and scales across every prospect, so the effort pays back fast.
- In Alta, the Pitch is the master input. Global instructions are always-on, the Knowledge base only fires when the Pitch calls for it, CRM examples are overridable reference, and well-named enrichment variables drive personalization.
Up next: When AI Does the Prospecting (Outbound)
You now know how to brief an agent well. Next we put it to work on the first motion. When AI does the prospecting, how a signal becomes a reason to reach out, and how Alta’s agent Katie turns that signal into a booked meeting.