I'm going to indulge myself a little with stories of grandeur. It'll sound a little braggy, so I apologize in advance. There is a point though, I promise.
The team in my last gig was completely awesome. Certainly they would have to be if I were to stay with them for four years. Not working with them is the hardest part of the RIF. What made it a great team is that we stayed ahead of everything put in front of us. Instrumentation was there from the start, performance was always meeting SLA's, and maybe most importantly, our backlog was fairly shallow most of the time. It's not that we didn't have tech debt, because you're always making certain compromises, but we didn't have a lot of it. We made mistakes early, iterated and delivered the better thing as fast as possible. And we did it without crunches or code freezes or exhausting everyone. My product partner was totally aligned, and while we debated things freely, there was never conflict or tension.
And even as we were sometimes sent on wild goose chases in the last six months, our core projects never got out of control. The team was using AI agents quite a bit by then. There were some PR's that churned a bit during this time, as the team was still feeling out the appropriate level of trust in the robots, and understanding what they were putting out, but I equate this to onboarding a new, junior engineer. That backlog was still not super deep.
As I've said many times, coding as a part of the entire SDLC is tiny relative to everything else. Deciding the "what" and "when" is a lot more work than the "how." Engineers should spend way more time understanding context, making design decisions, documenting, testing, reviewing each other's code and collaborating with other teams. Agentic coding frees up time to allow engineers to do all of that stuff, which makes them better equipped to direct the AI. Time spent coding, or making the machine do it, is even further reduced.
But as I said, that time savings is a sliver of the bigger cost. Include all of the stakeholders, and the what/why dwarfs the how. AI does not create a shortcut for this. It does create an opportunity.
My team was so effective because the what/why was so dialed in. My product manager and team worked so hard to understand the bare minimum thing to build, did it, and iterated. Freeing up the build time meant that we could focus more on the long-tail work in the backlog. The last system we built never went down in two years, and always met SLA's in terms of response times. It got even easier to stay on top of that when coding took less time.
This brings me to the bigger point. Some folks are so enamored with the shiny thing that they disregard outcomes. Yes, you can build faster, but what are the outcomes? If you shipped broken things in constant need of fire drills, that won't get better with AI. LLM's only mirror what they've seen, and most software isn't very good. Great, you have more PR's or more lines of code, but was that ever a useful measurement for humans? (Of course not!) The biggest opportunity with AI is that we have more time to get it right, address tech debt and keep that backlog tight. Use your found time to make your product robust and resistant to constant emergencies.
No comments yet.