ANTMAN: Adaptive Need Tracking for Multi-Agent Navigation in Large Information Spaces
Jerry Wang, Haibo Jin, Xiaopeng Yuan, Peng Kuang, Haohan Wang
University of Illinois Urbana-Champaign
Published September 27, 2026 · on the feed September 30, 2026 · ▲ 32 · arXiv 2609.33326
The abstract
Information-seeking agents increasingly operate over information spaces that are too large to process exhaustively. Yet many multi-agent systems organize computation around static partitions of the available space, causing coordination to grow with how information is segmented rather than with what the query still requires. We introduce ANTMAN, an adaptive coordination framework that treats evolving unresolved information needs as the unit of runtime coordination. ANTMAN maintains a revisable Need Graph that tracks unresolved requirements, accumulated evidence, prior attempts, and search progress, and uses this state to control worker selection, routing, and task-local recovery as new evidence is discovered. By separating the coordination policy from substrate-specific search interfaces, the same need-conditioned mechanism can operate across different information spaces. Experiments across multi-document question answering, controlled long-context scaling, and realistic structured navigation show that ANTMAN remains effective across settings, including when execution is delegated to substantially smaller worker models. Under a 16x increase in searchable context, ANTMAN increases active coordination by only 1.23x, compared with more than 15x for partition-driven baselines, while preserving strong answer quality.
The Brief
2:06A short podcast summary of the paper. Written and voiced by AI.
Transcript
ANTMAN. Adaptive Need Tracking for Multi-Agent Navigation in Large Information Spaces. By Jerry Wang and four others, from the University of Illinois Urbana-Champaign. Published on the twenty-seventh of September. No code repository is linked to it.
Information-seeking agents increasingly work over spaces of information too large to process exhaustively. Many multi-agent systems handle that by splitting the available space into static partitions and organising their computation around them. The consequence is that coordination grows with how the information happens to be segmented, rather than with what the query still requires.
ANTMAN, short for Adaptive Need Tracking for Multi-Agent Navigation, changes the unit of coordination. At runtime, what gets coordinated is the set of unresolved information needs, and those needs evolve as the search goes on. Coordination follows the question rather than the layout of the data.
At its centre is a revisable Need Graph. It tracks the requirements that are still unresolved, the evidence accumulated so far, earlier attempts, and how far the search has progressed. That state then controls which workers are selected, how work is routed, and how recovery happens locally within a task as new evidence is discovered. The coordination policy is kept separate from the search interfaces of any particular substrate, so the same need-conditioned mechanism can operate over different kinds of information space.
Experiments span multi-document question answering, controlled long-context scaling, and realistic structured navigation. ANTMAN remains effective across all three, including when execution is delegated to substantially smaller worker models. One number summarises the scaling behaviour. When the searchable context grows sixteen-fold, ANTMAN's active coordination rises by only one point two three times, while partition-driven baselines rise by more than fifteen times. Answer quality stays strong throughout.
Think of two teams searching a vast library. One divides the work by shelf, the other by open question. For the first, every new shelf means more coordination. For the second, coordination tracks only what is still unknown, and that is the property the sixteen-fold test isolates.
Paper Quest · 7 chapters · about 41 min
Play this paper
Learn it by doing: make the calls the authors made, guess the result before you see it, and defend it in a viva graded on your own words. Chapter 1 is right here.
Chapter 1 🧩 · The problem
More data should not mean more chaos. Here is the wall.
You build a team of language-model workers that answer questions over huge document sets, too large for any one model to read.
The road ahead
- 1 · The problem ~7 min Splitting a huge document set into fixed chunks keeps each worker small, but the team then grows with the data, not with the question.
- 2 · What came before ~3 min
- 3 · The key idea ~9 min Coordinate around the information the query still lacks, kept in a revisable graph, instead of around fixed slices of the input.
- 4 · Choose your own research ~3 min
- 5 · Build the method ~8 min ANTMAN loops: track open needs in a graph, route workers from it, update it with evidence, and recover locally when a need stalls.
- 6 · The experiment ~6 min You’ll guess: Under a 16x increase in searchable context, by what factor did ANTMAN's active coordination grow?
- 7 · The viva ~5 min Three questions from the examiners, answered in your own words. Win it for the paper’s trophy.
Free, with Google or an email address. Your stars and progress are kept.