The next article comes from two weblog posts by Drew Breunig: “How Lengthy Contexts Fail” and “How you can Repair Your Contexts.”
Managing Your Context Is the Key to Profitable Brokers
As frontier mannequin context home windows proceed to develop,1 with many supporting as much as 1 million tokens, I see many excited discussions about how long-context home windows will unlock the brokers of our desires. In spite of everything, with a big sufficient window, you may merely throw every part right into a immediate you may want—instruments, paperwork, directions, and extra—and let the mannequin care for the remainder.
Lengthy contexts kneecapped RAG enthusiasm (no want to search out the very best doc when you may match all of it within the immediate!), enabled MCP hype (join to each software and fashions can do any job!), and fueled enthusiasm for brokers.2
However in actuality, longer contexts don’t generate higher responses. Overloading your context could cause your brokers and purposes to fail in stunning methods. Contexts can develop into poisoned, distracting, complicated, or conflicting. That is particularly problematic for brokers, which depend on context to collect info, synthesize findings, and coordinate actions.
Let’s run by the methods contexts can get out of hand, then evaluate strategies to mitigate or fully keep away from context fails.
Context Poisoning
Context poisoning is when a hallucination or different error makes it into the context, the place it’s repeatedly referenced.
The DeepMind staff known as out context poisoning within the Gemini 2.5 technical report, which we broke down beforehand. When taking part in Pokémon, the Gemini agent would sometimes hallucinate, poisoning its context:
An particularly egregious type of this problem can happen with “context poisoning”—the place many elements of the context (targets, abstract) are “poisoned” with misinformation in regards to the recreation state, which might usually take a really very long time to undo. Because of this, the mannequin can develop into fixated on reaching not possible or irrelevant targets.
If the “targets” part of its context was poisoned, the agent would develop nonsensical methods and repeat behaviors in pursuit of a aim that can’t be met.
Context Distraction
Context distraction is when a context grows so lengthy that the mannequin over-focuses on the context, neglecting what it discovered throughout coaching.
As context grows throughout an agentic workflow—because the mannequin gathers extra info and builds up historical past—this collected context can develop into distracting moderately than useful. The Pokémon-playing Gemini agent demonstrated this drawback clearly:
Whereas Gemini 2.5 Professional helps 1M+ token context, making efficient use of it for brokers presents a brand new analysis frontier. On this agentic setup, it was noticed that because the context grew considerably past 100k tokens, the agent confirmed a bent towards favoring repeating actions from its huge historical past moderately than synthesizing novel plans. This phenomenon, albeit anecdotal, highlights an vital distinction between long-context for retrieval and long-context for multistep, generative reasoning.
As a substitute of utilizing its coaching to develop new methods, the agent grew to become fixated on repeating previous actions from its intensive context historical past.
For smaller fashions, the distraction ceiling is far decrease. A Databricks examine discovered that mannequin correctness started to fall round 32k for Llama 3.1-405b and earlier for smaller fashions.
If fashions begin to misbehave lengthy earlier than their context home windows are stuffed, what’s the purpose of tremendous massive context home windows? In a nutshell: summarization3 and truth retrieval. When you’re not doing both of these, be cautious of your chosen mannequin’s distraction ceiling.
Context Confusion
Context confusion is when superfluous content material within the context is utilized by the mannequin to generate a low-quality response.
For a minute there, it actually appeared like everybody was going to ship an MCP. The dream of a robust mannequin, related to all your companies and stuff, doing all of your mundane duties felt inside attain. Simply throw all of the software descriptions into the immediate and hit go. Claude’s system immediate confirmed us the way in which, because it’s largely software definitions or directions for utilizing instruments.
However even when consolidation and competitors don’t sluggish MCPs, context confusion will. It turns on the market may be such a factor as too many instruments.
The Berkeley Perform-Calling Leaderboard is a tool-use benchmark that evaluates the power of fashions to successfully use instruments to reply to prompts. Now on its third model, the leaderboard reveals that each mannequin performs worse when supplied with a couple of software.4 Additional, the Berkeley staff, “designed eventualities the place not one of the supplied capabilities are related…we count on the mannequin’s output to be no operate name.” But, all fashions will sometimes name instruments that aren’t related.
Looking the function-calling leaderboard, you may see the issue worsen because the fashions get smaller:
A hanging instance of context confusion may be seen in a latest paper that evaluated small mannequin efficiency on the GeoEngine benchmark, a trial that options 46 completely different instruments. When the staff gave a quantized (compressed) Llama 3.1 8b a question with all 46 instruments, it failed, though the context was effectively throughout the 16k context window. However after they solely gave the mannequin 19 instruments, it succeeded.
The issue is, if you happen to put one thing within the context, the mannequin has to concentrate to it. It could be irrelevant info or unnecessary software definitions, however the mannequin will take it under consideration. Giant fashions, particularly reasoning fashions, are getting higher at ignoring or discarding superfluous context, however we frequently see nugatory info journey up brokers. Longer contexts allow us to stuff in additional data, however this potential comes with downsides.
Context Conflict
Context conflict is while you accrue new info and instruments in your context that conflicts with different info within the context.
It is a extra problematic model of context confusion. The unhealthy context right here isn’t irrelevant, it straight conflicts with different info within the immediate.
A Microsoft and Salesforce staff documented this brilliantly in a latest paper. The staff took prompts from a number of benchmarks and “sharded” their info throughout a number of prompts. Consider it this manner: Typically, you may sit down and sort paragraphs into ChatGPT or Claude earlier than you hit enter, contemplating each vital element. Different instances, you may begin with a easy immediate, then add additional particulars when the chatbot’s reply isn’t passable. The Microsoft/Salesforce staff modified benchmark prompts to seem like these multistep exchanges:
All the knowledge from the immediate on the left facet is contained throughout the a number of messages on the correct facet, which might be performed out in a number of chat rounds.
The sharded prompts yielded dramatically worse outcomes, with a median drop of 39%. And the staff examined a variety of fashions—OpenAI’s vaunted o3’s rating dropped from 98.1 to 64.1.
What’s occurring? Why are fashions performing worse if info is gathered in levels moderately than all of sudden?
The reply is context confusion: The assembled context, containing the whole lot of the chat trade, comprises early makes an attempt by the mannequin to reply the problem earlier than it has all the knowledge. These incorrect solutions stay current within the context and affect the mannequin when it generates its ultimate reply. The staff writes:
We discover that LLMs usually make assumptions in early turns and prematurely try to generate ultimate options, on which they overly rely. In less complicated phrases, we uncover that when LLMs take a flawed flip in a dialog, they get misplaced and don’t recuperate.
This doesn’t bode effectively for agent builders. Brokers assemble context from paperwork, software calls, and from different fashions tasked with subproblems. All of this context, pulled from numerous sources, has the potential to disagree with itself. Additional, while you connect with MCP instruments you didn’t create there’s a higher likelihood their descriptions and directions conflict with the remainder of your immediate.
Learnings
The arrival of million-token context home windows felt transformative. The power to throw every part an agent may want into the immediate impressed visions of superintelligent assistants that would entry any doc, join to each software, and keep good reminiscence.
However, as we’ve seen, larger contexts create new failure modes. Context poisoning embeds errors that compound over time. Context distraction causes brokers to lean closely on their context and repeat previous actions moderately than push ahead. Context confusion results in irrelevant software or doc utilization. Context conflict creates inner contradictions that derail reasoning.
These failures hit brokers hardest as a result of brokers function in precisely the eventualities the place contexts balloon: gathering info from a number of sources, making sequential software calls, participating in multi-turn reasoning, and accumulating intensive histories.
Fortuitously, there are answers!
Mitigating and Avoiding Context Failures
Let’s run by the methods we are able to mitigate or keep away from context failures fully.
Every thing is about info administration. Every thing within the context influences the response. We’re again to the previous programming adage of “rubbish in, rubbish out.” Fortunately, there’s loads of choices for coping with the problems above.
RAG
Retrieval-augmented technology (RAG) is the act of selectively including related info to assist the LLM generate a greater response.
As a result of a lot has been written about RAG, we’re not going to cowl it right here past saying: It’s very a lot alive.
Each time a mannequin ups the context window ante, a brand new “RAG is lifeless” debate is born. The final vital occasion was when Llama 4 Scout landed with a 10 million token window. At that measurement, it’s actually tempting to suppose, “Screw it, throw all of it in,” and name it a day.
However, as we’ve already lined, if you happen to deal with your context like a junk drawer, the junk will affect your response. If you wish to be taught extra, right here’s a new course that appears nice.
Device Loadout
Device loadout is the act of choosing solely related software definitions so as to add to your context.
The time period “loadout” is a gaming time period that refers back to the particular mixture of talents, weapons, and tools you choose earlier than a degree, match, or spherical. Normally, your loadout is tailor-made to the context—the character, the extent, the remainder of your staff’s make-up, and your individual ability set. Right here, we’re borrowing the time period to explain deciding on probably the most related instruments for a given job.
Maybe the best strategy to choose instruments is to use RAG to your software descriptions. That is precisely what Tiantian Gan and Qiyao Solar did, which they element of their paper “RAG MCP.” By storing their software descriptions in a vector database, they’re in a position to choose probably the most related instruments given an enter immediate.
When prompting DeepSeek-v3, the staff discovered that deciding on the correct instruments turns into essential when you will have greater than 30 instruments. Above 30, the descriptions of the instruments start to overlap, creating confusion. Past 100 instruments, the mannequin was just about assured to fail their take a look at. Utilizing RAG strategies to pick out fewer than 30 instruments yielded dramatically shorter prompts and resulted in as a lot as 3x higher software choice accuracy.
For smaller fashions, the issues start lengthy earlier than we hit 30 instruments. One paper we touched on beforehand, “Much less is Extra,” demonstrated that Llama 3.1 8b fails a benchmark when given 46 instruments, however succeeds when given solely 19 instruments. The problem is context confusion, not context window limitations.
To handle this problem, the staff behind “Much less is Extra” developed a strategy to dynamically choose instruments utilizing an LLM-powered software recommender. The LLM was prompted to purpose about “quantity and sort of instruments it ‘believes’ it requires to reply the consumer’s question.” This output was then semantically searched (software RAG, once more) to find out the ultimate loadout. They examined this technique with the Berkeley Perform-Calling Leaderboard, discovering Llama 3.1 8b efficiency improved by 44%.
The “Much less is Extra” paper notes two different advantages to smaller contexts—diminished energy consumption and velocity—essential metrics when working on the edge (that means, operating an LLM in your telephone or PC, not on a specialised server). Even when their dynamic software choice technique failed to enhance a mannequin’s outcome, the facility financial savings and velocity beneficial properties had been definitely worth the effort, yielding financial savings of 18% and 77%, respectively.
Fortunately, most brokers have smaller floor areas that solely require a couple of hand-curated instruments. But when the breadth of capabilities or the quantity of integrations must increase, all the time think about your loadout.
Context Quarantine
Context quarantine is the act of isolating contexts in their very own devoted threads, every used individually by a number of LLMs.
We see higher outcomes when our contexts aren’t too lengthy and don’t sport irrelevant content material. One strategy to obtain that is to interrupt our duties up into smaller, remoted jobs—every with its personal context.
There are many examples of this tactic, however an accessible write-up of this technique is Anthropic’s weblog submit detailing its multi-agent analysis system. They write:
The essence of search is compression: distilling insights from an enormous corpus. Subagents facilitate compression by working in parallel with their very own context home windows, exploring completely different elements of the query concurrently earlier than condensing a very powerful tokens for the lead analysis agent. Every subagent additionally supplies separation of considerations—distinct instruments, prompts, and exploration trajectories—which reduces path dependency and allows thorough, impartial investigations.
Analysis lends itself to this design sample. When given a query, a number of brokers can establish and individually immediate a number of subquestions or areas of exploration. This not solely accelerates the knowledge gathering and distillation (if there’s compute out there), nevertheless it retains every context from accruing an excessive amount of info or info not related to a given immediate, delivering increased high quality outcomes:
Our inner evaluations present that multi-agent analysis techniques excel particularly for breadth-first queries that contain pursuing a number of impartial instructions concurrently. We discovered {that a} multi-agent system with Claude Opus 4 because the lead agent and Claude Sonnet 4 subagents outperformed single-agent Claude Opus 4 by 90.2% on our inner analysis eval. For instance, when requested to establish all of the board members of the businesses within the Info Expertise S&P 500, the multi-agent system discovered the proper solutions by decomposing this into duties for subagents, whereas the single-agent system failed to search out the reply with sluggish, sequential searches.
This method additionally helps with software loadouts, because the agent designer can create a number of agent archetypes with their very own devoted loadout and directions for how one can make the most of every software.
The problem for agent builders, then, is to search out alternatives for remoted duties to spin out onto separate threads. Issues that require context-sharing amongst a number of brokers aren’t notably suited to this tactic.
In case your agent’s area is in any respect suited to parallelization, make sure to learn the entire Anthropic write-up. It’s wonderful.
Context Pruning
Context pruning is the act of eradicating irrelevant or in any other case unneeded info from the context.
Brokers accrue context as they fireplace off instruments and assemble paperwork. At instances, it’s price pausing to evaluate what’s been assembled and take away the cruft. This could possibly be one thing you job your essential LLM with or you may design a separate LLM-powered software to evaluate and edit the context. Or you may select one thing extra tailor-made to the pruning job.
Context pruning has a (comparatively) lengthy historical past, as context lengths had been a extra problematic bottleneck within the pure language processing (NLP) area previous to ChatGPT. Constructing on this historical past, a present pruning technique is Provence, “an environment friendly and strong context pruner for query answering.”
Provence is quick, correct, easy to make use of, and comparatively small—just one.75 GB. You possibly can name it in a couple of traces, like so:
from transformers import AutoModel
provence = AutoModel.from_pretrained("naver/provence-reranker-debertav3-v1", trust_remote_code=True)
# Learn in a markdown model of the Wikipedia entry for Alameda, CA
with open('alameda_wiki.md', 'r', encoding='utf-8') as f:
alameda_wiki = f.learn()
# Prune the article, given a query
query = 'What are my choices for leaving Alameda?'
provence_output = provence.course of(query, alameda_wiki)Provence edited the article, chopping 95% of the content material, leaving me with solely this related subset. It nailed it.
One may make use of Provence or an analogous operate to cull paperwork or your entire context. Additional, this sample is a robust argument for sustaining a structured5 model of your context in a dictionary or different type, from which you assemble a compiled string prior to each LLM name. This construction would come in useful when pruning, permitting you to make sure the primary directions and targets are preserved whereas the doc or historical past sections may be pruned or summarized.
Context Summarization
Context summarization is the act of boiling down an accrued context right into a condensed abstract.
Context summarization first appeared as a software for coping with smaller context home windows. As your chat session got here near exceeding the utmost context size, a abstract could be generated and a brand new thread would start. Chatbot customers did this manually in ChatGPT or Claude, asking the bot to generate a brief recap that might then be pasted into a brand new session.
Nevertheless, as context home windows elevated, agent builders found there are advantages to summarization in addition to staying throughout the whole context restrict. As we’ve seen, past 100,000 tokens the context turns into distracting and causes the agent to depend on its collected historical past moderately than coaching. Summarization might help it “begin over” and keep away from repeating context-based actions.
Summarizing your context is straightforward to do, however exhausting to good for any given agent. Understanding what info needs to be preserved and detailing that to an LLM-powered compression step is essential for agent builders. It’s price breaking out this operate as its personal LLM-powered stage or app, which lets you accumulate analysis information that may inform and optimize this job straight.
Context Offloading
Context offloading is the act of storing info exterior the LLM’s context, normally through a software that shops and manages the information.
This is perhaps my favourite tactic, if solely as a result of it’s so easy you don’t imagine it is going to work.
Once more, Anthropic has write-up of the method, which particulars their “suppose” software, which is mainly a scratchpad:
With the “suppose” software, we’re giving Claude the power to incorporate a further pondering step—full with its personal designated house—as a part of attending to its ultimate reply… That is notably useful when performing lengthy chains of software calls or in lengthy multi-step conversations with the consumer.
I actually respect the analysis and different writing Anthropic publishes, however I’m not a fan of this software’s title. If this software had been known as scratchpad, you’d know its operate instantly. It’s a spot for the mannequin to write down down notes that don’t cloud its context and can be found for later reference. The title “suppose” clashes with “prolonged pondering” and needlessly anthropomorphizes the mannequin… however I digress.
Having an area to log notes and progress works. Anthropic reveals pairing the “suppose” software with a domain-specific immediate (which you’d do anyway in an agent) yields vital beneficial properties: as much as a 54% enchancment in opposition to a benchmark for specialised brokers.
Anthropic recognized three eventualities the place the context offloading sample is beneficial:
- Device output evaluation. When Claude must rigorously course of the output of earlier software calls earlier than performing and may have to backtrack in its method;
- Coverage-heavy environments. When Claude must observe detailed pointers and confirm compliance; and
- Sequential determination making. When every motion builds on earlier ones and errors are pricey (usually present in multi-step domains).
Takeaways
Context administration is normally the toughest a part of constructing an agent. Programming the LLM to, as Karpathy says, “pack the context home windows good,” well deploying instruments, info, and common context upkeep, is the job of the agent designer.
The important thing perception throughout all of the above ways is that context isn’t free. Each token within the context influences the mannequin’s conduct, for higher or worse. The large context home windows of recent LLMs are a robust functionality, however they’re not an excuse to be sloppy with info administration.
As you construct your subsequent agent or optimize an present one, ask your self: Is every part on this context incomes its maintain? If not, you now have six methods to repair it.
Footnotes
- Gemini 2.5 and GPT-4.1 have 1 million token context home windows, massive sufficient to throw Infinite Jest in there with loads of room to spare.
- The “Lengthy type textual content” part within the Gemini docs sum up this optmism properly.
- The truth is, within the Databricks examine cited above, a frequent method fashions would fail when given lengthy contexts is that they’d return summarizations of the supplied context whereas ignoring any directions contained throughout the immediate.
- When you’re on the leaderboard, take note of the “Dwell (AST)” columns. These metrics use real-world software definitions contributed to the product by enterprise, “avoiding the drawbacks of dataset contamination and biased benchmarks.”
- Hell, this whole listing of ways is a robust argument for why it’s best to program your contexts.
