Structure beats scale: $8K Stratego win, 30-point swing on agent orchestration
Every URL the pipeline pulled into ranking for this issue — primary sources plus the supporting and contradicting findings each Researcher returned. Inline citations in the issue point back here.
Sources
With most information hidden, the game Stratego had stumped AI—until now arstechnica.com
Adding in a second neural network that guesses the identity of hidden pieces was key.
The Dot and the Swarm oneusefulthing.org
Benefitting from the Bitter Lesson
References
MIT News news.mit.edu
Ataraxos reached superhuman levels using approximately one-hundredth of the training examples and one-thirtieth of the self-play games required by DeepMind’s DeepNash, with training costs under $8,000.
The Decoder the-decoder.com
AI techniques developed for games like poker definitely could not scale to Stratego’s explosion of possible universes; Ataraxos uses a three-pronged architecture — policy-value network, generative belief network, and decision-time search — to ‘zoom in’ on plausible board states.
SuperpowerDaily (summarizing Nature paper) superpowerdaily.com
Three-time world champion Vincent de Boer considers the AI’s inability to adjust its strategy across a long series of games a ‘major handicap,’ since a human like Niemeijer can learn and exploit an opponent’s patterns over time.
VMTech analysis vmtech.rs
In 2-player Hanabi, Ataraxos averaged 24.654/25 with 77.53% perfect games, surpassing SPARTA’s 24.61; it also defeated DouZero and PerfectDou at Dou dizhu, demonstrating the method generalizes beyond Stratego.
DeepMind blog (DeepNash, 2022) deepmind.google
DeepNash avoids modeling hidden pieces explicitly, instead using model-free Regularized Nash Dynamics to converge to an unexploitable Nash equilibrium — a fundamentally different design than Ataraxos’s generative belief approach.
Remio.ai developer writeup remio.ai
Human opponents described Ataraxos’s play as ‘preternaturally lucky,’ pulling off bluffs humans consider too risky; the system was trained on roughly 16 H100 GPUs for about a week.
Siddharth Yadav, ‘The Bitter Lesson versus the Garbage Can’ (sids.in) sids.in
Most organizations struggle to define their objectives clearly enough for an AI to optimize them; the Bitter Lesson only applies when the challenge can be represented as clean data.
Drew Breunig, ‘Does the Bitter Lesson Have Limits?’ (dbreunig.com, Aug 2025) dbreunig.com
In specialized domains like signal processing, hand-crafted methods such as FFTs remain more efficient than deep learning — compute does not dissolve every structure.
Sriram Narasimhan, ‘Why Multi-Agent Systems Fail in Production: They Forget’ (Medium) sriram-narasim.medium.com
Real-world failures often stem from organizational memory loss — agents forget earlier context or loop endlessly, leading to spiked error rates in regulated environments.
Machine Learning Mastery, ‘The Current State of Agentic AI’ machinelearningmastery.com
The same model can see a 30-point performance delta on GAIA depending solely on orchestration logic, and some frameworks are up to 50x more expensive than others for similar accuracy.
Uvik, ‘Agentic AI Frameworks’ survey uvik.net
Over 57% of organizations now have agents in production; LangGraph dominates stateful workflows while CrewAI leads role-based automation, replacing the earlier experimental ‘Swarm’ project.
Ethan Mollick, ‘The Overhang’ (One Useful Thing, prior essay) oneusefulthing.org
Capability has outrun deployment — the models can already do more than organizations have figured out how to use.
