For years, I wanted to start a business and was afraid of what I could lose. My father had often spoken about the risks, and that stayed with me. Eventually, I started. It became one of the best decisions I have made, but I am still learning how to run and improve what I am building.
I have also had to adapt to unfamiliar environments and learn how to make decisions when I did not understand everything around me. Listening helped. So did accepting that I would need to change some of my assumptions.
Those experiences do not give me a timetable for superintelligence. They do give me a reason to approach the question through decisions a business can make while the future remains uncertain.
What is the difference between AI, AGI and ASI?
AI is a broad term for systems that perform tasks involving capabilities such as language, prediction and problem-solving. Artificial general intelligence, or AGI, refers to broad capabilities across many tasks, although researchers disagree on the exact threshold. Artificial superintelligence, or ASI, refers to systems substantially beyond human capabilities across a broad range of cognitive work.
Autonomy is a separate question: how much can a system do without supervision? A tool can act independently within a narrow workflow without being generally intelligent. DeepMind’s Levels of AGI framework distinguishes performance, breadth and autonomy.
When could superintelligence arrive?
There is no established date. A 2026 research paper, From AGI to ASI, explores several possible routes, including scaling, new algorithms, recursive improvement and coordinated groups of AI systems. The researchers also examine obstacles and recommend preparing for multiple scenarios rather than relying on one trajectory.
This is research about possible future systems. It should not be read as proof that an AI tool available to a business today can independently run its strategy, marketing or customer relationships.
The International AI Safety Report 2026 describes improving but uneven capabilities, persistent reliability problems and a gap between controlled evaluations and real-world performance. Both progress and those limitations belong in the conversation.
What could this mean for marketing?
A plausible scenario is that more capable systems handle larger parts of research, creative production, experimentation and customer follow-up. Another is that customers increasingly delegate comparison and purchasing tasks. These developments could change how businesses compete for attention and how buyers evaluate offers.
Some supporting infrastructure is already being developed. Google’s Universal Commerce Protocol updates describe capabilities for shopping agents to work with carts and retrieve product information. That is evidence of development, not evidence that most customers already delegate their buying decisions.
If capabilities improve much more slowly, businesses still face today’s problems: unclear offers, slow handoffs and difficulty knowing which changes work. If they improve quickly, those same businesses will need to reassess what they can delegate. A useful preparation plan should leave room for both.
Four things a business can prepare now
- Make business knowledge usable. Keep service scope, prices, customer commitments and internal instructions current. Identify which information is sensitive and who is allowed to use it.
- Describe outcomes clearly. An instruction to “grow sales” leaves unanswered questions about margin, suitability, cancellations and service capacity. State the limits alongside the objective.
- Separate suggestion from authority. Decide which tasks a system may draft, which actions require approval and which it may perform independently after testing.
- Keep a record of experiments. Save the starting situation, the change, the tool used, the outcome and the problems encountered. Revisit the result when the system or workflow changes.
These are practical preparations, not a guarantee of advantage in an ASI future. They can make current AI experiments easier to evaluate and make later decisions less dependent on memory or enthusiasm.
Choose the next decision you can test
One thing I keep coming back to in my own experience is how you respond when something fails. You try, learn, shift and keep the reason for doing the work in view. That applies to a business experiment even when the wider technology story feels much bigger.
Choose a limited workflow and define what a better result would be. If the tool helps, extend its role carefully. If it does not, look at the evidence and change the approach. The value of the experiment is what it teaches you about your business.
