Capabilities · Trend
AI agents
Goal-directed systems that plan, use tools and act across software are the defining frontier of 2025–2026.
Forecast horizon
Agents move from demos to delegated work inside real workflows; reliability and oversight become the bottleneck.
Connections
Connections · 8
How this node ties into the rest of the map, and the evidence behind each link.
to · requires 8/10
Reasoning models Agents lean on deliberate reasoning to plan multi-step actions.
+3 growth to · applies to 8/10
Autonomous coding Software is the first domain where agents act end-to-end.
+3 growth to · requires 7/10
Long context & memory Acting over long horizons needs persistent memory.
+2 growth to · uses 7/10
Test-time compute Agents spend inference compute to search and self-correct.
+2 growth from · enables 7/10
Collapsing inference cost Cheap inference makes long agentic runs affordable.
+3 growth to · tracked by 6/10
Capability evaluations Autonomy raises the stakes for pre-deployment evaluation.
+3 growth from · tracked by 6/10
Capability evaluations Agentic autonomy is the hardest thing to evaluate.
+3 growth from · tracked by 6/10
Labor & automation Agentic automation reshapes task-level labor.
+2 growthSignal sources
Signal sources
Dated facts from primary sources in this direction.
Task horizon doubling Mar 2025
The length of software tasks AI agents can do autonomously at 50% reliability has doubled about every 7 months — and since 2024 closer to every ~3 months.
METR → Benchmarks saturating Apr 2025
In one year scores rose by 18.8, 48.9 and 67.3 points on MMMU, GPQA and SWE-bench; real-world software solve rate jumped from 4.4% to 71.7%.
Stanford HAI — AI Index 2025 → Autonomous coding 2025–2026
On SWE-bench Verified (500 real GitHub issues), autonomous coding agents reached ~80–86% by late 2025, up from under 50% in early 2025.
Epoch AI →