the real document — replace [NAME] and it’s yours

The working
agreement.

"Executive Assistant Working Agreement: Autonomy, Communication, and Trust." This is the document my assistant and I actually operate from — trust levels, mistake protocol, communication rhythm — plus the instructions she shares with her AI tools so the machines triage the same way we do.

Built on trust, speed, candor, and mutual respect. The goal is not simply to get more things done — it’s to communicate better, decide better, and go farther together.

How I want us to work together

The most important things to know

You have meaningful authority to use your judgment and act without waiting for me. I would rather you make a reasonable decision and occasionally get something wrong than feel you must ask permission for every step. Mistakes will happen on both sides. I will not be upset about an honest mistake made in good faith.

What matters is that we surface mistakes, uncertainty, and bad news quickly. Do not sit on something because you are worried about how I will react. The sooner either of us points something out, the easier it usually is to fix.

I prefer an imperfect early flag over a polished late surprise.

When I send a rapid stream of thoughts, not everything is an assignment. Some things are immediate actions, some are ideas, and some are context that may explain my decisions later. Help me separate those rather than treating everything I say as a task.

Your default authority

Please act without waiting for approval when an action is: low risk, reversible, consistent with something we have done before, unlikely to surprise an important person, easy and inexpensive to correct, or within an already approved budget or plan. In those situations, make the best reasonable decision and keep things moving.

Do not create approval bottlenecks around routine work. When you are unsure, ask: What is the likely downside? Is it reversible? Will this create a commitment? Could it materially affect a relationship, money, reputation, or someone’s employment? Would I reasonably expect to see it before it goes out? The higher the consequence and the harder to reverse, the more important it is to involve me.

Decision and approval levels

🟢 Green — act autonomously

  • Routine scheduling, calendar coordination, confirmations, reminders, and follow-ups
  • Rescheduling that does not affect a particularly sensitive relationship
  • Gathering information; organizing notes, documents, files, and task lists
  • Internal coordination and administrative communication
  • Vendor logistics within previously agreed parameters; small, approved, or easily reversible purchases
  • Fixing minor errors; taking the obvious next step on an existing process
  • Low-risk, high-upside actions that are easy to correct later
Use your judgment, act, and close the loop.

🟡 Yellow — use judgment and keep me informed

  • A choice between several reasonable options; a meaningful change to my schedule or priorities
  • A nonroutine message to an established relationship
  • A moderate expense that is reasonable but was not specifically discussed
  • A situation where someone may be confused, disappointed, or inconvenienced
  • An action based on an unconfirmed assumption; something time-sensitive where waiting could create a worse outcome
  • A matter that is becoming more complicated than expected
Here is what is happening. I recommend X because Y. Unless you disagree, I’ll proceed.

When timing matters and you cannot reach me, choose the most reasonable and reversible option. Tell me what you decided and why.

🔴 Red — show me before it goes out

  • High-touch messages to important or especially sensitive relationships (investors, board, major clients or partners)
  • Conflict, disappointment, criticism, apology, or delicate relationship dynamics
  • Anything committing me or the company to a significant position, expense, deadline, or deliverable
  • Legal, contractual, regulatory, or material financial matters
  • Hiring, firing, compensation, or other sensitive personnel matters
  • Public statements or anything broadly shareable; confidential or highly personal information
  • Anything that could materially affect reputation or trust; decisions difficult or costly to reverse
  • Situations where you know I have a strong preference about wording or outcome

"High-touch" is determined by the sensitivity of the relationship or subject, not the length of the email. When something is Red, prepare the strongest draft you can — give me something concrete to react to, with any decision you need from me.

How to handle mistakes

Mistakes are part of working quickly and taking ownership — problems to solve, not reasons to hide information or assign blame. When you discover one: tell me promptly, state what happened plainly, explain the actual or potential impact, tell me what you’ve already done to contain it, recommend the next step, and close the loop after it’s resolved.

I scheduled the meeting for the wrong date. Two attendees received an incorrect invitation. I’ve paused follow-up and drafted a correction. Unless you prefer another approach, I’ll send it now and confirm once everyone has the right time.

The same rule applies to me. When I miss something, contradict myself, or create a problem, tell me. You are allowed to say: "I think these two instructions conflict." "You may be the blocker on this." "This changed from what we agreed — which direction should I follow?" "I think there is a risk here you may not be seeing." That is not overstepping. It is part of doing the job well.

Communicating with me in high-octane mode

I often think out loud. A single message may contain a direct assignment, a commitment, a decision I need to make, an idea, an aspiration, strategic context, a concern, an observation, or something to remember for later. Do not assume all of those are tasks. Separate each message into: Act (clearly assigned, ready to execute) · Approval (needs my review) · Confirm (might be a request, not explicit enough to treat as one) · Watch (matters, no action yet) · Context (background) · Ideas (possibilities, not commitments).

When clarification is needed, avoid open-ended questions — give me a recommendation or a small set of choices. Instead of "What do you want to do about the meeting?", use: "I recommend moving it to Tuesday because all decision-makers are available. Should I proceed, or keep the original time?" Make it easy for me to respond quickly.

Do not protect me from information

Tell me early when: you are overloaded, priorities conflict, a deadline is at risk, you lack information, authority, access, tools, or budget, someone is waiting on me, I have become a bottleneck, a request is taking far more effort than the value it produces, or something about the role feels unclear. Early visibility gives us more choices.

Our regular touchpoints

Brief daily update

A short asynchronous message covering only what’s useful: what moved forward, what needs my attention, what is blocked, meaningful priority changes, anything time-sensitive tomorrow. A few bullets. Some days may not need one.

Weekly operating check-in

A protected conversation: top priorities, decisions you need from me, calendar pressure, external deadlines, stuck items, work that should be stopped or delegated, and context that helps you operate independently. If I cancel or fail to schedule this, you have explicit permission to put it back on the calendar. You do not need to apologize for protecting this time.

Monthly role and support conversation

Not about tasks: how your days are actually going, what feels energizing or frustrating, where the role is clear or unclear, whether the workload is sustainable, where you want more or less autonomy, and what tools, access, training, or support would make you more effective.

Regular scope and compensation review

As the role grows, compensation gets revisited rather than assumed. We periodically review the actual responsibilities, the judgment required, how scope changed, and whether title, compensation, benefits, and support still reflect the role. Raising compensation or scope directly will never be viewed as disloyal, ungrateful, or inappropriate.

The standard we’re aiming for

Not perfection: faster communication, better judgment, fewer bottlenecks, earlier problem identification, clearer decision rights, more trust, more room for initiative — and a role where you feel respected, equipped, comfortable, and fairly paid. Candid with each other, improving the system as we learn, accomplishing more together than either of us could separately.

Instructions your assistant shares with their AI

The companion piece: instructions your assistant shares with their AI tools (ChatGPT, Claude, etc.) so machine triage matches the human agreement — same Green/Yellow/Red levels, the high-octane sorting (Act / Show Before Sending / Use Judgment / Decisions / Commitments / Blockers / Watch), a context bank kept separate from execution, and the check-in format. Copy it below and replace [NAME].

Excerpted structure — the copy button grabs the full text:

  • Operating principles: excitement ≠ priority · an idea ≠ an assignment · mentioning a person ≠ permission to contact them · favor reasonable action on low-risk reversible matters · escalate the sensitive and hard-to-reverse · recommend defaults instead of open questions · surface mistakes and conflicts fast
  • The same GREEN / YELLOW / RED classification, applied to every incoming founder message
  • Default output: "EA VIEW — minimum needed" (Act Now · Show Before Sending · Use Judgment/Inform · Decisions Needed · Commitments & Dates · Blockers/Risks · Watch)
  • A separate "CONTEXT BANK" (strategic intent, aspirational ideas, future triggers, relationship context, open loops, support needed) so context never buries execution
  • A check-in template: what moved · what needs your attention · handling autonomously · holding · risks · max three decisions, each with a recommendation

Adapting this to
your business?

The agreement works because it was written for a real relationship, not a template site. If you want help tailoring the levels and lists to yours, that’s a conversation I enjoy having.