The speed of AI progress now hinges on whether the self-improvement loop “closes.” This shift toward recursive engineering could transition development from linear growth to a compounding feedback loop, where models autonomously build their successors. If the digital cycle closes, potentially collapsing software engineering timelines to 6–12 months window, AGI development could shift from linear to exponential growth.
The engineering loop that could collapse AGI timelines
At Davos 2026, the discussion centered on a single engineering question: Can AI meaningfully help build better AI fast enough without humans doing the heavy lifting? This “closed loop” refers to models performing end-to-end software tasks—designing, coding, and testing their own successors—to accelerate the next generation of intelligence.
The Architects: Leading the AGI conversation
The debate featured two titans: Dario Amodei, CEO of Anthropic, and Demis Hassabis, CEO of Google DeepMind. While both agree AGI is a credible path, they differ on how quickly this recursive loop can actually function without human intervention.
Diverging Timelines: A comparison
The core of the Davos debate was not if AGI happens, but when the loop reaches full autonomy.
| Metric | Dario Amodei (Anthropic) | Demis Hassabis (DeepMind) |
| AGI Arrival | 1–2 years (by 2027/2028) | 5–10 years (50% chance by 2030) |
| Software Loop | 6–12 months to full autonomy | Slower; cites “jagged intelligence” |
| Primary Friction | Purely an engineering cycle | Physical sciences & verification |
| Job Disruption | 50% of white-collar work in 1–5 yrs | “Uncharted territory” in 5–10 yrs |
Dario Amodei: The compounding engineer
Amodei frames AGI as a practical engineering outcome. For him, the software loop is already tightening as engineers at Anthropic move from writing code to simply overseeing model-generated production code.
I think we might be six to 12 months away from when the model is doing most, maybe all, of what software engineers do end-to-end.
Dario Amodei, CEO of Anthropic
Demis Hassabis: The physical speed limit
Hassabis identifies where the digital loop breaks against the friction of the real world. He argues that while code is easy to verify, breakthroughs in biology and physics require messy, real-world experiments.
Coding and math are easy because output is verifiable; science, biology, and theory creation introduce friction that slows the loop.
Demis Hassabis, CEO of Google DeepMind
Why this matters for business strategy
If the loop closes digitally, we face a sudden intelligence explosion. If physical bottlenecks persist, the transition remains more manageable.
- Founders: If the loop closes, software moats vanish. Focus on “proprietary cognition”—data that exists outside the digital loop.
- Business Owners: Prepare for a labor shift. Amodei warns that entry-level white-collar roles could be overwhelmed within one to five years.
- Professionals: Focus on verification. As AI generates more output, the value shifts from creation to judging what is actually correct.
Source: World Economic Forum 2026
