My thoughts on agentic coding after one week.

Agentic Coding

Vibe Coding Agentic Coding is what everyone is talking about these days. Companies like Replit, X.ai and most recently Anthropic are raising billions on the promise of AI-driven development. If you believe everyone on social media, these companies are printing money. It’s common to see people talk about getting second and third opinions on answers provided by the premium models at each company.

What used to take a team weeks to spec, build and deploy, now takes an developer individual a few hours, maybe days.

The part that gets all the attention — launching the product — is the easiest and most fun part of all of this. The constant feedback between the agent and the user is addictive and possibly dangerous. But once the product reaches the end user and starts to gain traction, the real work begins: bugs, feature requests, scaling, operations, and maintenance.

Which raises the question.. What do you say to the person saying that we’ll all be out of a job in the next 12 to 18 months?

Thanks to these agents, it’s impossible to not see articles about how someone thought of an idea and launched it over a weekend. What I want to see is the six-month update. How is that individual handling the pressure of running this new business? How are they triaging bugs while also building new features? Are they generating revenue, or are they back on the VC treadmill where raising money is the measure of success?

Consider Jmail, the site that makes Jeffrey Epstein’s publicly released documents browsable. It went viral, crossed 450 million pageviews, and the Vercel hosting bill exploded to nearly $50,000. It illustrates a point that applies to every product shipping fast right now: building and deploying with agents is only the first chapter. Operations, cost management, and infrastructure decisions are the chapters that determine whether a product survives. Anyone using agentic tools to ship quickly needs to be thinking about that from day one. How good are these agents at orchestrating them, and how well do they handle the operational challenges that come with scaling?

I am certain that my echo chamber is too small, so if you have examples of products that were built and shipped via coding agents, and are still running 6 months after launch, please share them with me. I’m curious to see how these products are performing in the real world.

This is not a criticism of the tools. It’s an observation that we’re still in the “launch” phase of this wave, and the “operate” phase hasn’t been stress-tested yet. I’m excited to see how we solve these problems and what comes of it.