Fri 13 Nov
LiveGet it free
ZOTPAPER

News without the noise

← All concepts

typed state representation

Agents often track their world as loose text or raw logs. A typed state representation keeps information in structured slots with defined kinds, like a labeled form. The system knows what belongs in each. This lets the agent compare options, validate decisions, and act without regenerating text. Different modules, such as planners or memory, share the same clean snapshot.

Across these works, typed states support better action scoring, hierarchical planning, residual forecasting, safe graph edits, Pythonic contracts, memory management, trace summaries, and case retrieval. Making state explicit and typed improves efficiency, controllability, and transfer, while cutting tokens and errors. It also allows deterministic auditability and lets agents and human observers consume the same structured view rather than raw streams.

Papers
25
Released code
2
First seen
July 2026
Latest
Nov 2026

16 papers in the last two months, against 8 in the two before.

Start here

Three papers to read first, in the order they came out
  1. 1Typed action encoding helps language models learn adaptable agent policies

    An architecture called Dyad adds a typed action encoder to LLMs, improving agent decision-making and general capabilities.

    Sept 2026
  2. 2AI agents track permissions better from updated records than long histories

    Maintaining an explicit state for an agent beats making a language model read the full event log, past a log length that grows with the model's strength.

    Nov 2026
  3. 3Visual puzzle solvers improve by building explicit problem states

    Explicit, revisable problem states, built from visual inspection and executed by generated programs, improve accuracy on visual reasoning benchmarks.

    Nov 2026

How fast it is moving

JunJulAugSeptOctNov so far

Often studied with

agent memoryA repository where an agent stores past interactions and outcomes to inform future decisions.5 togetherui groundingMapping natural language commands or descriptions to specific visual elements of a user interface.4 togetherruntime monitoringAudits an AI agent's live actions, internal states, or effects against constraints or policies to catch errors and attacks.4 together

Who is working on it

Microsoft 2

The papers

Most central to this idea first, not most recent.

Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.