One handles her admin. One coaches her sales. Both sound like you, and both belong to the firm.
The everyday helper. Captures her meetings, drafts her emails, and keeps her pipeline current.
The thinking partner. Reads her calls, recommends her next move, and keeps your tone in every conversation.
Every AI agent you build needs the same four things any new hire needs. A clear role. A clear objective. The context to do the job well. And the team and skills they can call on.
The position the agent fills.
The job title. "Sales coach." "Operations director." "Email assistant." The same way you would write a role description before posting a job.
What the agent is hired to deliver.
The outcome it exists to produce. The thing you would put on its 90-day review. If you can name the role but not the objective, the agent will not know what success looks like.
What the agent needs to know to do the job well.
Everything you would brief a new hire on before their first day. Five sub-pieces:
What the agent can call on beyond itself.
Two layers, often used together as your AI stack grows across the firm:
A skill is a folder with four files. The agent reads them when it needs to run the routine. Same shape every time means repeatable, predictable output.
Documentation of what the skill is for and when the agent should reach for it. Without this, the skill never gets used.
The exact procedure the agent follows. The same shape every time. Specific enough that the result is repeatable.
One or two samples of what a great output looks like. The agent uses these to calibrate against the standard.
What gets checked before the work ships. The agent runs the output against this list and fixes anything that misses.
For Brandy's V1, no skills are being built yet. The Coach handles whatever Brandy brings to it. The first time you catch Brandy giving the Coach the same instructions twice, that is the moment the work has earned a skill.
One folder. A handful of files. You drop the files into your Claude Projects and the tools work. You do not have to write any of these yourself.
R&D Advantage AI/ | |-- Shared Context/ | |-- R&D Advantage Business Context.md | |-- Russell (Luxury Academy) summary.md | `-- Hansen (Influence) summary.md | |-- Brandy's Assistant/ | |-- System prompt.md | `-- First test prompt.md | `-- Brandy's Coach/ |-- System prompt.md `-- First test prompt.md
The Shared Context files contain everything you saw in the tool cards above. Your business, your voice, your relationship-first approach, the Russell and Hansen frameworks. We package them as uploadable files instead of bullet points so the tools can actually read them. They get uploaded to BOTH tools, and they are reusable for any future tool you build for the firm. Build the context once. Reuse it forever.
Brandy's first version was a writing assistant. The new version is a coach. Five specific things changed.
You noticed this yourself when you tested it. The tool felt like a writing assistant, not a coach. It defaulted to Hansen on cold contact when Russell should lead. The instructions were a copy of yours, written for you as CEO, which made them a poor fit for Brandy's role.
The new tool leads with Brandy's role, defaults to Russell as the working register, and pulls Hansen in for relationship moments. The frameworks live in separate files the tool reads when needed. The system prompt itself is short and clear.
OldThe tool's identity (a calm, peer-level coach) was buried under a long list of rules near the bottom. NewThe first thing the tool reads is who it is and what its top priority is. Everything else builds from that.
OldThe tool was told to coach Brandy on six different things at once. Family office strategy. R&D Advantage sales. Incentive Partner sales. Proposal writing. Meeting prep. Call grading. Six jobs in one tool. NewThe Coach has one job: situational coaching. Other jobs become skills, or move to the Assistant tool.
OldDozens of always and never rules. The tool tried to follow them all at once and produced stiff, overcorrected output. NewThe tool gets the frameworks, the brand voice, and the priority. We let it choose. We tighten the rules only when it actually needs tightening.
OldYou noticed the tool was leaning Hansen on cold contact when most of your buyers want Russell. NewThe new prompt says it explicitly. Russell is the working register. Hansen is the relationship layer. The tool knows when to switch.
OldThe Russell pillars and Hansen principles were typed into the prompt itself. The prompt was huge and stale. NewRussell and Hansen each live in their own uploaded file. The prompt is short. Updating a framework means swapping a file, not rewriting instructions.
Both tools, ready for your review. About an hour on my end.
Five minutes. Nothing reaches Brandy until you sign off.
Both tools in her hands the same day you approve.
Your rewritten sales page.
A yes on using Christian Hansen's public posts.
The go-ahead to hand it to Brandy.