People thought my recent post about AI was simultaneously taking a fanboy position, as well as a poopy-pants skeptic position. So let me reiterate what I really meant: I think that it's a fantastic tool for developers, but coding was never the bottleneck in shipping software. Folks immediately jumped in to say "I get 10x productivity" as well as "It's all overstated." I was saying that the impact to the business was not what you think.
A retired business leader that I greatly admire (theme park nerds know) recently said that, "If you are in the hospitality business remember some meetings are more important than others," above an AI cartoon photo of him meeting kids at a theme park. Obviously he was pointing out that we need reminders that the things that influence and move the needle for the business are often not the things that we think. That's where I meant to go with that AI post.
While I continue to call for better data about AI impact on coding in a business context, instead of a developer context (hammer/nail problem), I'll offer my anecdote. Right before people really got into agentic coding, my most recent team shipped a great service, in production, in about two weeks. Sure, we were doing a gradual rollout, but we got to that point by ruthlessly limiting scope, challenging assumptions and getting to the core of the problem that we were trying to solve. Productivity was increased by those means. Two weeks later, user feedback suggested that we got it very wrong (by we, I mean my engineering team and product team). That was totally OK though, we pivoted and did something that delivered a ton of value. Over the next year we tweaked performance and observability, and discovered a bunch of edge cases that we had to account for.
Going back to that first iteration though, two weeks was a huge win. I was so proud of the team. We got it wrong, but our process of shipping and then adapting was so fast. The truth is that agentic coding would not have changed that outcome. Coding was never the bottleneck. Making a product, and innovating, requires user feedback and human judgment and wisdom. The robots can't do that yet, if they ever can.
Software engineering is so much more than coding. Product development is so much more than coding. If you want to eke out productivity gains, sure, AI tooling will help, but the bigger wins will always come from the process that includes your stakeholders, users and product partners. We can't frame productivity in coding alone. We must frame it in terms of outcomes and the overall process.
No comments yet.