We use AI across Pezigns to research ideas, plan projects, write copy, develop designs, build websites, and check the results. That includes projects for our own business as well as our clients.
Our website and a client promotional-products project help explain how we use AI in everyday business. They show two different ways it can help.
The practical challenge is keeping client requests moving while I’m also making decisions about our own business. I still make the business and creative decisions and remain responsible for what we deliver. AI helps by gathering information, preparing drafts, organizing project details, and carrying out tasks I’ve approved.
How we organize our AI team
Our setup includes 60+ AI agents across several platforms. By “AI agent,” I mean a software assistant given instructions, tools, and a particular responsibility—such as researching a topic, drafting website copy, or reviewing a design.
I set the priorities and make the business and creative decisions. Some agents help organize projects and identify the next steps. Others focus on specific tasks, such as writing, design, or website development. Review agents check the results and flag problems.
We give agents names to make their responsibilities easier to follow. For example, Marcus is the name we’ve given our design-focused AI agent—not a human employee. For this article, Marcus prepared the illustration and revised its headline and subtitle.
Each project uses the agents it needs. A business getting started with AI doesn’t need a setup this large; one useful, repeatable task is enough to begin.
Before assigning a task, we explain what we need, provide the latest files, and note which decisions are already made and which still need my approval. That gives the agent a clear starting point and gives me a specific result to review.
What that looks like in practice
For our own website, I decide how we describe Pezigns, what visitors need to know, and how the site should look and function. AI helps draft the text, build the pages based on the direction I’ve approved, and check them for problems.
Once I approve a page’s design and wording, the agent building it needs the latest text, images, and instructions. The agent checking the page then needs that same version. If something is wrong, the review needs to explain what to change and where.
For websites and software, testing is part of development rather than a single check at the end. We define what needs to work, check individual features and how they work together, and review complete tasks a customer would carry out. The checks also address usability, accessibility, performance, and behavior on supported devices.
When a problem is found, we record it and identify who is responsible for resolving it. The proposed fix is reviewed independently, tested on the current version, and checked again alongside related features. That helps catch changes that solve one problem but create another.
We keep the issue, changes, review findings, and test results connected. Making a change is not the same as confirming it works in the version people actually use. Our process keeps that final verification—and anything still unresolved—visible before an issue is marked complete.
For one client project, we used AI to summarize a request for promotional products, prepare an email recap for my review, and organize the next steps.
We saved the recap as an email draft, added a link to the shared project folder in our project notes, and recorded the follow-ups needed for shipping and tracking. That gave me the customer’s request, a draft to review, and a clear list of what needed to happen next.
That was the preparation stage. The customer later confirmed receipt, and I closed the project. Our responsibility to the customer extended through the project, not just the reply.
These examples show different uses for AI: helping develop our website and helping organize a customer request from the initial message through to follow-up.
Keeping project details together
When a task passes to another person or AI agent, they need enough information to continue without guessing. We keep the original request, approved decisions, latest files, and next steps together.
For a writing task, that means explaining who the reader is, what we want to say, and which facts we can support. For a website page, it means providing the approved text and design, plus a clear description of how the page should behave.
We share only the information needed for each task and limit access to client material. If we continue a task in another AI tool, we need to bring the latest files and a note of what is finished and what still needs doing.
When I return to a project, I need to see what changed, what was checked, and what still needs attention.
Why permissions and records matter
We organize AI this way for more than convenience. We want agents to operate in a secure environment, with clear limits on which tools and information they can use and which actions need approval.
Our standard is that every agent action should be logged and accounted for: what was requested, which agent handled it, what it changed, and what was approved or checked. That record is an audit trail—a history we can review to understand what happened, investigate mistakes, and identify anything that still needs attention. It should capture the necessary details without copying confidential information into every log.
What I still need to decide
I want the agents to prepare drafts, designs, and project information so I can make informed decisions. An email draft can be ready for my review before I authorize sending it. A website change needs to be checked and approved before it goes live.
Agents can continue tasks I’ve already approved. Anything waiting on my decision needs to stay on the project’s to-do list, with a clear explanation of what I need to decide.
This setup also needs regular attention. Instructions can be incomplete, tools can fail, and a polished result can still miss what was asked for. We need to check whether a draft is accurate and whether a page functions as intended—not just whether it looks finished.
Putting it to use in your business
AI helps us turn requests and ideas into something useful: research summaries, email drafts, website pages, and organized project notes. I can review those results and make the next decision with the relevant information in front of me.
If you’re new to AI, start with one task you repeat regularly, such as drafting a customer reply from notes you provide. Decide what a good result looks like, what AI can do, and what a person must review or approve. Check the result against the original request. When deciding whether the process is useful, count the time spent reviewing, correcting, and keeping it up to date—not just the time spent producing the first draft.
Our white paper, From Prompt to Practice: Inside the Pezigns AI Operating Model, takes a closer look at how we divide responsibilities, keep project information together, and check the results.
Download From Prompt to Practice: Inside the Pezigns AI Operating Model (PDF, 76 KB)
AI-first
Design with
agency.
We pair years of design experience with AI tools that support our research, development, testing, and deployment.