When Will Superintelligence Arrive? Why You Should Not Set a Date

There is no reliable date for when superintelligence will arrive, and the forecasts that exist disagree widely with each other. For a company, that changes the question: a plan that waits for a technology milestone tends to postpone decisions that can already be made with the systems available today. The most useful approach is to plan around what works now, with clear limits, records and regular reviews, and to treat any long-term forecast as a hypothesis to monitor.

Key takeaways

• Forecasts about superintelligence do not have enough basis to guide budgets, hiring or internal deadlines.
• Waiting for a technology milestone shifts the risk onto current operations, which keep running while the plan is on hold.
• A useful plan separates three things: official documentation, implementation experience and hypotheses.
• A pilot with a real workflow, records and a success metric agreed in advance teaches more than any forecast.
• The decision to expand, adjust or stop should be reviewed with evidence from your own operation, not with dates on a calendar.

Why forecast dates should not guide the plan

Predictions about when superintelligent systems will exist come from researchers, executives, commentators and science fiction authors, using different criteria and offering little way to verify them. Some extrapolate from test results, others from opinions about what remains to be discovered. Almost none define superintelligence in operational terms, and without that definition there is no way to compare one forecast with another. For someone who has to decide, a date without clear criteria works more as conversation than as a business assumption. The myths about superintelligence that come up most often in meetings usually start with exactly this kind of claim.

The practical problem shows up in two directions. Those who bet on a distant date tend to delay pilots, data organization and team training, and arrive at the expected moment with nothing ready to use. Those who bet on a near date tend to sign large contracts without defined scope, expose data without controls and switch vendors with every announcement. In both cases, the calendar makes the decision in place of the people who know the process.

What changes in planning when the milestone leaves the equation

Replacing the question of when it will arrive with the question of what our process needs today changes the entire plan. The horizon stops being an event and becomes a review cycle. Each decision needs an owner, a metric and a criterion for stopping.

Before any decision, it helps to separate three types of information. The first is official documentation from products and providers, which describes what the tool does, what limits it has and how it handles data. The second is implementation experience, meaning what happened when a real workflow was put into operation. The third is hypothesis, such as estimates about future capabilities. Hypotheses can guide what you watch, but they should not support financial or operational commitments.

The NIST AI Risk Management Framework organizes AI risk management into govern, map, measure and manage functions, and it can serve as a checklist of questions for the board. The OECD AI Principles address transparency, accountability and robustness, and they help define what to require from any vendor, regardless of how the technology evolves.

Five-step plan so you do not depend on a date

• Map the current process, including timings, volumes, known errors and who decides at each step.
• Define what the AI can do on its own, what it only suggests and what requires human approval.
• Limit data, tools and permissions to the minimum the chosen workflow needs.
• Pilot a real workflow with a record of every decision and a success metric agreed before the start.
• At the end of the agreed period, decide whether to expand, adjust or stop, and record the reason.

Anthropic, in its guidance on building effective agents, recommends starting with simple patterns and adding complexity only when it is justified, which fits this plan. To measure whether the agent does what it should, set up fixed evaluations before the pilot, as OpenAI's evals guide describes, and repeat them after every change to instructions, tools or model.

Illustrative scenario: WhatsApp support at a distributor

Illustrative scenario: a building materials distributor receives requests on WhatsApp for payment slip reprints and delivery status. Leadership debates whether to wait for more capable agents before changing customer support. The team decides on something else. It maps the most frequent types of question, chooses only a few for the pilot, sets the rule that the agent does not negotiate deadlines or discounts, and passes any case outside that scope to a person along with the conversation history. Respecting the WhatsApp 24 hour window is part of the sending rules. The logs show where the agent made mistakes and which questions went unanswered. At the end of the period, the team expands the scope to other question types and keeps negotiation with people. No future technology date entered the discussion, and the plan stayed useful whatever comes next.

Checklist for reviewing the plan each cycle

• Do the facts used in the decision come from official sources, or were they measured in your own operation?
• Was the metric agreed at the start met, partly met or not measured?
• Were there incidents involving personal data, wrong answers or improper access?
• Is the person with authority to pause the pilot named and clear on how to do it?
• Does the expansion respect the principles of Brazil's data protection law (LGPD), such as purpose, necessity and transparency toward the data subject?
• Has any long-term hypothesis changed in a way that affects the current choice?

If the review raises questions about legal obligations, follow the Brazil AI regulation closely with your legal team.

How Juan Carlo's method applies to this case

The method of mapping, limiting, piloting and handing over exists precisely to replace waiting with a cycle of evidence. First we map the process and its baseline, before any new technology enters. Then we limit data and actions to what is necessary. Next we pilot a real workflow, with records and a metric agreed with the company. At the end, we hand over the operation and decide together whether to expand, adjust or stop. This process does not depend on predicting when the technology will change, because each cycle produces evidence from the business itself. Anyone who wants to see how this work is carried out can read about how Juan Carlo works with companies.

Frequently asked questions

When will superintelligence actually arrive? No one has a verifiable answer, and the forecasts available disagree widely with each other. For a company, the useful question is what the process needs now and what can be controlled with the systems that already exist. An uncertain date should not be the criterion for a decision.

Should I wait for superintelligence before starting with AI? No. Small projects with limited data and clear metrics produce learning that stays useful regardless of what comes later. Waiting usually means only that the team loses the practice of leading changes and reviewing results.

How do I know whether a news story about superintelligence should change the plan? Check whether the claim comes from official documentation, a reproducible test or only an opinion. Only changes that affect your process, such as pricing, availability, legal requirements or performance measured in your operation, justify revising the plan. General forecasts should be noted as hypotheses and monitored, without creating commitments.

What should a company record to make this decision safely? Record the process baseline, the decisions the agent makes, the cases passed to people and the metric agreed at the start. These records show what worked and what needs to change, and they form the basis for deciding to expand, adjust or stop. Without them, any review becomes opinion.

Sources

• NIST AI Risk Management Framework
• OECD AI Principles
• Anthropic: Building effective agents
• OpenAI: Evals guide