How SPS Commerce Built an AI Software Factory on Port

Lead Software Engineer Mark DeBeer had just finished building SPS Commerce's software factory in Port. He labeled a Jira ticket and watched as a planning agent drafted the implementation, a developer approved the plan, and a coding agent wrote the code and opened a pull request. 

"It dawned on me that this is a new way of working that I have never seen in my career before," he said. "The possibilities of what we could do with Port were endless. I could see our backlog shrinking."

In its first four months, SPS Commerce’s agentic ticket workflow processed more than 500 tickets with zero incidents, proving the foundation for a scalable factory. The workflow is now expanding to support agentic use cases across more of the SDLC.

What should you know about SPS Commerce?

SPS Commerce is the world's largest retail network. The company connects roughly 50,000 retailers, suppliers, and logistics companies worldwide, with a platform powering more than $650 billion in annual transactions. If you've shopped at Walmart, Kroger, or Amazon, there’s a good chance SPS has been part of that transaction somewhere along the line. 

Keeping that network running takes hundreds of engineers across more than 3,000 services, and a developer productivity engineering team of about five people owns the Port implementation that supports them. At that scale, repetitive work like dependency migrations, bug fixes, and schema changes piles up faster than teams can clear it.

SPS saw an opportunity to move toward a software factory: an engineering model where developers and AI agents work together across connected, automated workflows to deliver software at scale. 

What did it take for SPS to move from isolated agents to a software factory?

As AI coding tools took hold, SPS teams began building their own agents for various use cases, including coding, planning, and migrations, each solving part of the software development lifecycle independently.

“A lot of teams were working in isolation, building out their own agents to meet their own needs, which is great,” Mark explained. “But in order to evolve beyond that, we needed to be able to not just reuse these but connect these agents to span the entire SDLC.”

To move toward a software factory, SPS needed to connect agents into cohesive, governed workflows, and give agents access to the same context as developers while maintaining visibility and control across the SDLC. 

How did Port become the horizontal foundation for SPS’s software factory?

SPS evaluated multiple platforms and chose Port because it met all its requirements:

“We were looking for an AI software factory platform, and we realized that Port hit all the major things we needed. It had a centralized context lake, the governance, and the orchestration so it tied all the processes and information about what was happening in the SDLC together. It had dashboards so that we could see the ROI and monitor what was happening. And it was a single source of truth for both agents and developers, which makes it easy to debug and understand what's happening with these complex agentic workflows,” explained Mark. 

What does an agentic workflow look like inside SPS’s software factory?

SPS designed its first agentic workflow the way it would onboard a junior engineer: agents draft plans and write code, and developers review key decisions before work moves forward. The workflow removed repetitive work while keeping developers accountable for the outcome.

The first workflow centered on the Jira ticket-to-pull-request journey. A developer labels a Jira ticket, and Port sends it to a planning agent. The agent drafts an implementation plan using Port’s Context Lake, which brings together data from the tools developers already use, including Jira, Azure DevOps, and Kubernetes. The developer reviews and approves the plan through a Port self-service action, and Port hands the approved plan to a coding agent, which generates the code and opens a pull request in GitHub. The developer then reviews and merges the PR, remaining accountable for the outcome. Behind the scenes, Port orchestrates each handoff, posts status updates to Slack and dashboards, and collects developer feedback to help the team understand where agents excel and where they need improvement.

For Mark, that first ticket moving end-to-end through the agentic workflow demonstrated something bigger: repetitive work could increasingly be handled by agents, freeing up developers to focus on more complex engineering work.

What results is SPS seeing from its AI software factory?

In the first four months, SPS demonstrated that agentic workflows could scale across real development tasks:

• 500+ Jira tickets processed through SPS’s agentic ticket workflow
• ~400 story points of work processed
• Zero incidents related to the workflow
• More than 50% of developers who used the workflow reported it as helpful, demonstrating early adoption in the new way of working

For SPS leadership, zero incidents was particularly important. SPS monitors agent workflows through Port alongside its existing telemetry and incident-management systems, giving the team visibility into both agent performance and operational impact.

3 key findings: What SPS learned building its software factory

Connected workflows scale better than isolated agents. Individual agents can automate individual tasks, but a software factory requires connecting agents, developers, and tools into end-to-end workflows. Orchestration turns isolated AI use cases into a system that can work across the SDLC.

Shared context makes agents more effective. Giving agents access to the same real engineering context as developers helps them plan and act with a more complete understanding of the SDLC.

The “junior-developer” framework tells you where humans belong. SPS treated agents like junior developers, placing human approval where engineers already review work: before implementation and before merge. This keeps developers accountable without adding unnecessary friction. 

What’s next: How SPS is expanding its software factory across the SDLC

What began as a ticket workflow is becoming a blueprint for SPS’s broader software factory. The team is expanding into PR review and build-failure use cases, with agent-to-agent workflows also on the horizon, including observability agents that could eventually invoke coding agents directly. 

“We’ve proven that agents work well for planning and coding,” Mark said. “From there, we’re expanding out further in the SDLC.”

Sign up for the Port Platform here.