Skip to content
hidra.Blog · engineering and applications
EngineeringIdeas and perspectives5 min read

When building gets easier, what becomes worth building?

AI invites us to rethink small teams, purpose-built tools and interfaces: how might it change what we choose to design and use?

Different tools and interfaces emerge from a shared engineering core
An idea can take different forms: an interface, a working tool, a physical system.

The ideas we leave on the shelf

Many businesses have a small collection of tools it would be useful to have. A screen that guides a procedure. A way to bring together data from different systems. A device that saves someone from repeating the same operation. These ideas often remain notes: the benefit seems too narrow to justify the work needed to build them.

Artificial intelligence introduces an interesting possibility. When it helps us explore, implement and revise a solution with less effort, it can shift the threshold at which an idea becomes worth trying. Even a need shared by only a few people might justify a carefully designed response.

The most interesting change may concern that selection process: which ideas begin to look feasible? Which needs do we finally notice after treating them as too small to deserve a project?

A small team can hold a broader project together

A complete system crosses several disciplines: electronics, firmware, communications, interfaces, data and testing. For a small team, moving between these layers involves learning, coordination and implementation work. A substantial part of the effort goes into connecting the pieces.

AI tools can take on bounded tasks, prepare alternatives and help bridge different environments. This gives a compact group an opportunity to handle a broader scope while keeping decision-makers close to one another.

An engineer can devote more attention to the relationships between components: what should happen, which information is needed, where responsibility passes from one part to another, and how a correct outcome can be recognised. They can guide implementation without personally writing every line or building every screen.

Expertise remains essential to understanding and evaluating the output. Critical decisions require knowing when to involve a specialist. The organisational question becomes: which capabilities need to stay closely coordinated for the system to retain a coherent direction?

The prototype becomes part of the conversation

When an exploratory version is easier to prepare, a discussion can begin with something people can use. Someone tries a sequence, pauses at an unexpected point or asks for different information. The prototype reveals needs that an initial description may have missed.

This can also change the relationship between designers and future users. Requirements develop through examples and observations. Two ways of carrying out a task can be compared before committing to full development.

A convincing demonstration and a reliable everyday tool still require different kinds of work. Field testing, error handling and maintenance need a place in the project. A useful prototype makes those decisions more concrete.

Interfaces that follow the work people do

Many applications ask people to adapt to the program’s structure: open forms, find data and transfer results. If a dedicated view becomes less expensive to build, we can try starting with the task itself.

Imagine a technician checking a machine. One screen could bring together the procedure, relevant measurements and the record of the outcome. An operator might see only the information needed for a particular shift, with controls that fit the situation.

These are design possibilities to explore with the people doing the work. A conversation may suit an open-ended search; a guided sequence may serve a procedure better; a physical button may be the right choice for a frequent action. The interaction deserves to be chosen alongside the function.

AI can help design these interfaces while remaining in the development tools. The result can run on conventional software, deterministic commands and defined behaviour. This is another way products can change through AI.

Products can take more specific forms

A product intended for many users must reconcile different needs. A tool for one particular activity can choose a narrower scope and fit that situation more closely. If preparing it takes less work, some highly specific solutions may become more accessible.

We might imagine a small service connecting two existing systems, a panel dedicated to one department or a device that performs one function well. Its value may lie in continuity between things that already exist, fewer handovers and ease of use.

This perspective also raises a practical question: who will look after all these different tools? Development cost is one part of the decision. Updates, dependencies, data access and the ability to make changes over time help determine whether a solution is worthwhile.

Easier building makes choosing more important

We can produce many variants quickly and still make life more complicated for their users. More development capacity calls for a criterion: which activity becomes clearer? Which step disappears? Who can now do something that previously required an intermediary?

Productivity also needs to be examined in context. In its February 2026 developer-research update, METR describes selection and measurement difficulties that limit the interpretation of its results. A single percentage says little when the people, tasks and kinds of work being attempted change. METR’s methodological update helps explain this difficulty.

For design, this suggests a further question: are we carrying out work already planned, or exploring solutions we would previously have dismissed? Those activities call for different ways of assessing the outcome.

Which ideas now deserve an experiment?

This is a useful starting point for a series of reflections: observing how teams, interfaces, business tools and the expectations of technology users change.

For a closer look at tasks, tools and autonomy, Anthropic’s guide to agentic systems distinguishes software-defined paths from processes in which a model decides how to use tools. It is a useful distinction when considering which activities to delegate and which controls to retain.

The possibility of building more easily invites us to look again at everyday work: procedures people work around, spreadsheets passed between colleagues, information copied repeatedly, and requested functions that were never built. Some needs may deserve an experiment; others may have a simpler answer than new software.

Which tool had we stopped imagining because it seemed too demanding to build?