How software architects and project managers can leverage agentic AI

You may have heard of vibe coding. But what about vibe software architecture design, or vibe project and technical management?

I’m here to tell you that teams can “vibe” these aspects of software development, too. Using AI agents, software architects, project managers and technical managers can accelerate their workflows, just as developers are now using AI to help with code implementation and integration processes.

Allow me to explain by detailing how the teams I work with are already putting agentic AI technology to work in areas like software architecture and project management.

Extending agentic AI beyond development roles

Agentic AI is the use of software agents — meaning programs that can carry out tasks autonomously, with guidance from AI models — to automate complex tasks.

Starting about a year ago, software developers began cluing into the benefits of agentic AI as a way of streamlining tasks like writing and testing code.

And now, other stakeholders within the software development process are starting to realize that they can take advantage of AI agents, too. For example:

• Software architects can use agentic AI to design application architectures and components. AI agents can generate design plans much, much faster than humans. They are also adept at picking up on details (like data transit and encryption requirements, for example) that humans might overlook when planning a complex architecture.
• Project managers can leverage AI agents to formulate plans that describe who will do what during a software development project. Agents can also create resource dependency lists and timelines. Here again, agents can complete these tasks much faster than humans could achieve on their own, while also factoring in a wide variety of considerations that may be overwhelming for a human to manage manually.
• Technical managers, whose main job is to align technical plans with business priorities, can take advantage of AI agents to generate insights such as estimates for project scope and budget.

To be clear, I’m absolutely not suggesting that AI agents can replace humans in these roles. Software development teams will still need architects, project managers and technical managers for the foreseeable future. But by using agents to kick off and automate workflows, these stakeholders can work faster and at an increased level of scale. For example, it’s reasonable to expect AI to succeed in generating a software architecture that is 80 percent complete and accurate, significantly reducing the time that a software architect has to spend reviewing and updating plans manually.

The challenges of agentic AI for software architecture and management

As is the case when using AI agents to help with coding, software architects, project managers and technical managers should expect to run into some special challenges when integrating agentic AI into their workflows.

One is that AI can make inaccurate assumptions, especially when the humans who guide it don’t include complete details within prompts. If the result of work completed by AI agents is inaccurate, architects or managers will need to tweak their prompts and try again. Indeed, iteration is key to getting AI agents to produce efficient, reliable designs and plans.

A second challenge — and one that is impossible to solve through simple iteration — is agents’ lack of awareness of bureaucratic obstacles that may apply to a particular organization. For example, an agent might assume that it can modify DNS records automatically. But at some companies, DNS changes require a human to issue a formal request, which must, in turn, receive review and approval. This is an example of why humans must review AI-generated software architectures and project plans and update them based on context that AI doesn’t know about.

Real-world examples of agentic AI in software architecting and project management

As I mentioned, my organization is already using AI agents to help with software designs, project management and technical management. Here are two recent examples of initiatives we completed with help from agentic AI:

• Implementing security log management on Azure: A client wanted to centralize the ingestion and analysis of security logs on the Azure cloud. Using AI agents, my team scoped the project and generated a list of roles and skill sets (including solution architect, IaC engineers and data engineers) required to complete it. The agents also generated a preliminary architecture for the system, along with an estimated budget. While we needed to adjust the plans following manual review, the major benefit of using agentic AI was that we were able to get all stakeholders on the same page quickly, achieving a level of seamless coordination that tends to be elusive when working with multiple people representing diverse roles and parts of the business.
• Technical editor agent software: To streamline internal processes, we decided to create a “technical editor agent” capable of reviewing folders of technical documentation content for clarity, accuracy, consistency and usability. Here again, AI agents were able to generate a very detailed technical specification, including an internal API design. They also created a project effort estimation and a preliminary task list for implementing the system. As a result, we had a complete project plan within hours, instead of the days or weeks it would have taken to plan for even a small-scale tool like this without the help of agentic AI.

AI agents and the future of software development projects

Agentic AI remains a fast-changing domain, and how teams leverage AI agents will no doubt continue to evolve. I think of the examples I’ve mentioned above as “proofs of concept” showing that AI agents have an important role to play beyond coding, rather than the be-all, end-all of what agentic AI will do when applied to multiple facets of the software development process.

Nonetheless, the results we’ve already achieved by leveraging AI agents to help create software architectures and manage software projects highlight the importance of thinking holistically about how technical teams leverage agentic AI. The headlines still tend to focus on vibe coding, but clearly, agentic AI is not just for coders.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?