Skip to content
KatSquare
AINews

We put AI agents inside the build. Here is what actually changed.

The timelines on our process page just got shorter. This is the honest account of where the time went, where it didn't, and what the senior people do now that the first draft of everything is free.

Published

09 Oct 2026

Reading time

2 min

By

KatSquare

For most of this year every agency on earth has been saying it uses AI. Most of them mean a copywriter with a chat window open. We mean something narrower and more useful: coding agents that work inside our repositories, against our tests and our linting rules, with an engineer reviewing every line before it merges. This post is about what that did to a real project timeline, because we just rewrote the one on our process page and we owe you the reasoning.

Where the time went

Discovery used to be a fortnight because reading is slow. A competitor teardown means opening forty sites, screenshotting them, and writing up what they do and don't do. An analytics review means exporting a year of data and staring at it. Agents now do the first pass of both in an afternoon, and the senior person's job becomes checking the summary against reality and deciding what it means. Two weeks became two to four days, and the written point of view at the end is better, not worse, because the thinking time was not spent on the collecting.

Build is the big one. Our engineers now work in one-week increments instead of two, because the boring half of every ticket, the migration, the form validation, the test scaffolding, the third API client this month, is drafted by an agent while the engineer is still reading the spec. The engineer's time goes to the part that needs a human: the data model, the edge cases, the performance budget, the review. A sixteen-week build is now a six-week build with the same people and the same standard.

Where it didn't

Define got shorter by days, not weeks, and we are fine with that. Positioning, information architecture and product requirements are the decisions that cost the most to reverse later, and an agent cannot sit in a room with a founder and notice that two directors want different businesses. We refuse to rush that phase, and we would be suspicious of anyone who says they can.

Launch is not faster either, and should not be. Core Web Vitals, accessibility, redirects, analytics and error tracking are a checklist, and checklists were already fast. What changed is that the audit is thorough every time, because the tedious parts are automated and nobody skips them at ten at night.

The first draft of everything is now free. Taste, judgement and accountability are not, and that is what you are paying for.

What we will not do

  • Ship code no engineer has read. Every merge has a human reviewer, and their name is on it.
  • Let an agent talk to your customers unsupervised. Drafts, yes. Sends, no.
  • Quote the old timeline and pocket the difference. The shorter phases are on the process page now, in public, so you can hold us to them.
  • Pretend the model is the product. The model is a very fast junior. The product is still what the senior people decide.

If you want the long version for your own project, the discovery call is the same as it ever was: an hour, free, and we tell you what we would do before you have signed anything.

MoreAlso from the studio
Rather talk than read?

We'll give you the opinion directly.

A 30-minute call, no deck, no pitch theatre. You get a written point of view whether or not you hire us.

Direct
Typical response

Within one working day, IST (UTC+5:30).