Tools make more actions possible
In March 2025, OpenAI introduced the Responses API, tools for web search, file search and computer use, an Agents SDK, and ways to trace agent workflows. These building blocks were designed to help developers connect model reasoning with actions in software. Read OpenAI’s announcement.
OpenAI’s tools make a familiar starting point for a wider question. As AI systems become more capable, how do their actions produce useful changes in a person’s life? A software action can be one part of the answer. It may still depend on people, resources and decisions outside that software.
A projection makes the conditions visible
A projection describes a possible future and the conditions it depends on. It can show what changes for each participant, which contribution is missing and what would count as evidence. Changing a condition changes the possible future. It does not make that future happen.
A learner could become a repairer.
The learner knows the basics, but has not practiced the repair.
Keep studying the theory.
A course outline can explain the repair. It cannot provide the tools or hands-on practice.
Repair a bicycle independently.
Guided practice could turn knowledge into a demonstrated skill, and open up new work.
Observe the repair, test the bicycle and record what the learner can do without help.
A person’s goal crosses several owners
Here is a hypothetical RealMaker example. Someone wants to learn bicycle repair so they can maintain their own bike and take on suitable paid work. They have a little experience and limited free time. A local instructor offers practice sessions. A workshop has tools, and a community group can provide bikes to work on.
An AI-generated learning plan would be helpful. A workable arrangement must also fit the person’s schedule, the instructor’s availability, the workshop’s rules and the cost. If the person cannot attend the proposed session, the plan needs another route.
The instructor, workshop and community group each supply something the learner needs. The arrangement works only if teaching, tools, practice and available time come together.
Private help before network participation
The proposed RealMaker Node is a private workspace that helps the person understand their goals, existing skills, resources and next actions. The person should get value from that workspace before being asked to share information with others.
They could then choose to share a specific request: a suitable practice opportunity at an available time. The proposed RealMaker World would help connect that need with contributions other participants have chosen to offer. The person’s full financial situation or private history need not become public.
Each owner still decides. The learner accepts a booking. The instructor agrees to teach. The workshop authorizes the use of its tools. Any software action must operate within those permissions.
Show the difference between a plan and a result
Completing a booking is evidence that a session was arranged. It is not evidence that a skill was learned. Attendance, practice and an observed result tell different parts of the story. The model should record those differences and use them to improve the next recommendation.
The same pattern extends beyond learning: a supplier accepts work, a shipment moves, a customer receives a result, or a payment settles. RealMaker aims to connect these changes without needing to own every tool or organization involved.
RootKnow’s research question is whether this shared process helps independent participants finish useful work more reliably than existing approaches. More automation would not count as success if it adds errors, unwanted commitments or burdens that exceed its benefit.
RealMaker is in development. The learning scenario is hypothetical. OpenAI and Sam Altman are referenced for public context; no affiliation or endorsement is implied.