There is something remarkable about a tool becoming part of the process that builds its next version. Artificial intelligence is entering that chapter. Alongside helping people make things, it is contributing to the work behind new AI. Software becomes a tool for creating the tools we will use next.

That shift makes the coming year an exciting prospect. If we can delegate parts of a project today, what becomes possible when those parts fit together more smoothly? Autonomous AI opens a compelling conversation about creativity, practical capability and who gets to build.

From answering questions to completing an assignment

Consider a straightforward brief: build a booking tool for a local service business. The experience becomes more useful when a system can follow the whole path from understanding the request to preparing a solution, checking it and returning something people can try. The useful unit of interaction becomes a working result.

That direction is visible in Claude Code's dynamic workflows. Anthropic describes systems that break down an assignment, coordinate agents and verify their contributions. This gives technical work an organizing structure through which connected tasks can be delegated.

Its compiler-building experiment provides a tangible example: 16 agents developed a program capable of compiling Linux 6.9. The researcher supplied the environment and tests. The system's autonomy operated within that framework.

The striking part is the change in the experience. An initial request can set a building process in motion. For someone with an idea, the interaction starts to feel like a workshop: propose something, work through it and come back with an artifact that invites testing and improvement.

AI becomes part of the machinery of progress

In When AI builds itself, Anthropic reports that Claude authored over 80% of the code merged into its codebase as of May 2026. This is an internal measure of merged code, rather than all work. The company places fully autonomous successor development in the future.

What makes this compelling is the relationship between generations of tools. A tool helps produce part of the environment in which later tools will be developed. If that contribution makes experimentation easier, a promising idea could reach a useful test sooner. Each improvement creates room to explore the next.

The supporting infrastructure is advancing too. Anthropic's November 2, 2026 announcement of message threads describes sending only new messages on each turn to reduce latency in long sessions. The approach leaves token usage unchanged.

Such improvements can look modest in isolation. Together, they invite a practical vision of technology that keeps pace with an extended assignment. Progress also comes from the small engineering decisions that make an experience smoother and help more steps fit into a useful workflow.

Self-sufficiency begins with carrying out the work

A useful way to describe this stage is operational self-sufficiency within defined tasks: making progress independently under an agreed brief and working conditions. That description makes the achievement easier to appreciate. People contribute purpose, context and the criteria for deciding whether the result is useful.

To follow the progress, look at what systems can complete. METR relates task difficulty to the time a human expert would need, at a given probability of success. That duration describes the task, rather than how long an agent runs. Its evaluations focus primarily on software, machine learning and cybersecurity.

For a business owner or creative team, the interesting question is how much useful work can be delegated, at what quality. A system that completes a sequence, carries context forward and produces something verifiable could change the relationship between an idea and the effort needed to explore it.

Four possibilities to watch by November 2027

The following scenarios are our editorial interpretation of the developments above, looking twelve months ahead from November 2026. They are possibilities to watch. Their arrival will depend on reliability, cost, integration and ease of use.

• More complete projects from an initial brief. Moving from a description to a working application could become easier. A local business might try a scheduling tool or an interactive product guide, put it in front of customers and learn from their reactions before committing to a larger build.
• Tools that fit the way a team works. A plausible next step is simpler customization around a real workflow. A creative studio could adapt its approval process; a retailer could improve order preparation. The practical gain would be making useful changes with less friction between an operational need and a working tool.
• More experiments behind the technology itself. If preparing and running tests becomes more dependable to delegate, researchers could explore more alternatives. The progress to watch would be a greater ability to turn technical questions into comparable experiments and results that people can evaluate.
• More people able to make something useful. Clearer, more accessible experiences could bring tool building within reach of professionals who understand a problem but have limited programming experience. Teachers, designers and founders could have more opportunities to turn that understanding into prototypes other people can try.

The most useful signs will be easy to recognize: how often a brief needs repeating, how much correction a delivery requires and how easily someone can use the result. Those observations make technological progress tangible to the person trying to get something done.

What that sense of possibility means for a business

Imagine a small company whose customers ask similar questions before buying. Its first project could be a simple product guide that organizes those answers and helps people find a suitable option. More autonomous AI could make it easier to construct a prototype, while the team supplies actual customer questions and shapes the experience.

The opportunity would be earlier learning: seeing which information is missing, which questions recur and which part of the journey deserves attention. That learning can be more valuable than adding features. At La Fábrica, this evolution invites us to imagine digital experiences grounded in a specific need and built around a useful response.

The coming year offers an extraordinary horizon for exploring that possibility. Picture smaller teams trying ideas they had left on the backlog, professionals making tools for their own work and more room to focus on what makes a product valuable to the people using it.

AI helping build the next AI opens a new chapter in technological creation. The wonder lies in a growing capacity to make things: an idea takes its first working form, that form teaches us something and the lesson informs another version. A year from now, many more people could have the opportunity to follow that path.