Ideas

Those experiments helped refine a concern Vishal first explored personally in 2014.

Instead of demanding complete digitization first, could a business begin with only a location, category and indicative presence, perhaps built conversationally from text, voice and photographs?

Locally operated nodes could coordinate discovery within a city or trading area. Buyer-authorized enquiries would seek confirmation from selected businesses, with searches extending to neighbouring nodes only with permission.

The aim is to support local livelihoods while preserving direct relationships, merchant choice and compatibility with open-commerce systems. The wider network and its livelihood benefits remain hypotheses to test.

Our first experiments used trusted source material together with multimodal AI to create a more adaptive tutoring workflow: compress familiar material, stop where understanding breaks, explain prerequisites when needed, and return to the original learning objective.

That leads to a larger systems question.

Could a learning system model the target capability as a concept graph, estimate which parts the learner already understands, temporarily branch into missing prerequisites and then return to the main objective?

Career OS explores the same idea from another direction: repeated job-market gaps can become learning signals rather than disappearing as isolated rejections.

The broader goal is not simply a better AI tutor.

It is to explore systems that connect goal → prerequisite → learning → evidence of usable capability.

Existing agent standards increasingly support discovery, task state and agent-to-agent interaction.

Our interest sits one level above that:

What exactly moves when ongoing responsibility moves?

A responsibility may carry a goal, authority limits, context, escalation rules, human delegates, expiry conditions and a record of what already happened.

We are also interested in group coordination.

Many small requests force several people to context-switch even when their personal agents could safely perform the mechanical work and ask only when human approval is genuinely needed.

The design objective is not maximum autonomy.

It is continuity without loss of agency.

Someone is already travelling toward where a non-urgent item needs to go.

Physical capacity exists, but coordination does not.

This theme explores mobility as a time-varying capacity graph.

Trusted ride matching could begin inside existing social or professional circles and widen only when needed. Public transport could remain the high-capacity backbone while peer or commercial services help coordinate the edges. Parking could become temporarily discoverable instead of remaining statically assigned.

For non-urgent goods, time flexibility creates another possibility: fulfilment may occasionally piggyback on movement that was already going to happen rather than creating a dedicated trip.

The difficult questions are not only routing.

They are trust, incentives, privacy, custody, accessibility, regulation and whether the system measurably reduces duplication rather than merely shifting it around.

The interesting opportunity was not to replace the cobbler.

It was to connect capabilities that normally live apart:

professional guidance → digital specification → parametric design → local 3D fabrication → local footwear integration → validation and iteration

Flexible materials such as TPU, increasingly accessible digital fabrication and AI-assisted technical exploration make such workflows more plausible.

The research question is not whether customized footwear already exists; it does.

It is whether the capability can be safely unbundled, allowing specialist knowledge to remain specialist while fabrication and integration become more geographically distributed.

That same model may eventually apply to other low-volume personalized physical products.

Some AI tasks can remain deterministic.

Some can run on small local models.

Some deserve larger cloud models.

Others can wait until a device is charging, idle or connected.

Old phones and computers may still provide useful always-on infrastructure. Modern NPUs can expose local inference to trusted applications. Delay-tolerant workloads can potentially shift in time or location according to resource availability.

At a more speculative edge, we are interested in whether flexible computation can sometimes move toward otherwise underused energy rather than assuming all energy must move toward permanent computing infrastructure.

The principle is simple:

Use the right amount of intelligence, on the right resource, at the right time.