# Emerging Directions

# Emergent Directions for Agentic AI

## Large Action Models (LAMs)

Models designed to reason about actions, workflows, and consequences. Used for:

● Browser automation

● Enterprise workflows

● Multi-step operations

● Transaction execution

## Model Specialization

<div align="left" dir="ltr" id="bkmrk-role-best-model-type"><table><colgroup><col width="311"></col><col width="312"></col></colgroup><tbody><tr><td>### Role

</td><td>### Best Model Type

</td></tr><tr><td>Parsing

</td><td>Small / Fast

</td></tr><tr><td>Drafting

</td><td>Small / Medium

</td></tr><tr><td>Planning

</td><td>Medium / Large

</td></tr><tr><td>Verification

</td><td>Large / Low-temp

</td></tr><tr><td>Compliance

</td><td>Large / Deterministic

</td></tr><tr><td>Long Context

</td><td>Long-context specialist

</td></tr></tbody></table>

</div>## Context Engineering

Context engineering is now widely treated as the successor discipline to prompt engineering. It is the deliberate design of everything a model sees on every inference call: system prompt, user input, retrieved documents, conversation history, tool definitions, and long-term memory. The shift happened for three reasons: agents fail on state-management, not phrasing; context windows grew past 1M tokens, making naive "stuff everything in" both expensive and worse (the "lost in the middle" effect); and "context rot", degraded performance as context grows became a measurable, actively-managed phenomenon rather than a curiosity.

The Four-Layer Context Stack

<div align="left" dir="ltr" id="bkmrk-layer-contents-refer"><table><colgroup><col width="176"></col><col width="301"></col><col width="145"></col></colgroup><tbody><tr><td>### Layer

</td><td>### Contents

</td><td>### Reference Cadence

</td></tr><tr><td>System context

</td><td>Instructions, persona, output schema, tool definitions

</td><td>Per deployment

</td></tr><tr><td>Persistent context

</td><td>Memory of prior sessions, user preferences, learned facts

</td><td>Per session / weekly

</td></tr><tr><td>Retrieved context

</td><td>RAG hits, search results, DB lookups, document chunks

</td><td>Per turn

</td></tr><tr><td>Workflow / state context

</td><td>Prior agent decisions, current plan, tool outputs so far

</td><td>Per step

</td></tr></tbody></table>

</div>