From Idea to Digital Product: What Changes When Machines Work Alongside You as a Team?

Experienced developers expected AI tools to make them about 24% faster. The measured result went the other way: they were 19% slower, and yet afterwards they believed the tools had made them about 20% faster. This paradox comes from a 2025 study by METR, and it reminds us that a feeling of speed is not a measurement of it.

What do the experiments say?

No single study settles the matter, but together they paint a more accurate picture than either enthusiasm or pessimism.

In a controlled experiment published in 2023, with researchers from GitHub among others, programmers were asked to implement an HTTP server in JavaScript, and the group using the assistant finished 55.8% faster. It was a defined task with clear boundaries.

In the "Cybernetic Teammate" field experiment, 776 professionals at Procter & Gamble worked on real product-development challenges, and an individual using AI came close to the performance of a team working without it. The researchers also noted that the solutions became more balanced between the technical and commercial sides.

The METR study, by contrast, covered 16 experienced open-source developers and 246 tasks in mature projects, and there the tool proved a burden rather than a help. The small sample calls for caution in generalising, but it is enough to show that the benefit of a tool follows the type of task and the experience of its user.

From a tool you ask to a worker you assign

What is new in recent years is that these systems no longer merely answer; they can be assigned a chain of steps: they plan, call tools and check their own work. Anthropic distinguishes between a "workflow", in which models and tools are orchestrated through predefined code paths, and an "agent", in which the model directs its own process and tool use. It advises starting with the simplest solution possible and adding complexity only when needed, because an agent's autonomy means higher costs and the potential for compounding errors. For more, see the article "AI Agents: When the Machine Decides the Next Step for You".

Where does scarcity move?

When production becomes cheap, scarcity is likely to move from the ability to execute to the ability to choose well: which need deserves a solution, for whom, and at what quality? This is analytical inference, not the finding of any one experiment, and its logic is that the tool itself is available to everyone. Perhaps someone who knows their community and lives its daily concerns, whether families, students or charities, is better placed to see the real need than someone who starts from the tool.

Reservations before multiplying projects

Cheap to build is not cheap to run. Every product that lives on after launch needs support, updates, protection and promotion, and these burdens are not removed by the tool. The market is also likely to fill with similar products as they become easier to make, and no one will notice what solves no real problem. If your product collects user data, its safekeeping is your responsibility; for more, see the article "Privacy in the Age of AI".

It is also a matter of trust not to build what is impermissible in Islamic law or offends the values of society, and to be honest with your users about what your product promises.

What to do in practice

• Start from a real pain you have witnessed, not from a tool that impressed you.
• Keep the prototype small, and show it to real users before any expansion.
• Write down the steps of the work, delegate only what repeats, and review the output yourself.
• Time yourself with and without the tool, as the METR team did; impressions are not enough.
• Keep whatever touches data, money and sensitive decisions under your own review.

Reliance on God follows taking the means; it does not replace them.