The future of development started yesterday

The world of development is not going to change tomorrow. It is not even changing today.

It changed yesterday.

As recently as last year, a large share of AI-assisted development demos consisted of generating an interface: a landing page, a dashboard, a few cards, a sidebar and two charts.

It was impressive, but it often remained a visual demonstration.

Today, what I am seeing is different. Complete games are starting to emerge in a few hours or a few days. Specialized tools are being developed for a single need. Editors, engines, level generators, physics systems, 3D tools and business applications are appearing constantly.

There are now too many of them for me to reasonably list them all.

The real shift: accessing knowledge and putting it into action

What seems revolutionary to me is not, ultimately, that artificial intelligence can write code.

It is the gradual disappearance of the delay between a question and its execution.

I want to understand how to move an object correctly around a sphere? I can immediately access the relevant vectors, quaternions, transformations and methods, then test them directly in my game.

I want to add physics? The engine already exists, its documentation exists, the mathematical principles exist, and the agent can help me connect all of that to my architecture.

I want to build a tool specific to my project? I no longer necessarily need to spend several weeks learning every detail before getting a first working version.

The knowledge already existed. What has changed radically is its operational accessibility.

First, we built tools for building

From the start, I thought an initial phase would be devoted to this: using LLMs to create tools that would then make it possible to create other things.

That phase is now firmly established.

DCCs, or Digital Content Creation tools, no longer necessarily have to be huge general-purpose applications. We are seeing much more specialized tools emerge: scene editors, asset generators, procedural creation interfaces, placement tools, texture pipelines or environments built around a single project.

And now, some of these tools directly integrate an agent capable of contributing to their use.

OpenAI is pushing this logic quite far with Codex. The tool is now presented as an agent capable of handling an engineering task end to end: modifying a codebase, developing a feature, performing a migration, running tests and preparing the result for review. Source: OpenAI — Codex.

With Symphony, OpenAI goes even further in orchestration. In the internal project that led to this tool, the team says it imposed a particular rule: the repository had to be built without code written directly by a human, with every line generated by Codex. Symphony was then developed to orchestrate several agents around the work to be done. Source: OpenAI — Symphony.

It is probably the next tool I am going to test.

3D makes what is happening especially clear

In video games, the shift is becoming spectacular because many of the building blocks already existed.

Three.js has long made it possible to build a 3D world in a browser. Engines such as Cannon.js or cannon-es provide physics. The mathematics required for trajectories, collisions, cameras and procedural generation have been documented for years.

But previously, using all this knowledge correctly required a tremendous amount of time.

Today, it can be assembled almost at the pace at which problems are formulated.

Tools capable of directly producing the elements needed for a 3D world are now being added to the mix.

Microsoft is developing TRELLIS.2, for example, a pipeline capable of transforming an image into an exportable 3D asset. The project notably offers GLB outputs and different generation resolutions. Source: Microsoft — TRELLIS.2.

A project such as img2threejs takes a different and particularly interesting direction: reconstructing the object visible in an image as readable, editable procedural Three.js code, rather than simply producing a mesh file. Source: img2threejs.

We are therefore moving from a world in which we spent hours looking for the right asset, tutorial or piece of code to one in which a large part of those elements can be understood, adapted or produced at the exact moment they are needed.

Two hours do not replace twenty years

Of course, we need to remain reasonable.

A game developed in two hours does not replace twenty years of development. It does not automatically have the content, polish, art direction, balancing and thousands of decisions accumulated in a major production.

But that is not really the interesting comparison.

What strikes me is that, in two hours of intensive development, it is becoming possible to reach a level of execution that would previously have required several days, sometimes several weeks, simply because one had to search, learn, test, understand APIs and solve each problem in succession.

That research time has not disappeared entirely.

It has been compressed.

The dashboard has changed too

The exact same thing applies to a business application.

A dashboard is not a collection of pretty cards with numbers.

Behind the interface are data, relationships, permissions, filters, migrations, logging, tests, business rules and sometimes an entire architecture.

Here again, it is no longer only pixels that can be produced quickly.

It is now the structure behind them as well.

And now?

The more projects I develop, the less I feel that the real revolution is automatic code generation.

Code is only the visible part.

The revolution may simply be this:

we have brutally reduced the distance between knowing something, understanding how it works and being able to put it into action.

A curious person can now explore physics, 3D rendering, databases, shaders, procedural generation, software architecture or fields that would previously have taken them months to approach.

It does not replace expertise.

But it makes expertise accessible, makes it possible to experiment with it and, above all, to produce with it much more quickly.

And when thousands of developers, creators and simply curious people gain this capability at the same time, the result becomes visible everywhere: an utterly staggering number of games, tools and applications.

That is why I no longer think we are waiting for a development revolution.

It has already begun.

1. OpenAI — Codex: https://openai.com/codex/
2. OpenAI — Open source Codex orchestration with Symphony: https://openai.com/index/open-source-codex-orchestration-symphony/
3. Microsoft — TRELLIS.2: https://github.com/microsoft/TRELLIS.2
4. img2threejs — Image to procedural Three.js: https://github.com/img2threejs/img2threejs