Module 7: Orchestration and Memory › Agent-to-Agent Interop
GitHub Login Glossary

Agent-to-Agent Interoperability

You've built orchestration within a single system, a router that delegates to specialists, all running in one process under one graph. That covers most code assistant use cases. But as agent-based systems grow in organizations, a new question emerges: what happens when agents built by different teams, running in different processes, or even operated by different organizations need to collaborate?

This lesson introduces the protocol landscape for agent interoperability. Our goal here isn't to build a complex cross-system architecture. We can build all sorts of interesting solutions once we master this concept. Our goal here is to understand the distinction between MCP, A2A, handoffs, and orchestration so you can make the right choice when the need arises.

What you'll learn

  • Distinguish four coordination patterns: MCP (agent-to-capability), A2A (agent-to-agent), handoffs (control transfer), and orchestration (routing logic)
  • Identify when cross-system agent communication solves a real problem that simpler patterns can't
  • Understand what the A2A protocol provides and where it fits in the agent interoperability stack
  • Apply a decision framework for choosing between internal orchestration and cross-boundary protocols

Concepts

MCP (Model Context Protocol) — a protocol for connecting an agent to external capabilities: tools, data sources, and resources exposed by a server. We've been using MCP implicitly throughout this curriculum whenever our agent calls tools. The key mental model: MCP is agent-to-capability. The agent is the actor; the MCP server provides functions the agent can call. The server doesn't have its own goals, plans, or state — it's a capability surface.

A2A (Agent-to-Agent protocol) — a protocol for connecting one agented system to another as a peer. Unlike MCP, both sides of an A2A connection are agents with their own goals, context, and decision-making. A2A defines how agents discover each other, exchange tasks, negotiate capabilities, and stream results. Where MCP says "here are the tools you can call," A2A says "here's another agent you can delegate to or collaborate with."

Handoff — a control transfer between agents inside one application or runtime. We built handoffs in the previous two lessons: the orchestrator passes control to a specialist, the specialist does its work, and control returns to the orchestrator. Handoffs happen within a single trust boundary and a shared state space. The orchestrator can see the specialist's full state and output.

Orchestration — the broader control logic that decides routing, delegation, state management, retries, and synthesis. Orchestration encompasses handoffs but is bigger: it includes the routing decision, the state that persists across steps, the retry logic when a specialist fails, and the synthesis step that combines outputs. You can have orchestration without A2A (everything runs locally) or A2A without a central orchestrator (agents collaborate as peers).

The four-way distinction

These four patterns operate at different levels:

PatternWhat it connectsTrust boundaryState sharingExample
MCPAgent to capability serverAgent trusts the server's toolsAgent controls all stateYour agent calling search_code via an MCP server
HandoffAgent to agent within one runtimeShared trust, shared processFull state visibilityOrchestrator passing control to the code specialist
OrchestrationOverall control flowOne system owns the flowCentralized state graphLangGraph routing to specialists and synthesizing
A2AAgent to agent across boundariesSeparate trust domainsNegotiated exchangeYour code assistant delegating a security review to a separate team's security agent

The most common confusion is between MCP and A2A. The distinguishing question: does the other side have its own goals and decision-making? If it's a tool server that exposes functions, that's MCP. If it's an autonomous agent that interprets your request, plans its own approach, and returns results on its own terms, that's A2A.

Walkthrough

When cross-system agent communication matters

Most systems don't need A2A. Internal orchestration with handoffs covers the majority of multi-agent use cases. A2A becomes valuable in three specific situations:

1. Organizational boundaries. Your team's code assistant needs to delegate a security review to the security team's agent. You can't (and shouldn't) fold their agent into your system. They maintain it, they control its policies, and they update its rules independently. A2A lets your agent send a task and receive results without coupling to their implementation.

2. Capability boundaries. Your agent needs to interact with a system that has its own autonomy, like a CI/CD agent that decides how to run tests, a deployment agent that manages its own rollback policies, or a monitoring agent that interprets alerts according to its own rules. These aren't tools you call; they're systems that make their own decisions based on your request.

3. Scale boundaries. When the number of specialists grows beyond what one orchestrator can manage, or when specialists need to discover each other dynamically rather than being hardcoded in a graph, A2A's discovery and negotiation protocols provide structure that static orchestration can't.

If none of these situations apply (all your agents run in one process, under one team's control, with shared state), internal orchestration and handoffs are simpler and more debuggable.

What A2A provides

The A2A protocol defines several key capabilities:

Agent discovery. Agents publish "agent cards," structured descriptions of what they can do, what inputs they accept, and what outputs they produce. Other agents can discover these cards and decide whether to delegate work. This is analogous to how MCP servers publish tool schemas, but at the agent level.

Task exchange. A2A defines a task lifecycle: one agent creates a task, sends it to another agent, and receives updates as the task progresses. Tasks can be synchronous (wait for a result) or asynchronous (get a notification when it's done). The protocol handles the messaging, and your code handles the business logic.

Capability negotiation. Before sending a task, agents can negotiate what's possible. "Can you review Python code for security vulnerabilities?" "Yes, I can review Python 3.8+ code for OWASP Top 10 categories." This prevents wasted work on tasks the receiving agent can't handle.

Streaming results. For long-running tasks, A2A supports streaming partial results. Your agent can show the user progress while the peer agent works, rather than blocking until the full result is ready.

What A2A doesn't replace

A2A isn't a replacement for MCP, handoffs, or orchestration. It's a complement for specific boundary-crossing situations:

  • MCP still handles tool access. Your agents still use MCP to call tools, read files, and query databases. A2A sits above the tool layer.
  • Handoffs still handle internal routing. Within your orchestration graph, specialists still pass control via handoffs. A2A is for when you're crossing a process or organizational boundary.
  • Orchestration still handles control flow. Your orchestrator still decides routing, manages state, and synthesizes results. A2A is a communication channel, not a replacement for your control logic.

A decision framework

When you're deciding how to connect agents, work through this sequence:

  1. Can a single agent handle it? If yes, don't split. (We covered this in the first lesson of this module.)

  2. Can internal handoffs handle it? If all the agents run in your process, under your control, with shared state, handoffs within your orchestration graph will cover most situations you'll encounter.

  3. Do you need tool access across a boundary? If you need another system's functions but not its decision-making, use MCP. Expose the tools via an MCP server and call them from your agent.

  4. Do you need another system's autonomous decision-making? If the other side has its own goals, policies, and reasoning, and you can't or shouldn't fold that into your system, A2A is the right fit.

Most code assistants will stop at step 2 for a long time. A2A becomes relevant when the system grows beyond a single team's scope or when integration with autonomous external agents is a real requirement, not a speculative one.

A lightweight A2A example

To make this concrete, here's a minimal example of what an A2A interaction looks like. Your code assistant wants to delegate a security review to a separate security agent:

# This is a conceptual example showing the A2A interaction pattern.
# The actual A2A protocol implementation depends on the SDK/library you use.

# 1. Discover the security agent
security_agent_card = {
    "name": "security-reviewer",
    "description": "Reviews code for security vulnerabilities",
    "capabilities": ["python-security-review", "dependency-audit"],
    "endpoint": "https://internal.example.com/agents/security-reviewer",
    "input_schema": {
        "type": "object",
        "properties": {
            "code_snippet": {"type": "string"},
            "language": {"type": "string"},
            "review_type": {"type": "string", "enum": ["owasp", "dependency", "full"]},
        },
    },
}

# 2. Create a task
task = {
    "type": "security-review",
    "input": {
        "code_snippet": "def handle_upload(file_path: str): ...",
        "language": "python",
        "review_type": "owasp",
    },
    "callback": "https://your-agent.example.com/tasks/callback",
}

# 3. Send the task and receive results (async)
# In a real implementation, this would use the A2A SDK client
# result = a2a_client.send_task(security_agent_card["endpoint"], task)

# 4. The security agent processes independently and returns:
result = {
    "status": "completed",
    "findings": [
        {
            "severity": "high",
            "category": "path-traversal",
            "description": "file_path parameter is not validated against directory traversal",
            "recommendation": "Use pathlib and validate against an allowed base directory",
        }
    ],
    "summary": "1 high-severity finding: path traversal vulnerability in handle_upload",
}

The key difference from a tool call: the security agent decided how to review the code, what rules to apply, and how to structure its findings. Your agent sent a request; the security agent exercised its own judgment. That's the agent-to-agent boundary that A2A formalizes.

Exercises

  1. Review your current orchestration graph. For each specialist, decide: is this a handoff (internal, shared state) or could it ever become an A2A boundary (separate team, separate policies)? Most will be handoffs, and that's expected.

  2. Identify one hypothetical scenario where your code assistant would benefit from communicating with an external agent. Write an agent card for that external agent describing its capabilities, input schema, and what it would return.

  3. Take one of your existing MCP tool calls and ask: what would change if this tool were replaced by an A2A agent? What decisions would the agent make that the tool currently doesn't? When would the tool be the better choice?

  4. Draw a diagram of your system showing which connections are MCP (agent-to-tool), which are handoffs (internal control transfer), which are orchestration edges, and where A2A boundaries might exist in the future.

Completion checkpoint

You have:

  • A clear understanding of the four-way distinction: MCP, A2A, handoffs, and orchestration
  • The ability to identify which pattern fits a given integration scenario
  • At least one concrete scenario where A2A would add value over simpler alternatives
  • A diagram showing the protocol boundaries in your current system

Reflection prompts

  • In your current system, is there any integration that's awkward as a tool call (MCP) but would be more natural as an agent interaction (A2A)?
  • What are the debugging and observability challenges of A2A compared to internal handoffs? How would you trace a task that crosses an agent boundary?
  • The A2A protocol is still evolving. What's the risk of building on it now vs. waiting for it to mature? How would you manage that risk?

What's next

Thread and Workflow Memory. Coordination is one axis; persistence is the next. The next lesson gives the system memory inside a session and across a workflow.

References

Start here

  • A2A Protocol — the current Agent-to-Agent protocol documentation and specification overview

Build with this

Deep dive

Your Notes
GitHub Sync

Sync your lesson notes to a private GitHub Gist. If you have not entered a token yet, the sync button will open the GitHub token modal.
Glossary
API (Application Programming Interface)Foundational terms
A structured way for programs to communicate. In this context, usually an HTTP endpoint you call to interact with an LLM.
AST (Abstract Syntax Tree)Foundational terms
A tree representation of source code structure. Used by parsers like Tree-sitter to understand code as a hierarchy of functions, classes, and statements. You'll encounter this more deeply in the Code Retrieval module, but the concept appears briefly in retrieval fundamentals.
BM25 (Best Match 25)Foundational terms
A classical ranking function for keyword search. Scores documents by term frequency and inverse document frequency. Often competitive with or complementary to vector search.
ChunkingFoundational terms
Splitting a document into smaller pieces for indexing and retrieval. Chunk boundaries significantly affect retrieval quality. Split at the wrong place and your retrieval will return half a function or the end of one paragraph glued to the start of another.
Context engineeringFoundational terms
The discipline of selecting, packaging, and budgeting the information a model sees at inference time. Prompts, retrieved evidence, tool results, memory, and state are all parts of context. Context engineering is arguably the core skill of AI engineering. Bigger context windows are not a substitute for better context selection.
Context rotFoundational terms
Degradation of output quality caused by stale, noisy, or accumulated context. Symptoms include stale memory facts, conflicting retrieved evidence, bloated prompt history, and accumulated instructions that contradict each other. A form of technical debt in AI systems.
Context windowFoundational terms
The maximum number of tokens an LLM can process in a single request (input + output combined).
EmbeddingFoundational terms
A fixed-length numeric vector representing a piece of text. Used for similarity search: texts with similar meanings have nearby embeddings.
EndpointFoundational terms
A specific URL path that accepts requests and returns responses (e.g., POST /v1/chat/completions).
GGUFFoundational terms
A file format for quantized models used by llama.cpp and Ollama. When you see a model name like qwen2.5:7b-q4_K_M, the suffix indicates the quantization scheme. GGUF supports mixed quantization (different precision for different layers) and is the most common format for local inference.
HallucinationFoundational terms
When a model generates content that sounds confident but isn't supported by the evidence it was given, or fabricates details that don't exist. Not the same as "any wrong answer"; a model that misinterprets ambiguous instructions gave a bad answer but didn't hallucinate. Common causes: weak prompt, missing context, context rot, model limitation, or retrieval failure.
InferenceFoundational terms
Running a trained model to generate output from input. What happens when you call an API. Most AI engineering work is inference-time work: building systems around models, not training them. Use "inference," not "inferencing."
JSON (JavaScript Object Notation)Foundational terms
A lightweight text format for structured data. The lingua franca of API communication.
Lexical searchFoundational terms
Finding items by matching keywords or terms. Includes BM25, TF-IDF (Term Frequency–Inverse Document Frequency), and simple keyword matching. Returns exact term matches, not semantic similarity.
LLM (Large Language Model)Foundational terms
A neural network trained on large text corpora that generates text by predicting the next token. The core technology behind AI engineering; every tool, pattern, and pipeline in this curriculum runs on top of one.
MetadataFoundational terms
Structured information about a document or chunk (file path, language, author, date, symbol type). Used for filtering retrieval results.
Neural networkFoundational terms
A computing system loosely inspired by biological neurons, built from layers of mathematical functions that transform inputs into outputs. LLMs are a specific type of neural network (transformers) trained on text. You don't need to understand neural network internals to do AI engineering, but knowing the term helps when reading external resources.
Reasoning modelFoundational terms
A model optimized for complex multi-step planning, math, and logic (e.g., o3, o4-mini). Slower and more expensive but better on hard problems. Sometimes called "LRM" (large reasoning model), but "reasoning model" is the more consistent term across provider docs.
RerankingFoundational terms
A second-pass scoring step that re-orders retrieved results using a more expensive model. Improves precision after an initial broad retrieval.
SchemaFoundational terms
A formal description of the shape and types of a data structure. Used to validate inputs and outputs.
SLM (small language model)Foundational terms
A compact model (typically 1-7B parameters) that runs on consumer hardware with lower cost, latency, and better privacy (e.g., Phi, small Llama variants, Gemma). The right choice when privacy, offline operation, predictable cost, or low latency matter more than peak capability.
System promptFoundational terms
A special message that sets the model's behavior, role, and constraints for a conversation.
TemperatureFoundational terms
A parameter controlling output randomness. Lower values produce more deterministic output; higher values produce more varied output. Does not affect the model's intelligence.
TokenFoundational terms
The basic unit an LLM processes. Not a word. Tokens are sub-word fragments. "unhappiness" might be three tokens: "un", "happi", "ness". Token count determines cost and context window usage.
Top-kFoundational terms
The number of results returned from a retrieval query. "Top-5" means the five highest-scoring results.
Top-p (nucleus sampling)Foundational terms
An alternative to temperature for controlling output diversity. Selects from the smallest set of tokens whose cumulative probability exceeds p.
Vector searchFoundational terms
Finding items by proximity in embedding space (nearest neighbors). Returns "similar" results, not "exact match" results.
vLLM (virtual LLM)Foundational terms
An inference serving engine (not a model) that hosts open-weight models behind an OpenAI-compatible HTTP endpoint. Infrastructure layer, not model layer. Relevant when moving from hosted APIs to self-hosting.
WeightsFoundational terms
The learned parameters inside a model. Changed during training, fixed during inference.
Workhorse modelFoundational terms
A general-purpose LLM optimized for speed and broad capability (e.g., GPT-4o-mini, Claude Haiku, Gemini Flash). The default for most tasks. When someone says "LLM" without qualification, they usually mean this.
BaselineBenchmark and Harness terms
The first measured performance of your system on a benchmark. Everything else is compared against this. Without a baseline, you can't tell whether a change helped.
BenchmarkBenchmark and Harness terms
A fixed set of questions or tasks with known-good answers, used to measure system performance over time.
Run logBenchmark and Harness terms
A structured record (typically JSONL) of every system run: what input was given, what output was produced, what tools were called, how long it took, and what it cost. The raw data that evals, telemetry, and cost analysis are built from.
A2A (Agent-to-Agent protocol)Agent and Tool Building terms
An open protocol for peer-to-peer agent collaboration. Agents discover each other's capabilities and delegate or negotiate tasks as equals. Different from MCP (which connects agents to tools, not to other agents) and from handoffs (which transfer control within one system).
AgentAgent and Tool Building terms
A system where an LLM decides which tools to call, observes results, and iterates until a task is complete. Agent = model + tools + control loop.
Control loopAgent and Tool Building terms
The code that manages the agent's cycle: send prompt, check for tool calls, execute tools, append results, repeat or finish.
HandoffAgent and Tool Building terms
Passing control from one agent or specialist to another within an orchestrated system.
MCP (Model Context Protocol)Agent and Tool Building terms
An open protocol for exposing tools, resources, and prompts to AI applications in a standardized way. Connects agents to capabilities (tools and data), not to other agents.
Tool calling / function callingAgent and Tool Building terms
The model's ability to request execution of a specific function with structured arguments, rather than just generating text.
Context compilation / context packingCode Retrieval terms
The process of selecting and assembling the smallest useful set of evidence for a specific task. Not "dump everything retrieved into the prompt."
GroundingCode Retrieval terms
Tying model assertions to specific evidence. A grounded answer cites what it found; an ungrounded answer asserts without evidence.
Hybrid retrievalCode Retrieval terms
Combining multiple retrieval methods (e.g., vector search + keyword search + metadata filters) and merging or reranking the results.
Knowledge graphCode Retrieval terms
A data structure that stores entities and their relationships explicitly (e.g., "function A calls function B," "module X imports module Y"). Useful for traversal and dependency reasoning. One retrieval strategy among several, often overused when simpler metadata or adjacency tables would suffice.
RAG (Retrieval-Augmented Generation)Code Retrieval terms
A pattern where the model's response is grounded in retrieved external evidence rather than relying solely on its training data.
Symbol tableCode Retrieval terms
A mapping of code identifiers (functions, classes, variables) to their locations and metadata.
Tree-sitterCode Retrieval terms
An incremental parsing library that builds ASTs for source code. Used in this curriculum for code-aware chunking and symbol extraction.
Context packRAG and Grounded Answers terms
A structured bundle of evidence assembled for a specific task, with metadata about provenance, relevance, and token budget.
Evidence bundleRAG and Grounded Answers terms
A collection of retrieved items grouped for a specific sub-task, with enough metadata to evaluate whether the evidence is relevant and sufficient.
Retrieval routingRAG and Grounded Answers terms
Deciding which retrieval strategy or method to use for a given query. Different questions need different retrieval methods.
EvalObservability and Evals terms
A structured test that measures system quality. Not the same as training. Evals measure, they don't change the model.
Harness (AI harness / eval harness)Observability and Evals terms
The experiment and evaluation framework around your model or agent. It runs benchmark tasks, captures outputs, logs traces, grades results, and compares system versions. It turns ad hoc "try it and see" into repeatable, comparable experiments. Typically includes: input dataset, prompt and tool configuration, model/provider selection, execution loop, logging, grading, and artifact capture.
LLM-as-judgeObservability and Evals terms
Using a language model to evaluate or grade the output of another model or system. Useful for scaling evaluation beyond manual review, but requires rubric quality, judge consistency checks, and human spot-checking. Not a replacement for exact-match checks where they apply.
OpenTelemetry (OTel)Observability and Evals terms
An open standard for collecting and exporting telemetry data (traces, metrics, logs). Vendor-agnostic.
RAGASObservability and Evals terms
A specific eval framework for retrieval-augmented generation. Measures metrics like faithfulness, relevance, and context precision. One tool example, not a foundational concept. Learn the metrics first, then the tool.
SpanObservability and Evals terms
A single operation within a trace (e.g., one tool call, one retrieval query). Traces are made of spans.
TelemetryObservability and Evals terms
Structured data about system behavior: what happened, when, how long it took, what it cost. Includes traces, metrics, and events.
TraceObservability and Evals terms
A structured record of one complete run through the system, including all steps, tool calls, and decisions.
Long-term memoryOrchestration and Memory terms
Persistent facts that survive across conversations. Requires write policies to manage what gets stored, updated, or deleted.
OrchestrationOrchestration and Memory terms
Explicit control over how tasks are routed, delegated, and synthesized across multiple agents or specialists.
RouterOrchestration and Memory terms
A component that decides which specialist or workflow path to use for a given query.
SpecialistOrchestration and Memory terms
An agent or workflow tuned for a narrow task (e.g., "code search," "documentation lookup," "test generation"). Specialists are composed by an orchestrator.
Thread memoryOrchestration and Memory terms
Conversation state that persists within a single session or thread.
Workflow memoryOrchestration and Memory terms
Intermediate state that persists within a multi-step task but doesn't survive beyond the workflow's completion.
Catastrophic forgettingOptimization terms
When fine-tuning causes a model to lose capabilities it had before training. The model gets better at the fine-tuned task but worse at tasks it previously handled. PEFT methods like LoRA reduce this risk by freezing original weights.
DistillationOptimization terms
Training a smaller (student) model to reproduce the behavior of a larger (teacher) model on a specific task.
DPO (Direct Preference Optimization)Optimization terms
A method for preference-based model optimization that's simpler than RLHF, training the model directly on preference pairs without a separate reward model.
Fine-tuningOptimization terms
Updating a model's weights on task-specific data to change its behavior permanently. An umbrella term that includes SFT, instruction tuning, RLHF, DPO, and other techniques. See the fine-tuning landscape table in Lesson 8.3 for how these relate.
Full fine-tuningOptimization terms
Updating all of a model's parameters during training, as opposed to PEFT methods that update only a small subset. Requires significantly more GPU memory and compute. Produces the most thorough adaptation but carries higher risk of catastrophic forgetting.
Inference serverOptimization terms
Software (like vLLM or Ollama) that hosts a model and serves inference requests.
Instruction tuningOptimization terms
A specific application of SFT where the training data consists of instruction-response pairs. This is how base models become chat models: the technique is SFT, the data format is instructions. Not a separate technique from SFT.
LoRA (Low-Rank Adaptation)Optimization terms
A parameter-efficient fine-tuning method that trains small adapter matrices instead of updating all model weights. Dramatically reduces GPU memory and compute requirements.
Parameter countOptimization terms
The number of learned weights in a model, commonly expressed in billions (e.g., "7B" = 7 billion parameters). Determines memory requirements (roughly 2 bytes per parameter at FP16) and broadly correlates with capability, though training quality and architecture matter as much as size. See Model Selection and Serving for sizing guidance.
PEFT (Parameter-Efficient Fine-Tuning)Optimization terms
A family of methods (including LoRA) that fine-tune a small subset of parameters instead of the full model.
Preference optimizationOptimization terms
Training methods (RLHF, DPO) that use human or automated preference signals to improve model behavior. "This output is better than that output" rather than "this is the correct output."
QLoRA (Quantized LoRA)Optimization terms
LoRA applied to a quantized (compressed) base model. Further reduces memory requirements, enabling fine-tuning on consumer hardware.
QuantizationOptimization terms
Reducing the precision of model weights (e.g., FP16 → INT4) to shrink memory usage and increase inference speed at some quality cost. A 7B model at FP16 needs ~14 GB VRAM; quantized to 4-bit, it fits in ~4 GB. Common formats include GGUF (llama.cpp/Ollama), GPTQ and AWQ (vLLM/HuggingFace). See Model Selection and Serving for format details and tradeoffs.
OverfittingOptimization terms
When a model memorizes training examples instead of learning generalizable patterns. The model performs well on training data but poorly on new inputs. Detected by monitoring validation loss alongside training loss.
RLHF (Reinforcement Learning from Human Feedback)Optimization terms
A training method that uses human preference signals to improve model behavior through a reward model. More complex than DPO (requires training a separate reward model) but offers more control over the optimization objective.
SFT (Supervised Fine-Tuning)Optimization terms
Fine-tuning using input-output pairs where the desired output is known. The most common fine-tuning approach.
TRL (Transformer Reinforcement Learning)Optimization terms
A Hugging Face library for training language models with reinforcement learning, SFT, and other optimization methods.
Consumer chat appCross-cutting terms
The browser or desktop product meant for human conversation (ChatGPT, Claude, HuggingChat). Useful for experimentation, but not the same as API access.
Developer platformCross-cutting terms
The provider's API, billing, API-key, and developer-docs surface. This is what you need for this learning path.
Hosted APICross-cutting terms
The provider runs the model for you and you call it over HTTP.
Local inferenceCross-cutting terms
You run the model on your own machine.
ProviderCross-cutting terms
The company or service that hosts a model API you call from code.
Prompt cachingCross-cutting terms
Reusing computation from repeated prompt prefixes to reduce latency and cost on subsequent requests with the same prefix.
Rate limitingCross-cutting terms
Constraints on how many API requests you can make per unit of time. An operational concern that affects system design and cost.
Token budgetCross-cutting terms
The maximum number of tokens you allocate for a specific part of the context (e.g., "retrieval evidence gets at most 4K tokens"). A context engineering tool for preventing any single component from dominating the context window.