Everyone’s building AI agents. Fewer people are asking the question that actually determines whether those agents work: Who are you, and what do you care about? It turns out that’s the hardest problem and the most important one to solve first.
You’ve watched the demos, read the LinkedIn posts, maybe even started a free trial of something. You’re ready to automate, to delegate, to finally get out from under the inbox. And then… you stall.
Because the first question the agent needs answered — who are you, and what actually matters to you? — turns out to be the hardest one in the room.
This is the insight that gets buried under every conversation about AI adoption. We talk about tools, platforms, prompts, and workflows. We debate Claude versus ChatGPT, cloud versus local, Telegram versus Slack. And almost nobody talks about the thing that determines whether any of it works: the quality of self-knowledge you bring to the table before you touch a single setting.
Before getting into the files, it helps to understand what you’re actually building. AJ Thomas, executive coach and agentic AI builder, uses a biological metaphor that makes the whole thing less intimidating.
Every agent needs an exoskeleton, a shell to live in, either in the cloud or on your local machine. It needs a brain, the LLM (Claude, Gemini, GPT) that does the reasoning. It needs bones and muscle, the configuration layer that governs identity, memory, and behavior. It needs arms and legs, a messaging interface like Telegram or Slack where you actually interact with it. And it needs hands, the API connections that let it reach into your email, calendar, and other tools and take action on your behalf.
The part most people skip is the soul. Not metaphorically, literally: a set of documents that tell the agent who it’s working for and what that person actually cares about. That’s the layer that turns a capable tool into a capable delegate.
AJ Thomas has built sixteen AI agents, twelve of them for coaching clients. When she walks someone through the process, she doesn’t start with platforms or APIs. She starts with three documents she calls the soul file, the identity file, and the user profile — plain text files that answer, in specific and personal terms, who this agent is supposed to serve and how.
a→ The soul file holds your operating principles, the values and standards that should govern how your agent acts on your behalf.
→ The identity file gives the agent its voice, character, and disposition.
→ The user profile covers the practical reality of your life: your schedule, your priorities, your most important relationships, the things that drain your energy, and the communication patterns that define you.
None of this requires technical skill; it requires reflection.
Before one of the most demanding weeks of her professional calendar — managing a major conference, multiple speaking engagements, and an advisory board launch simultaneously — Thomas spent significant time loading her agent, Atlas, with exactly this kind of context. Her latest 360 review, her core leadership principles, her ten most important contacts, her quarterly goals, the two projects that mattered most right now.
The result? Atlas booked her a $5,000 speaking engagement she’d missed in her inbox, secured commitments from six of nine target advisors, and coordinated her calendar in real time while she was on stage. Not because the technology was sophisticated, but because the technology knew enough about her to act like her.
There’s a reason AI adoption rates remain stubbornly low despite the hype. It’s not skepticism, it’s not a lack of access. It’s that most people are trying to implement a tool they haven’t yet personalized, and an unpersonalized AI agent is just a faster way to get generic output.
The instinct when building an agent is to give it everything, every skill, every integration, every capability available. More inputs, more power. But an agent without a clear hierarchy of priorities and a defined sense of purpose doesn’t become more capable when you load it up. It becomes paralyzed because capability without character isn’t an upgrade.
What the agent needs, what any delegate needs, is a clear mandate. And a clear mandate requires you to have one first.
This is where senior leaders, and particularly high-achieving women, actually have a significant advantage. Years of leadership development, executive coaching, 360 reviews, values clarification work, all of that is raw material. Your Enneagram type, your communication preferences, your decision-making frameworks, your hard-won understanding of how you work best: these aren’t soft extras, they’re the architecture of a capable agent.
One of the more quietly radical ideas to emerge from this conversation is a reframe of how we talk about AI oversight. The standard advice is to “keep the human in the loop” as if the default state is full AI autonomy that we’re carefully dialing back. Thomas inverts this entirely.
The human is always in the loop. The question is when AI enters yours.
This shift matters more than it might seem. The first framing positions humans as a check on AI. The second positions AI as a tool humans intentionally deploy. One is defensive. The other is architectural. And if you’re trying to build something that actually works for you — not just something that runs — the architectural mindset is the only one that gets you there.
If you’re waiting until you understand Railway and Homebrew and OAuth scopes before you begin, you’ll be waiting a long time. The technical layer is learnable, and much of it can be delegated. What can’t be delegated is the work that happens before any of that.
Start with a document. Write down your values, your priorities, your communication style, the things you never want to miss, the tasks that consistently drain you, the goals you’re working toward this quarter. Write down what “acting like you” actually means.
That document is your soul file. It’s also, not coincidentally, the kind of clarity that makes every other form of delegation human or AI dramatically more effective.
The technology is ready. What it needs from you isn’t expertise — it’s the kind of deep self-knowledge that great leaders have been building their entire careers. You’re more prepared for this than you think.
