What 'good' looks like across the delivery lifecycle

Most conversations about AI at work focus on a single role or a single task. A developer using it to write code. A tester using it to generate scenarios. A BA using it to draft user stories.

That's useful. But it's not the full picture.

When you zoom out and look at how AI is being used across an entire project, something more interesting emerges. Not a tool bolted on at one stage, but a thread running through the whole thing, from the first conversation about what needs to be built right through to the documentation that follows delivery. This is what it looks like in practice at Optimation.

Discovery and specification

The work starts before a line of code is written. Jeshua Hertzke, our Engineering Team Lead, describes a shift in where thinking time goes:

"The heavy lifting now goes into precisely articulating what needs to be built and why, because AI handles execution well, but only when given the right context."

The upfront investment in clarity and the thinking that used to feel like overhead has become the most valuable thing a person can contribute. AI is being used from early discovery on Miro boards through to specification authoring, with the human's job being to give it enough direction to produce something worth using.

Development

Anthony Brandon, one of our senior developers, describes a workflow that would have been difficult to imagine a short time ago. He starts by having Claude read a user story or bug directly from the project management tool. Claude understands the issue, then codes a solution. Anthony reviews and tests it, they iterate together, and at the end, the same AI that wrote the code also writes the unit and integration tests, documents the changes, updates the user story with a summary, and creates wiki documentation if required.

"The work we do these days is far more complex than the past, because we can now do it with AI. Some of the stuff we are adding now would have been parked in the 'it would take too long or is too hard' basket. But not anymore."

He's also building toward something further: using MCP servers so that AI can test changes in a browser, log in as a test user, take screenshots, and review console output, iterating until a fix is confirmed. The goal is a closed loop where AI handles the full change-test-validate cycle for appropriate tasks.

Testing

On the test side, the picture is similar. Aaron McLean describes a shift in his role from document creator to reviewer:

"I am no longer creating the document. I am more reviewing them and making changes where applicable."

He applies an 80/20 rule, with AI generating the bulk of the test scenarios and his expertise shaping the final 20%.

He's also doing things he couldn't do before. Six months ago, creating automated test suites in Visual Studio Code wasn't part of his toolkit. Now it is.

Neena D'Souza leads Optimation's test team and takes a clear-eyed view of what AI enables and where the limits are. The team uses Claude and Copilot to generate test scripts, documentation, and wiki pages. Efficiency has improved noticeably. But the validation step stays: "AI can occasionally introduce invalid scenarios," Neena notes, so thorough review remains part of the process. The speed gain doesn't come at the expense of quality control. It comes because quality control is still there.

Documentation

Across every stage, documentation is the thread that ties it together, and it's where AI is often doing some of its most consistent work. Anthony Brandon has it writing unit and integration test documentation, updating acceptance criteria, and creating wiki pages as part of the same workflow that produces the code. Aaron McLean is producing a level of technical documentation for clients, including detailed automated test suite write-ups, that would have been slower and harder to achieve before.

This is the pattern: AI doesn't replace the delivery. It handles the recording of the delivery, freeing people to stay focused on the work itself.

What this actually means

The reason this matters isn't efficiency for its own sake. It's that the scope of what's achievable has shifted.

Work that was parked because it was too complex or too time-consuming is now on the table. Documentation that would have been thin or rushed is now thorough. Test coverage that would have required more resource is now within reach.

But Jeshua puts his finger on the thing that holds it all together: AI has become pervasive across the delivery lifecycle, from Miro boards to Azure pipelines. "That's actually the biggest misconception I see," he says.

"People think adopting AI is the hard part. It's not. The real challenge is directing it with enough domain knowledge and governance to get correct outcomes, not just fast ones."

The technology is accessible. What's harder to shortcut is the expertise needed to use it well. Every example in this post, Aaron knowing which 20% to change, Neena insisting on validation, Anthony reviewing code before accepting it, Jeshua investing in precise articulation at the front end, comes back to the same thing.

Good outcomes don't happen because AI was involved. They happen because the right people were directing it.

If you're thinking about where AI fits into your own delivery process, or where it should, we'd like to talk. Get in touch.

Contributors:

• Jeshua Hertzke | Engineering Team Lead

• Neena D'Souza | Test Lead

• Anthony Brandon | Senior Developer

• Aaron McLean | Senior Test Analyst