The AI Futures Project has published a major update to its forecasting work, introducing a unified “AI Futures Model” intended to predict when systems will reach key capability milestones such as “Automated Coder” (full automation of an AI lab’s coding workforce) and, eventually, superintelligence.

In the update, the authors explain that their newest model shifts their median estimate for full coding automation later compared with their earlier “AI 2027” work, largely because they are now less optimistic about how quickly partial AI-driven R&D acceleration compounds into rapid, self-reinforcing progress. A core idea is to ground forecasts in measurable capability trends (including coding-related benchmarks), while explicitly modeling constraints like compute growth limits, investment ceilings, and organizational bottlenecks that could slow trajectories.

They also emphasize transparency: the model exposes assumptions and parameters so readers can see what drives outcomes and adjust inputs if they disagree. The team has also released an interactive website where users can explore the model’s structure, scenarios, and forecast distributions.

The bigger takeaway is that AI timeline debates are moving from vibes to frameworks: even if any single model is wrong, having an explicit, testable structure makes it easier to update beliefs as new evidence arrives — and to identify which variables matter most for policy, safety, and business planning.