Skip to content

Concept track

The concept track explains the foundational principles behind coding agents. Each page examines a core concept, such as agent architecture, context limits, or verification strategies, in a generalized form independent of any specific dataset or tool. You can read these modules sequentially or reference individual topics as needed.

Every concept page follows a consistent structure: theoretical background, general methodology, verification guidelines, common failure modes, and a Try it section linking directly to practical exercises in the demo tracks (see Track navigation below).

Topic areas

Section Focus
Background Core agent mechanics, differences from conversational models, and capabilities in research workflows
Setup and safety Harness selection, execution environments, and security practices for data and credentials
Practice Essential operational habits: task scoping, data ingestion, context hygiene, durable memory, steering, verification, and cost control
Themes and reference Cross-cutting principles and a reference glossary

Recommended entry points

Research need Recommended reading
Getting started with coding agents Agent vs. chat, followed by What code makes possible
Handling confidential or sensitive data Privacy and data handling, Secrets, and Sandboxing
Diagnosing inconsistent or unreliable outputs Scoping work, Context, and Verification
Managing compute and token expenses Costs and Context
Creating reusable workflows Memory and Tools

Contents

Background

Concept What it covers
Agent vs. chat A standard chat model answers from training memory. An agent interacts with a live computing environment to inspect files, execute code, and produce verifiable artifacts.
What code makes possible Coding agents use executable code as a universal interface for research tasks, making text-based source formats and programmatic scripts far more reliable than manual GUI interactions.

Setup and safety

Concept What it covers
Harnesses and models An agent system separates the inference engine (the model), the orchestration software (the harness), and the execution target (the environment). Understanding this separation clarifies system capabilities and boundaries.
Environment Computational research requires isolated execution environments to prevent dependency conflicts and guarantee that findings can be reproduced on other systems.
Sandboxing and isolation Coding agents execute commands with your user permissions. Enforcing appropriate isolation boundaries protects local files and prevents indirect injection attacks.
Privacy and data handling Data read by an agent is transmitted to the model provider as context. Determining whether data is permitted to leave your infrastructure must precede deciding what tasks to delegate.
Secrets Credentials must never enter prompts, session logs, or version control. Use isolated environment variables and enforce immediate revocation protocols if exposed.

Practice

Concept What it covers
Scoping work Scope requests to sizes that can be independently verified. Require structured plans before executing complex multi-step pipelines.
Inputs Language models process information as text. Selecting appropriate ingestion pathways and source formats preserves data structure and prevents extraction errors.
Outputs Request text-based source files and compile them to presentation formats. Source files are inspectable, version-controlled, and reproducible.
Context Context represents the model's active working memory. Because every turn re-transmits the complete conversation, long sessions increase latency, cost, and attention drift.
Memory Language models do not retain memory across sessions. Preserving knowledge requires externalizing guidelines, documentation, and tools into persistent files.
Tools Turning interactive prompts into parameterized scripts creates deterministic, testable tools that run identically on every execution.
Steering Balance supervisory approval modes with task risk, intervene early when execution drifts, and codify recurring corrections into standing rules.
Verification Language models can be confidently incorrect. Require empirical evidence, verify mathematical invariants, and use clean-context peer review to audit findings.
Version control Version control provides an immutable audit trail and safety net, allowing researchers to inspect every line changed by an agent and roll back errors instantly.
Costs Language model pricing is determined by token volume. Because conversation history re-transmits on every turn, session length compounds costs quadratically.

Themes and reference

Concept What it covers
Use AI recursively Apply the agent recursively to improve your collaboration workflows: drafting plans, refining instructions, reviewing code, and packaging tools.
Simple text formats win Plain-text source formats maximize parsing accuracy and enable precise line-by-line editing and version control.
Manage context carefully Working memory is finite, costly, and ephemeral. Maintain focused sessions and externalize knowledge into durable files.
Glossary Definitions of foundational terms and concepts used across the workshop.

Track navigation

  • Demo · Clinical applies these concepts sequentially to a complete worked example (Clinical: clinic network).
  • Demo · Imaging applies these concepts sequentially to a complete worked example (Imaging: chest radiographs).
  • Tutorial · Clinical interleaves the concepts with hands-on exercises in one progression (Clinical: clinic network).
  • Tutorial · Imaging interleaves the concepts with hands-on exercises in one progression (Imaging: chest radiographs).