When people hear the word Codex, or AI coding agent, the first reaction is often simple:
“This is a tool for developers.”
I used to think the same way.
My work is Project Coordination. Most of my time is spent around requirements, timelines, tasks, test cases, documents, progress tracking, and connecting business, product, design, and development. I am not the person who directly writes code to build the product.
So when I started exploring Codex, my goal was not to use AI to become a developer. What I wanted to know was much more practical:
Can someone doing Project Coordination use a coding agent to reduce manual work and work more effectively?
After using it for a while, my answer is yes.
What surprised me most was that the biggest value of Codex in my work was not its ability to write code. It was its ability to read context, work across multiple sources of information, and carry out a sequence of actions until there is a complete result.
In other words, I stopped asking Codex for answers and started handing over pieces of work for it to execute.
From a Chatbot to an Execution Layer
Before Codex, my usual AI workflow was quite familiar: ask a question, receive a piece of content, copy it into a document, adjust the structure, and then continue the remaining steps manually.
With Codex, that workflow started to change.
Instead of asking, “What should a good ticket include?”, I can provide the requirement, the guideline, and the expected format, then ask Codex to analyze the requirement, write the ticket, and prepare it in the right structure.
If I had to describe Codex with a simple image, I would compare it to a junior member of the team. I give it a clear task, provide the context and the rules, then let it do the work.
Of course, I still review the result and remain responsible for the final output. But instead of doing every small step by hand, I can spend more attention on three questions: what needs to be done, what logic must be followed, and what the final result should look like.
This is also why AI adoption is not just about having access to a tool. As Linnoedge discussed in Why Employees Don’t Use AI: It Wasn’t Fear, It Was Time, the real change starts when AI becomes part of an actual workflow, not just something people occasionally open in a browser.
From Logging Plane Tickets Manually to Reviewing Results
One of the most practical use cases for me is task management on Plane.
After a discussion, or once a requirement has been agreed, I often need to create tickets with a clear title, description, acceptance criteria, priority, label, assignee, relationships with other tasks, and related references.
One or two tickets are not a problem. But a feature can be divided into many smaller tasks. When several features are being implemented at the same time, the work that looks as simple as “logging tasks” can start taking a significant amount of time.
There is also a small contradiction in this kind of work: it is highly repetitive, but it still requires accuracy.
In my current workflow, I can give Codex the requirement and guideline, ask it to analyze the content, split it into tasks, draft tickets in the defined format, and prepare the next steps on the system. My work shifts from opening each ticket and copying content line by line to reviewing a batch of results.
For a large number of tasks, the time saved is very clear. This is where I started to see Codex as an execution layer in my workflow, instead of only a question-answering tool.
Good tickets still start with clear requirements. Linnoedge has written before about how unclear requirements can create more difficulty than complex code itself, and that point becomes even more important when AI is helping prepare the work.
The Value Is Also in Keeping Everything Aligned
During product development, requirements rarely stay still.
Business may change the logic. Designers may update the flow. Developers may discover a new constraint. QA may find an edge case that was not mentioned before.
A small change can create a whole chain of follow-up work: update the ticket, adjust acceptance criteria, revise test cases, change documentation, and check whether related materials are still consistent.
If done manually, this easily becomes a long copy-and-paste loop. The more artifacts are involved, the easier it is to miss something.
With Codex, I can describe the new logic and ask it to review related content, identify what is affected, and propose or apply updates. A human still needs to check the logic, but I no longer need to spend most of my time finding and editing every affected line myself.
AI helps create new content. It also keeps what already exists consistent, and that is the part I had underestimated.
Tasks That Are Not Hard, but Can Consume Days
Test cases are a good example.
When a flow in the application changes, I may need to review a long list of existing test cases: which test cases are still valid, which steps need to be updated, which expected results changed, which edge cases need to be added, and which scenarios are no longer relevant.
Documents are similar. A single product change may appear in a requirement document, functional specification, test case, task description, and internal guideline at the same time.
Before, if the scope was large enough, reviewing and updating all of these materials could take several working days. Codex helps me turn that process into a more structured flow:
New context → review documents → identify affected parts → update → review the result.
The difference is not that humans disappear from the process. The difference is that humans spend their time confirming the logic instead of repeating hundreds of manual operations.
This matters especially for testing. As Linnoedge explained in Why all tests pass but code still breaks in production, a process can look green while still missing the real assumption that needed to be checked.
From Raw Project Data to Decision-Ready Information
Reporting is another use case where I find Codex useful.
A Project Coordinator often needs to answer questions such as: which tasks are blocked, which features are not finished, how many bugs are still open, which tasks are past the deadline, how the sprint is progressing, or how each team member’s workload looks right now.
The data may already exist in a task management system, but filtering, summarizing, and turning it into a readable report still takes time.
Codex helps shorten the distance between raw project data and information that can be used for decisions. I can ask it to collect data based on specific criteria and create a summary for a daily meeting, sprint review, or project update.
AI becomes useful when it turns existing project data into something the team can decide from.
This also connects closely with scheduling. A report does more than state status; it helps the team see risks early, just as good estimation and schedule ownership help development teams make better decisions before the deadline arrives.
Codex Helps Me Ask Better Questions to Developers
Beyond automation, Codex also helps me approach technical topics more clearly.
A requirement that looks simple from the business side is not always simple to implement. For example, in an application for mothers and babies, the requirement may sound like this: send a notification to remind the user to do an activity at the selected time.
Behind that simple sentence, many questions appear. What happens if the user changes time zone? What if notification permission is turned off? What if the activity is already completed before the reminder time? What if the user changes the reminder time? What if the same account is used on multiple devices?
In the past, I might have needed to ask developers quite early before realizing these points. Now, Codex helps me do an initial investigation first: analyze the requirement, identify dependencies, list edge cases, and prepare the right questions to clarify.
That does not make me someone who can review code. But it helps me understand the problem better and ask higher-quality questions when working with developers.
That quality of communication matters. Linnoedge’s article on communication mistakes in IT makes a similar point: vague updates and unclear questions slow the whole team down, while specific context helps everyone move faster.
What Changed Was Not My Role, but My Workflow
After my own experience, I do not think Codex will turn a Project Coordinator into a developer. I also do not think the goal of AI is to completely replace a role in the team.
What changed most clearly was the way work is divided.
Before, much of my time went into manual operations: creating and editing tickets, updating test cases, revising documents, summarizing reports, and checking data across different places.
Now, more and more of that work can be handed to AI to prepare first. My role gradually shifts toward:
Define the request → set the rules → provide context → review the output → make the decision.
This is also why I like comparing Codex to a junior teammate. If I give it a vague request, the result may be wrong. If I provide good context, clear rules, and a specific expected output, it can handle a meaningful amount of work.
A Project Coordinator does not necessarily need to know how to code to use Codex. But I think we will gain more advantage if we understand system logic, dependencies, workflows, and how to work with an AI agent.
The important skill is no longer only “knowing how to use AI.” It is knowing how to delegate work to AI: how to break down a problem, provide enough context, build rules, review the result, and recognize which parts can be automated and which parts still require human judgment.
For me, Codex is useful not because it helps me do a developer’s job. It is useful because it reduces repetitive work, helps me process a larger amount of information, and gives me more time for the parts that truly require coordination, thinking, and decision-making.
AI may not change our job title. But it can change a lot about how we do that job every day.
Want to put an AI agent into your team’s actual workflow?
Linnoedge is a software development company in Ho Chi Minh City. The way our coordinators work with Codex is the same way we help client teams put AI agents into real workflows, instead of handing people one more tool to open. Happy to talk through what that would look like for your team.
Book 30 minutes →
Chau Phan
PM / IT-Director · Linnoedge Inc. · LinkedIn↗
Soft girl in love with software. Rookie IT Coordinator turning hard-learned lessons into simple tips to make work accessible & enjoyable.
Bringing AI agents into non-engineering roles is part of what we cover in AI training for teams.