What a Delivery Lead Does When AI Takes the Transactional Work

Four years ago, I wrote a Spin post about my favorite meeting agenda tool. I was very excited to share it because it was a time-saver for me and I hoped other delivery leads would think so too. Reading it now, it reminds me where most of a delivery lead’s week used to go.

Notes. Agendas. Status reports. Contracts. I was good at the assembly work, and there was a lot of it. Then AI tools took most of it off my plate in the span of about a year.

That left me with a question I hadn’t expected. With newfound time on my hands, how could I best serve my client and my team? The answer depended on what my project was missing.

At Atomic, many projects get a product strategist and a designer alongside the developers. Mine doesn’t. The budget doesn’t allow it, which is a common reality in client work. Without those roles, developers end up carrying questions that sit outside their core craft. Is this the right problem? Which feature matters most to the business this quarter? Who is the user, really?

I don’t design UI or UX, and I’m not pretending to. What I can do is teach the ways of thinking those roles bring, so the developers can answer those questions themselves. That’s where the hours are going. It’s also where AI handed me a brand-new problem to solve.

The Work I Didn’t Miss

Meeting notes came first. In client meetings I was trying to give the client and the team my full attention while typing. After the meeting, I’d sort the notes into themes and send each piece to the right person. Then I’d update the backlog with the new discussion items.

Client status reports and engagement contracts came next. Every week I pieced progress together from tickets, chat threads, reports, and my own notes. Every new engagement or contract extension meant turning messy discovery conversations into clear contract language. Necessary, but assembly.

Now an agent captures the meeting, and I ask it for whatever cut I need. A summary by developer, by theme, by enhancement request. Reports and contracts arrive as drafts. I still read every word before a client does, but reviewing takes minutes instead of a few hours.

I’ll admit I was a little attached to that work. It was visible, and it was finished at the end of each week. I once wrote a whole blog post about using something besides an agent to write my agendas. I wouldn’t do that today.

What the team can do now that it couldn’t before

Design thinking came first. We define the user’s problem from their point of view before anyone proposes a solution. We ideate. Then we put a low-fidelity MVP artifact in front of the client for approval before the build starts.

That last step paid for itself quickly. Reviewing one artifact, the client pointed out a group of users we hadn’t accounted for. Better to learn that from a sketch than from a release.

User personas run through all of it. A persona turns an argument about opinions into a conversation about a person.

Then I introduced the value vs. effort matrix. Developers use it when they plan feature roadmaps for the subsystems they own. Each feature gets plotted by how much the client’s business values it right now and how much work it will take. High value, low effort goes first. Low value, high effort gets questioned out loud.

One pair of developers plotted the enhancements they’d generated in a design thinking exercise. The matrix surfaced a handful of small, high-value items that had been buried under bigger work. They shipped those first, and the client got wins right away.

That’s the part I care about most. The business conversation now happens in the developers’ hands. They don’t need me in the room to ask whether a feature is worth it.

The problem AI handed us

Not everything AI brought was relief.

I’d like to say I spotted it in a dashboard. I didn’t. It showed up in standups and retros, the same frustration in different words. Reviews were piling up, and people felt they couldn’t get to their own work. AI was producing spec documents and code faster than ever, and every one of them still needed a developer to read it. The writing got faster. The reading didn’t.

So I took the design thinking we’d been using on client problems and pointed it at ourselves. We ran a workshop. The first job was defining the core problem, and it wasn’t the number of PRs. Nobody knew how deep a review needed to be, so everyone reviewed everything as if it were critical.

Then we rewrote each pain point as a How Might We question. “Every review takes forever” became “How might we decide which changes need a deep review and which need a lighter pass?”

Afterward, the team agreed on which changes get a deep review and which get a lighter one. Not everything gets the same scrutiny anymore. No tool was going to notice that problem for us, and no tool was going to fix it.

What I’ll write in four more years

If AI is giving you time back, notice where it goes. It’s easy to spend it on more coordination and more meetings. On a lean project, the better use is helping the team think more clearly about the problem in front of them.

Conversation

Join the conversation

Your email address will not be published. Required fields are marked *