← Back to insights

From Autocomplete to Teammate: How AI Is Reshaping IT

In a few years AI went from an autocomplete feature to a teammate that can finish whole tasks. Here is what we see from the desk of a small studio.

In short

  • AI moved from completing a line of code to doing whole tasks.
  • Writing code got much cheaper. The bottleneck moved to decisions, review, and consistency.
  • Small teams gain the most, as long as they have tidy working habits.

A few years ago, the most magical thing a coding assistant could do was guess the next line. We remember how it felt: tab, tab, tab, and a function appeared without being typed. Like a small magic trick.

Now, on an ordinary Tuesday night, we write one paragraph of instructions, close the laptop for dinner, and come back to a branch that already holds the changes, the tests, and a note on why each decision was made. Those two moments are only a few years apart.

Four steps up in a short time

From the user’s seat, the jumps came one after another:

  1. Autocomplete. Guesses the next line or function while you type.
  2. Chat. You ask, it answers and writes a snippet. You paste it in.
  3. Agents. It reads the project, runs commands, edits many files, then checks its own work.
  4. Agent teams. Several agents share the job: one researches, one writes, one tests.
↗ More and more done by AI itself1Autocompleteguess a line2ChatQ and A3Agentdo the task4Agent teamsplit roles
Each step is not just faster. It changes the kind of work people do around it.

On the first step people write and the machine helps. On the third, people explain and check. It sounds small on a brochure, but it is big in daily life.

The bottleneck moved

The main obstacle used to be typing and searching. Now that writing code is far cheaper, the obstacle has moved to things that used to hide behind it: deciding what is worth building, judging the result, and making sure today’s ten tasks follow the same style as last month’s ten.

Writing code got cheaper. Judging code did not.

The industry is starting to find words for this too. Anthropic, for example, writes that the agent patterns that work in practice are usually simple and composable, not elaborate frameworks (Building Effective Agents). There are also small standards such as AGENTS.md: one file that tells an agent how to work in a project.

Why this is good news for small teams

A small studio like ours used to lose on headcount. Now part of that gap has closed. Two founders can look after a product that once needed a larger team, and still talk to the client directly. There is a catch: speed without habits just produces faster mess.

What we do differently

  • Write project rules in one file the agent reads at the start of every session.
  • Split tasks into small pieces that are easy to check.
  • Run tests and lint automatically before a person sees the result.
  • Keep design decisions in writing, not in someone’s head.

We cover the last two habits in the next article: skills and harnesses for consistent AI results. And the question we hear most, will AI replace engineers and designers, has its own piece.

If you are planning a digital product and want to know how we use this way of working, see our services or tell us your idea.

Start a project

We reply within one working day.

What do you need?
Budget