After the Build

Your timeline is probably filled with posts about Agentic development — adopting new workflows, getting agents to build software in as few prompts as possible, one-shot dreams and all that. There are even experiments looking to completely replace company workforces with fully autonomous, self-sufficient agentic systems that can go from product idea to marketing and everything in between. But for most of us, that’s not where we are. So this one’s for the normies.

I picked up agentic software development a while back, and my current favourite workflow is GitHub’s Spec Kit. I find it useful because it fits naturally into how I already think about building software — scoping out a feature, planning, clarifying before writing a single line of code. I’ve built three projects with it. And while I’ve gained a lot of speed and, dare I say, “productivity,” one thing looms large over every project I build: Operations and Product Ownership.

Before we all started adopting agents, we worked in teams — each person holding onto specific responsibilities. A typical team had UI/UX, Project Management, Product Ownership, Software Engineers, DevOps Engineers, and more. But when you’re building solo and prepping a product for market, you’re forced to wear every single hat.

I’ve been working on a personal project, and I’ve recently hit the point where I need to get users on it. I’m comfortable with the standard deployment side of DevOps, but I suddenly found myself needing to do Observability and Product Owner-type tasks too. I needed a way to measure and visually track how the infrastructure is doing and how users are behaving on the platform.

At this point, how I got here — using agents to build the thing — barely matters. What matters now is: is this application actually valuable to people? With observability, I can see how my servers are performing, whether I need to pay for more capacity, and if I do decide to upgrade, I’ve got the data to back that decision.

There’s also the challenge of cost. The project isn’t making money yet, and I don’t have spare cash to throw at it, so I need to squeeze as much value as I can out of what I have. That’s led me to dig into platforms like Grafana and PostHog.

I used to be part of a DevOps group at a previous job, and though I spent most of my time on pipeline work, I now wish I had paid more attention to observability.

One other benefit of living in the AI era: almost every product now ships with an AI assistant baked in. I was able to build a useful dashboard in Grafana using its embedded assistant — I described what I wanted to see, asked it to analyse my ingested logs, and got a working dashboard without knowing much about Grafana at all. That said, it doesn’t eliminate the need to actually understand the tool. The more you understand it, the more value you can pull from it.

So, what’s the takeaway? Agentic code is just the beginning. Operations, cost, and >growth — that’s where the real work starts.