AGI-on the cusp
-
GPT-6 Astra: how big is the leap, and what does it mean for AI and monetisation?
The leap from GPT-5 to GPT-6 Astra looks increasingly significant. This is not simply another incremental improvement in a chatbot. The important change is the breadth of what the model can do reliably: reason across complex problems, work autonomously for longer periods, use tools, handle large amounts of information and increasingly execute multi-step tasks rather than simply telling you how to do them.That matters because it moves AI much closer to general-purpose intelligence. We are getting to the point where the distinction between a highly capable specialist system and something approaching AGI becomes increasingly difficult to maintain. It does not necessarily mean that full AGI has definitively arrived, but the trajectory suggests that we may be very close to the threshold.
More importantly, there is little evidence that we have reached the ceiling.
The capabilities of these models are improving simultaneously across reasoning, coding, mathematics, research, planning, memory, multimodal understanding and autonomous tool use. Each improvement also makes the others more valuable. A model that can reason better can use software more effectively; better software use allows it to undertake longer tasks; longer tasks generate more useful feedback and create more opportunities for automation.
This creates an enormous potential economic opportunity.The obvious applications today — coding assistants, customer service, document processing, research and productivity — are probably only the beginning. As models become more reliable, they can increasingly perform entire workflows rather than individual tasks.
For example, an AI system could eventually:
run a company's finance function;
prepare and reconcile accounts;
complete tax returns;
analyse investments and portfolios;
conduct investment research;
monitor markets continuously;
write and test software;
run compliance processes;
undertake scientific and medical research;
design new materials and chemical compounds;
manage procurement and supply chains;
operate sophisticated customer-service functions;
act as a personal executive assistant;
and eventually coordinate other AI agents and human employees.
That is a vastly larger addressable market than today's AI software market.
The really important point is that monetisation does not depend upon finding one killer application. If AI becomes sufficiently capable, it can be embedded into thousands or millions of existing economic activities. A relatively small amount of automation multiplied across almost every industry becomes an enormous pool of potential revenue.This is why the opportunity could be much larger than the market currently assumes.
If GPT-5 represented a major step towards highly capable AI, GPT-6 Astra potentially represents another step towards AI that can actually do the work. And if the underlying capability curve continues, there is no obvious reason to assume that GPT-6 is anywhere near the ultimate ceiling.
The bigger investment question, therefore, is not simply "How good is GPT-6?" It is:
How much of the world's intellectual and administrative work can increasingly capable AI eventually perform — and how much economic value will accrue to the companies providing the intelligence, compute, infrastructure and applications required to do it?If we are genuinely approaching AGI while still being relatively early on the capability curve, the potential economic impact is enormous. The applications that can be monetised may be limited far more by imagination, reliability and deployment than by the underlying technology.
In that scenario, today's AI applications could look like the very early stages of a much larger technological transformation.