AI agents in the studio's work: what gets automated and what stays with people

AI in development is discussed as if it will either replace everyone or is useless. In practice it's more boring and more useful: AI takes on part of the routine, while responsibility and decisions stay with people. We work with agents every day and want to show what that looks like without promises.

Where AI really saves time

• ·Drafts of texts and documentation: feature descriptions, instructions for the client, answers to typical questions. A person edits the meaning instead of typing from scratch.
• ·Breaking down what comes in: a long email or spec turns into a list of tasks with questions that need clarifying.
• ·Tests and checks: the agent suggests which cases should be covered and finds forgotten branches of logic.
• ·Routine code edits: renames, types, repeated patterns — things it's a shame to spend a specialist's time on.
• ·Log and error analysis: a failure turns into a clear description of "what happened and where", not a wall of stack trace.

Where we don't hand the decision to a machine

• ·Architecture and technology choices: the cost of a mistake is too high to make it on a suggestion.
• ·Estimating time and money: an estimate is a commitment to the client.
• ·Priorities: what to do first and what can wait — decided by a person and the business context.
• ·Security and personal data: what can be sent outside and what can't is a rule, not an improvisation.

How this relates to price

Routine that used to take hours now takes minutes — which means fewer billable man-hours in a project. This isn't a "discount because of AI", it's honest economics: we spend less time on mechanical work and more on what affects the result.

If a contractor promises "an AI website in an hour" — ask what exactly is automated. Usually it's the sale that's automated, not the work.

What's worth automating for you

• ·If you answer the same questions every day — that's the first candidate for a bot or a knowledge base.
• ·If data is moved by hand from a form into a spreadsheet and then into an invoice — that's an integration, not "one more employee".
• ·If a report is assembled by hand once a week — that's automation: the data is already in the system, it just needs to be reconciled.

Automation doesn't start with AI. It starts with the question "which operation repeats most often". AI plugs in where text needs understanding, not where you need a copy machine.