Progress to AGI
Our honest, opinionated take on how close AI is to artificial general intelligence. Updated as the landscape evolves. Take it for what it is — one team's informed perspective.
This is not a scientific measurement. The percentages below reflect AnovaGrowth's subjective assessment based on our experience building and evaluating AI systems. Reasonable people disagree on these numbers. That's the point — this is meant to spark discussion, not settle it.
Aggregate across 6 capability dimensions
Where AI stands today
Reasoning
62%Chain-of-thought, logical deduction, multi-step problem solving. Strong on well-defined problems, still struggles with truly novel reasoning.
Planning
45%Long-horizon planning, goal decomposition, strategy formation. Getting better with agentic frameworks, but real-world planning remains brittle.
Learning
38%In-context learning is impressive. True continual learning — remembering and building on past interactions without retraining — is still limited.
Creativity
55%Generating novel combinations within learned patterns. Strong in text and image generation. Less clear whether this qualifies as genuine creativity.
Adaptability
42%Handling unexpected situations, transferring skills between domains, recovering from errors. Models are improving but still domain-dependent.
Autonomy
30%Operating independently over long periods without human intervention. Current agents need guardrails and can't reliably self-correct over extended workflows.
How we got here
Transformer architecture published
The foundation that made modern AI possible. Attention mechanisms changed everything.
GPT-3 demonstrates few-shot learning
The first time a single model could handle wildly different tasks with just a prompt. A turning point.
ChatGPT brings AI to the mainstream
Not a research breakthrough per se, but the moment the general public started paying attention.
GPT-4 and multimodal capabilities
Vision, reasoning, and tool use in a single model. The gap between AI and human-level performance started narrowing visibly.
Agentic AI and reasoning models emerge
Models that plan, use tools, and execute multi-step tasks. Still early, but the trajectory toward autonomous AI agents is clear.
Frontier models approach expert-level on benchmarks
Performance on many traditional benchmarks is saturating. The question is shifting from "can AI do this?" to "can AI do this reliably, at scale?"
What AGI means to us
We define AGI as an AI system that can learn, reason, and perform at or above human level across any intellectual task — without needing task-specific training. That bar is high. Current systems are impressive but narrow.
We think the path to AGI isn't a single breakthrough — it's a gradual accumulation of capabilities. Each year, systems get meaningfully better at reasoning, planning, and adapting. The question isn't whether, but when.
In the meantime, we focus on what's practically useful today — building AI that solves real problems right now while keeping an eye on where everything is heading.
Progress to AGI — Common Questions
Curious about where AI is heading?
Whether you're planning an AI strategy or just following the field, we're happy to share what we're seeing from the trenches.