> ## Content Index
> Fetch the complete content index at: https://www.abnerballardo.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# As a CIO, I Used an AI Agent to Make My Communication Guidance Reusable
- URL: https://www.abnerballardo.com/en/post/as-a-cio-i-used-an-ai-agent-to-make-my-communication-guidance-reusable/
- Published: 2026-10-26T12:00:00.000Z
- Updated: 2026-10-26T12:00:00.000Z
- Description: I made recurring communication guidance available through an AI agent so my team could keep preparing messages while I was occupied.
- Author: Abner Ballardo
- Tags: Tech Leadership, Organizational Systems

[Leer en Español](https://www.abnerballardo.com/es/post/como-cio-puse-mis-criterios-de-comunicacion-en-un-agente-de-ia/)

When I was CIO of a bank, many of the messages my team wrote for other executives crossed my desk first. Some drafts could have used more context, or a clearer explanation of why the issue mattered to the person reading it.

Through my own work with those executives, I had learned what each of them needed to understand, what background they already had, and what helped them weigh a request. Much of that reached my team one draft review at a time.

That made my team dependent on my availability. I had other responsibilities, and my feedback didn't always come quickly. Meanwhile, people could be left wondering whether they had missed something, explained it wrong, or written something that would only create confusion.

I built an AI assistant to put that guidance within reach while the message was still being written. My team could use it to challenge a draft and keep going while I was busy with something else.

I'm now sharing a public reconstruction of it, **Executive Communication Preparation**, as part of [Agentic Work Patterns](https://github.com/AbnerBallardo/agentic-work-patterns?ref=abnerballardo.com), my collection of patterns for applying AI to recurring work inside organizations.

## The guidance inside my draft reviews

Correcting a fact helps with one case. Knowing how much background a particular recipient needs helps with many.

So I wrote that recurring guidance down as recipient profiles: what each person already understood, what needed explaining, and what evidence or level of detail worked for them. With those profiles, the agent had something specific to work from.

The profiles captured what I had learned from working with those people. A job title alone couldn't have told the agent how to help.

In an [earlier article](https://www.abnerballardo.com/en/post/as-a-cio-i-used-an-ai-agent-to-prepare-decisions-not-make-them/), I wrote about an agent that helped my team think an issue through before bringing it to me. This second agent covered a different part of the work. Once I had looked at the situation and given direction, how could the team write to someone else using what I had learned about that person?

**The answer was to reuse what carried from one case to the next: knowing what each recipient needed to understand.**

## The agent needed more than a draft

To help with a message, the agent had to know who would read it and what that person needed to know. It also needed the team's account of the situation and the direction I had given.

The pattern keeps those three inputs separate. The recipient profile describes how to explain the work. The team brings what it knows today: facts, evidence, options, and open questions. And whoever gave direction (in my case, me) says what the message should achieve, what has actually been decided, and whether the draft has to come back for review.

With that context, the agent can challenge an explanation before polishing it. It can ask why something matters, what backs up a claim, or what the reader is actually being asked to do. Checking the facts is still the team's job. A profile can't tell the agent what is true today.

## A fictional regulatory requirement

Here is a **fictional banking example, informed by my experience**.

A bank receives a regulatory requirement with a fixed deadline. Business and technology teams assess the work. Their proposal needs more investment and people who are already committed elsewhere. Meeting the deadline means deciding on funding and priorities.

The team first uses a decision-preparation agent to lay out the issue for the bank's CIO: options, recommendation, consequences. The CIO reviews it and concludes that the investment and the shift in priorities are beyond his authority. He asks the team to write to the bank's CEO and to the regional CIO, and to bring both drafts back to him before anything goes out.

His guidance about these two recipients already existed before any of this came up. As the team works on each message, it brings the agent the relevant profile, its assessment, and the CIO's instructions.

A first draft might say:

> We need additional investment and revised priorities to meet the regulatory deadline.

For the CEO, the CIO's guidance ties any request for investment to existing business plans. That sentence doesn't make the connection. Which commitments would move? What would that mean for plans the CEO already owns? When is a decision needed?

Those questions give the team something concrete to work on. The next version has to say which commitments are affected and what follows, along with what is being asked: the investment and a decision on priorities.

The same sentence leaves the regional CIO with a different question: what does the bank need from the region? Here the guidance asks for an explanation of regulatory exposure and technology dependencies. The agent can ask which dependency needs regional support, what intervention is being requested, and what still hinges on the CEO's decision.

With those answers, the team can sharpen the request. Where its assessment comes up short, it has to find the answer or leave the uncertainty in plain view. The agent shouldn't invent an estimate, a commitment, or an approval just to finish the message.

The CIO still reviews both drafts. But while he's busy, the team keeps working from expectations he has already written down, and he doesn't have to explain them again. The next case can reuse the same guidance, with its own facts and new direction.

Neither message can drop what matters: the requirement and its deadline, the resources needed, the commitments that move, and what is still uncertain. The CEO needs to understand the regulatory exposure. The regional CIO needs to see which business commitments would be pushed aside. In this example, complying isn't optional. What's open is funding and priorities, and each executive decides within their own authority.

## Would a written guide be enough?

Writing the guidance down would already help. A guide or template may be enough when the task is familiar and the person writing has all the information.

The reason to try an agent is the back-and-forth. In the example, someone could fill a "business impact" field with a vague line about shifting priorities. A template would accept it. The agent can ask whether that tells the reader enough, and help work out what is still unclear.

I can't claim this design works better than training, templates, or another kind of review. It can also miss a gap or suggest an explanation the evidence doesn't support. Its value is in the thinking it prompts.

## What still depended on me

The independence I wanted for my team was practical: keep working on a message without waiting for me. I could hand some of my guidance to the agent and keep the judgments that really needed me.

I did see people get familiar with how the agent approached communication. I didn't formally assess whether that learning lasted or how people reacted more broadly, and I didn't measure time savings or productivity.

If I had asked to see a draft again, it came back to me. If the situation or the proposed commitment changed, the team might need new direction. And answering the agent's questions never meant permission to send.

The person writing the message still owned its facts and wording, and a person, not the agent, chose the channel and sent it. The pattern I'm sharing also spells out a safeguard for the profiles themselves: the agent can't change a recipient's profile on its own because of one conversation. Any change needs approval from whoever owns that guidance.

## Start with the correction you keep making

If you're a leader thinking about trying this, I'd start with one correction you keep making. Write down the need behind it and who it applies to. Then try that guidance on a fictional case, with some information left out on purpose.

You don't have to start from scratch. The [Executive Communication Preparation](https://github.com/AbnerBallardo/agentic-work-patterns/blob/main/patterns/executive-communication-preparation/README.md?ref=abnerballardo.com) pattern includes the agent's instructions, a profile template, and fictional examples. It's a reconstruction informed by my field experience, separate from the private version I used, and its [evaluation summary](https://github.com/AbnerBallardo/agentic-work-patterns/blob/main/patterns/executive-communication-preparation/evaluations/README.md?ref=abnerballardo.com) is clear about what has and hasn't been tested.

Watch what the agent asks before it writes anything. The real test is whether its questions expose what still needs explaining, without hiding uncertainty or skipping a review that's required.

**Which of the corrections you keep making could your team apply with an agent's help, and which judgments still need you personally?**