Skip to content
Install
Back to Customers

Automating the SDLC, one agent at a time

WEX-logo
Industry
Fintech
Location
Portland, ME
The Story

Automating the SDLC, one agent at a time

The software development lifecycle (SDLC) at any company is full of opportunity to innovate and streamline, but that’s easier said than done. It is inherently a series of handoffs that require thorough review with plenty of context. Yet WEX sees each stage as ready for reinvention. WEX is now building toward a fully automated SDLC: a connected pipeline of AI agents covering requirements, code generation, review, security, and deployment, with humans reviewing direction and intent rather than implementation detail.

Bill Mayhew, VP of Benefits Platform Engineering at WEX, was on a product call when the team realized they needed a deep dive into a specific backend process in order to move forward. Normally, that would have meant contacting a lead architect and scheduling a breakout session.

Instead, Bill opened the codebase and ran a prompt.

What came back was a plain-English description of exactly how the process worked — its objectives, the workflow, and what it was accomplishing.

He shared it with the Product Manager on the call. From there, it went to the product lead, then to a live demo for the WEX Health & Benefits team, and eventually to a session for the entire product organization.

It was clear that Augment wasn't just a coding tool; it was a way to make complex system logic accessible to everyone — from engineers to leadership — at a much faster pace. By streamlining how the Platform Engineering teams synthesized information, it opened up a bigger opportunity: Where else could they use this technology to accelerate the delivery cycle.

Where it started: context at scale

A global leader in intelligent payment solutions, WEX processes billions of dollars in fleet, health, and benefits payments annually. For a team managing this level of technical complexity and scale, the priority was finding a tool capable of reasoning across the entire environment.

When Bill started evaluating AI coding tools in early 2024, the key question wasn't capability in the abstract. It was whether any tool could understand their code and reason across a large, interconnected system.

"We have very large, complex codebases,” Bill said. “We started asking questions about the codebases from Augment. It became really apparent the power of the tool in terms of context and its ability to decipher what's going on."

For a team actively refactoring applications into a new united platform, codebase awareness was the prerequisite for everything that came next.

What automation has already unlocked

That first product call wasn't a one-off; the pattern held under pressure.

WEX’s claims processing system — the backend handling FSA reimbursements and benefits submissions — is a complex environment. A major refactor of the webforms within that system was initially scoped as a seven-to-nine-month project, but WEX decided to test a different approach.

They gave an engineer five days, a focused prompt workflow, and Augment’s codebase context. By the end of day five, the results were in: new RESTful APIs mirroring the existing system's behavior, a connected UI already in QA, and full test coverage validating that every database output matched old to new.

Something that was going to take a team the majority of a year to move forward— one engineer was able to make significant progress in just five days, Bill said. I would call it an outstanding result. We shaved months off our time to test with customers, without impacting the WEX quality they’ve come to expect.
Bill Mayhew

Bill Mayhew

VP of Benefits Engineering

The pattern generalized further. Outside of major project roadmaps, Bill and team were finding additional efficiency gains by using Augment to optimize systems and address a spectrum of requests made by partners, who build more directly with WEX to provide Consumer Driven Health offerings. Like any large-scale engineering organization, WEX manages a high-volume roadmap where people are prioritized for core product innovation.

Using Augment, WEX engineers can now address partner requests and secondary backlog items with unprecedented speed. In one instance, a developer fed a specific, long-standing request into Augment with the relevant codebase context. The tool identified the root cause and generated a suggested fix with full test coverage in minutes.

This shift allows WEX to prioritize high-detail maintenance work. It empowers the team to clear the long tail of requests while keeping the focus on the primary product roadmap.

An agentic SDLC, built piece by piece

The software development lifecycle (SDLC) at any company is full of opportunity to innovate and streamline, but that’s easier said than done. It is inherently a series of handoffs that require thorough review with plenty of context.

Yet WEX sees each stage as ready for reinvention.

WEX is now building toward a fully automated SDLC: a connected pipeline of AI agents covering requirements, code generation, review, security, and deployment, with humans reviewing direction and intent rather than implementation detail.

Stage 1: Product intake

A product manager submits a use case. An agent breaks problem statements down by domain mapping, identifying which teams own which capabilities, defining accountability boundaries, and preventing scope sprawl. It outputs epics, stories, and tasks formatted to a standard that downstream agents can consume.

The key design decision: ticket quality is upstream of everything else. WEX is deliberate about formatting standards so that work landing on a team's backlog is already structured for Augment to work with.

Stage 2: Code generation

Engineers open VS Code, reference a ticket, and Augment handles code generation. The context pipeline gives Augment what it needs to write with architectural awareness, not just local file context.

Stage 3: Code review

The code review agent is live and being evaluated. WEX is tracking comment acceptance rates, types of issues caught, and noise reduction. It gives engineers a first pass that cleans up the obvious issues before a teammate sees the PR. What they're still learning: where the agent adds signal versus where it generates friction. That calibration is ongoing.

Stage 4: Security and QA (in design)

An InfoSec agent running static and dynamic application security testing checks automatically on every PR, feeding into a QA agent that validates expected behavior. WEX already uses Snyk within VS Code for in-editor feedback; the plan is to formalize a security gate that catches what escaped in real time.

Stage 5: Deployment (on the roadmap)

Deployment agents close the loop. At that point, the human role in the pipeline is to write the intent clearly at the top, review the result at the bottom, and trust the system to handle everything in between. While WEX works thoughtfully toward that model, Bill remains clear-eyed about the path ahead.

Fully autonomous SDLC — PR to production. That's the North Star. Knowing that the reality is it'll be a while before we get there. But without stating that as an objective, we'll never make progress. So we're just going to chip away each piece of the SDLC as much as we can.
bill-mayhew-headshot

Bill Mayhew

VP of Benefits Engineering

How WEX is measuring it

WEX tracks Product Innovation Velocity (PIV), which is time from ideation to customer delivery. While Bill is tracking these efficiency gains through broader velocity metrics, the most compelling evidence of Augment’s impact is found in these day-to-day wins: completing a multi-month refactor in less than a week, resolving complex partner requests in a single afternoon, and empowering product managers to navigate sophisticated system flows independently

WEX's advice: start with the cases where the delta is unambiguous. Build tracking along the way.

Try it yourself

The patterns WEX is using aren't proprietary. Here are three you can run today.

1. Make a legacy system legible

If you've inherited a codebase and need to understand a specific process before you can change it, or if a PM needs to understand what a backend actually does, this is the fastest on-ramp.

markdown
Prompt:
You are a Technical Product Owner that needs to understand how <workflow_name> workflow works within this legacy code base.
Analyze the codebase and structure the analysis with these outputs:
1. Plain English description of the workflow and dependencies
2. A Mermaid diagram of the workflow
3. Snippets of the code including filenames that are relevant
4. 4. Suggestions on how to improve the workflow
5. A markdown file summarizing the above

What to expect: A detailed description of the workflow with the outputs requested, workflow description, objective summary, plain-English narrative. The difference from just reading code is significant: Augment synthesizes intent across the system, not just syntax within a file.

2. Scope a refactor before you commit to one

Before estimating a migration, use Augment to map what you're actually dealing with. Feed it the legacy code and ask it to identify the surface area, dependencies, side effects, implicit contracts, before writing a single line of new code.

markdown
Prompt:
You are an expert software engineer helping me refactor code safely and efficiently.
The refactor requires a need to upgrade <third party library> from version <x.xx> to <x.xx> due to <reason for upgrade>. Analyze the code and suggest the refactoring plan.
1. Identify breaking changes in detail.
2. Describe the risk associated with the upgrade.
3. Call out where the upgrade may introduce defects.
4. Create the migration path steps with an expected timeline.
5. Create atesting plan to validate that the upgrade doesn't impact existing functionality.

What to expect: A detailed upgrade path identifying associated risks, an expected timeline with explicit steps, and tests that ensure the upgrade doesn't impact expected behaviors. Note what it gets right and where human judgment is still required.

3. Close a long-tail bug ticket with full test coverage

Take a bug report or a screenshot of a JIRA ticket, feed it to Augment with the relevant codebase context, and ask for a diagnosis, a fix, and tests that prevent regression.

markdown
Prompt:
You are a senior engineer that was presented with a defect from a stakeholder.
The reported defect is:
<defect description>
1. Identify the root cause of the reported defect and make a suggested fix.
2. Suggest the fix to prevent the bug while preserving intended behavior.
3. Call out risks associated with the suggested fix.
4. Create tasks necessary to address the defect and the test cases to catch this bug and tests for the additional fix for my review.

What to expect: Augment identifies the defect clearly and provides a task list to remediate along with complete test coverage, including tests for the suggested fix. The value is in reviewing rather than writing from scratch.

Get Started

Give your codebase the agents it deserves

Install Augment to get started. Works with codebases of any size, from side projects to enterprise monorepos.