Where the Time Goes When You Build With Claude Code
October 31, 2026 · 2 min read
Most of the time on an AI-assisted build doesn’t go to writing the code.
Generating is the fast part
Claude Code will put together a working dashboard faster than I can finish describing what I want on it. That part is real. It’s also the part that looks impressive in a demo, which is why people talk about it.
What it can’t notice
It can’t notice that the ad platform stamps a conversion on the click date while the CRM stamps it on the close date, so those two numbers were never going to reconcile. It can’t notice that the reason last quarter looks flat is a UTM somebody typo’d back in March. Neither of those is a coding problem. They’re problems with what the code is being fed.
From the outside, a working dashboard looks finished. The charts render, the totals look plausible, and nothing prompts anyone to look closer. A plausible number is the dangerous kind. A broken chart gets noticed within a day, and a figure that’s quietly a bit off can sit in front of an owner for months.
That’s where the time goes: getting the inputs honest, deciding what a “lead” means everywhere the word shows up, and writing the checks that catch it when a source quietly changes shape underneath you.
The split
On my own builds I’d put it at roughly a quarter of the time generating and the rest on plumbing and validation around it. That’s an estimate, not a measurement, and it moves with how messy the sources are. A clean export shifts the balance toward generating, and a CRM that three people have filled in three different ways pushes it the other way.
The checks themselves are small: a test that totals match between two systems, an alert when a source’s columns change, a rule that flags any lead with no source. None of them are exciting to write. Plumbing is the part that decides whether anyone can trust the output six months from now, and it’s the part that doesn’t get posted about.
What that phrase tells you
Which is why “we built it with AI” tells you almost nothing about whether the thing is any good. It describes how fast the first version appeared. It says nothing about whether the numbers reconcile, whether “lead” means one thing everywhere, or what happens when a source changes next quarter.
For the builders here: where does the time go on your projects?
Written by Michael Masner
Building AI-powered marketing and data systems for small and mid-sized businesses.
Get in touch →More Insights
Thirty Days Is the Wrong Window
October 29, 2026
The platforms report monthly, so a month becomes the unit. For a business with a long decision cycle and a busy season, thirty days says as much about timing as it does about performance.
Read more →
The Calls Nobody Answered
October 27, 2026
For a service business, the most expensive part of marketing can be the calls nobody answered. Those leads show up nowhere in your reporting, and they're usually the cheapest thing to fix.
Read more →
Vague Briefs Get Vague Marketing
October 23, 2026
Most of the conversation about AI is about which tools to use. The step before any of them is knowing what you want clearly enough to say it, and that applies to a campaign brief as much as it does to code.
Read more →