Managing AI-Assisted Software Projects

The volume of code has exploded with AI

AI has made coding much faster, but the rest of the software delivery process hasn't necessarily sped up at the same pace. Software developers can now build several fully fledged features in the time it used to take to only write a few lines of code. However, someone still needs to coordinate that work, make sure everything is tested, and help clients review and approve what's being delivered.
In this blog, we explore how project managers can adapt to AI-assisted software development, drawing on insights from Sebastian Balderramo, an experienced engineering project manager at Ewents. We discuss why project management becomes even more important as development accelerates, how to account for AI-related rework when planning timelines, and which metrics help teams track progress, manage expectations, and maintain software quality.

Someone still needs to manage the work

One of the main advantages of AI in software development is the amount of time it can save. Based on our team’s experience, some tasks that could previously take 3-4 months can now be completed in just a few of weeks or even days.
However, while AI is great, it can still produce a large volume of code that could be completely useless if a senior engineer doesn’t correct or validate it. In Sebastian’s estimate, “60 to 70 percent of what it gives you is useful, and there’s another 30 to 40 percent you have to throw out. That 30 to 40 percent is exactly where you lose time.” So, while developers may save time during implementation, they still need to spend time checking whether the result actually works and matches what they were trying to build.

From the points above, the main implication is that AI is also changing how companies need to estimate cost and timelines of any software development project that is built using AI.

Estimating AI-assisted software projects

Let's start with the fact that a feature that previously took two days to develop might now take just a few hours with AI. That's a huge improvement, but writing code is only one part of delivering software. The engineering team still needs to review the code, make sure it works with the rest of the application, test different scenarios, and fix any issues before it goes into production.
So, when estimating AI-assisted software projects, project managers can't simply assume that faster coding means the entire project will take less time. They need to consider the full development process, including the time saved by AI, the additional work needed to validate its output, and how familiar the team is with the technology and existing codebase. Our team usually asks a few questions during the estimation phase:

• How much of the work can realistically be completed faster using AI?

• How much time will engineers and QA need to validate the AI-generated code?

• What happens if AI produces incorrect code or works with incomplete or inaccurate data?

• How familiar are engineers with the codebase, and how effectively can they use AI tools?

Poor data quality can also affect the estimate. As Sebastian warns, “If you give AI something that's wrong, it's going to keep giving you wrong results.” In one project discussed during our interview, incorrect inputs only became apparent during testing, forcing developers to revisit work they thought was finished and delaying delivery.

QA is still very important

QA is another important part of the process and, based on Sebastian’s experience, one of the roles that tends to be underestimated even as teams start using AI more heavily. Indeed, he is adamant about it: “For me, the role that should always, always be there is QA. I’m not ready to trust 100 percent that AI will discover every test case and get it right,” Sebastian says. “
For now, it has to be a person who does that validation.” The client still expects the product to work correctly, which means someone needs to test the functionality and identify problems before they reach production. Otherwise, as he notes, “errors pop up, or something breaks in production and you have to log on late at night or on a weekend to fix it.” Just for reference, GitHub’s guidance recommends reviewing and testing generated code, since suggestions can contain errors or security issues.
Another important point is that faster development can increase the amount of functionality waiting for QA testing. As Sebastian jokes, “you can be pushing a ton of code, a ton of functionality, and then you have 200 million bugs afterwards.” The team still needs to check whether changes work as expected and whether the business flows remain correct. In Sebastian’s latest project migrating databases to AWS, for instance, the client performs the final validation: “We deliver the data work, the client validates that everything is OK, and it goes to production.”

How to manage everyone’s expectations

Another challenge is managing changing client requests. Since AI allows developers to build features much faster, it's tempting to act on every new idea or request from the client. However, if priorities keep changing, developers may end up abandoning work they've already started or building features that aren't actually needed, wasting time and slowing down the project.
One of the lessons that comes from managing these projects isOne of the lessons Sebastian has learned managing these projects is not to react immediately to every request. Agreeing too quickly can mean asking the engineering team to work on something one day and changing direction the next because the client’s priorities have changed.“With complicated clients, first of all: patience,” he says. Early on, he admits, he was too reactive: “I’d say yes, yes, we’ll do that right away. So I’d put the team on one thing, and the next day the client would change course.”
Therefore, the project manager’s main responsibility should be is to understand the request before sending it directly to development:

• Understand what the client is actually trying to solve, rather than reacting only to the feature being requested.

• Keep priorities clear for the engineering team, especially when multiple ideas can now be implemented quickly.

• Avoid unnecessary rework, because faster development also means the team can move in the wrong direction faster.

Before adding a request to the development plan, the PM needs to clarify the problem it addresses and discuss how it affects current priorities. That conversation helps the team avoid building something that may need to change again the next day.

How to keep track of projects with clear metrics

Another challenge with AI-assisted development is keeping track of everything being built. As Sebastian explains, “before, a delivery was a login, and that’s it. Now it’s the login, it’s user registration, it’s a complete module in a week or two. The client often can’t fully absorb it, and that’s where the PM comes in.”
With so much more work happening at once, project managers need to make sure the team is actually making progress and delivering what was promised. Even with AI speeding things up, PMs can still use a few key metrics to understand how the project is going:

• Story points and sprint velocity: How much work can the team realistically get done in each sprint? These metrics help PMs plan future work and set realistic deadlines.

• Completed vs. committed story points: Is the team actually finishing what it planned? This helps identify when the team is taking on too much work or falling behind.

• Burn-down charts: How much work is still left in the sprint? These charts help spot delays early so the team can address them.

• Defects per sprint: How many bugs are being found? This helps PMs understand whether faster development is coming at the expense of quality.

As Sebastian explains, if the client assigns work directly to developers, the PM may not have full visibility into what's being done, making it harder to measure progress accurately. To ensure the success of the project, it is critical for everyone to follow these metrics closely.

Every dev team should have a project manager

If you read up to this point, we’d like you to leave with one insight: AI is helping developers build more software in less time, but that also means more work to coordinate, test, and deliver. Because AI is massively increasing the amount of code being pushed, project managers need to think more like Product Owners, understanding the client's business, setting priorities, and making sure teams are building the right things.
At Ewents, we believe project managers are now more important than ever. As companies try to cut costs and use more AI, reducing project management can end up creating more problems. AI can help developers build software faster, but without strong project management and QA, teams risk pushing out unreliable code across the organization.
If your organization is hiring managers to coordinate AI-driven development, contact Ewents today.

Frequently Asked Questions (FAQ)

• Do software teams still need project managers when they use AI?

  Yes. AI can help engineers build more functionality in less time, but someone still needs to coordinate that work, resolve bottlenecks, and keep priorities clear. The project manager also helps the client understand what is ready, what still needs testing, and how new requests affect the delivery plan.

• How should teams estimate AI-assisted software development?

  Estimates should include implementation, review, testing, and possible rework. AI may shorten a development task, but engineers still need to check whether the output works and meets the requirements. Estimates become more reliable as the team learns the system and understands where AI saves time and where it requires more supervision.

• How can project managers keep scope under control when AI speeds up development?

  Before turning a new request into development work, the team needs to understand the business problem behind it. Faster implementation can make it tempting to act on every idea, even when priorities are still changing. Discussing the request, its impact, and what it would displace in the current plan helps avoid unnecessary work.

• Is QA still necessary when developers use AI?

  Yes. AI can help generate tests and automate parts of testing, but the team still needs human review to check functionality, business flows, and cases that automated tests may miss. When developers produce more features in less time, there is also more work for QA to validate before release.

• Should AI-generated code go directly into production?

  AI-generated code should go through review and testing before production. Developers need to check their changes, and the team needs to validate that the functionality meets the requirements before including it in a release. GitHub’s guidance also recommends reviewing and testing generated code because suggestions can contain errors or security issues. Who handles final validation depends on the project and the responsibilities agreed with the client.

• How can poor data affect an AI-assisted software project's timeline?

  Incorrect or incomplete data can lead to wrong results and additional work. If those problems only become visible during testing, functionality that appeared finished may need to be reviewed and reworked. Checking data quality and business rules early helps teams identify problems that could otherwise delay delivery.

• Which metrics help track AI-assisted software delivery?

  Story points, team velocity, completed versus committed work, and defects reported per sprint can help teams follow progress and quality. These measures need to reflect the work the team actually owns and tracks. When assignments are split across several client teams, the project manager also needs visibility into dependencies and validation stages to understand what is ready for delivery.