The next article comes from two weblog posts by Drew Breunig: “How Lengthy Contexts Fail” and “How one 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 goals. In spite of everything, with a big sufficient window, you possibly can merely throw all the things right into a immediate you would possibly want—instruments, paperwork, directions, and extra—and let the mannequin maintain the remaining.

Lengthy contexts kneecapped RAG enthusiasm (no want to search out the very best doc when you possibly can 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 functions to fail in suprising methods. Contexts can change into poisoned, distracting, complicated, or conflicting. That is particularly problematic for brokers, which depend on context to collect data, synthesize findings, and coordinate actions.

Let’s run by means of the methods contexts can get out of hand, then evaluation strategies to mitigate or totally 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 workforce 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 often hallucinate, poisoning its context:

An particularly egregious type of this challenge can happen with “context poisoning”—the place many elements of the context (targets, abstract) are “poisoned” with misinformation concerning the recreation state, which may typically take a really very long time to undo. Consequently, the mannequin can change into fixated on attaining unattainable 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 objective that can not 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 data and builds up historical past—this amassed context can change into distracting reasonably 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 an inclination towards favoring repeating actions from its huge historical past reasonably than synthesizing novel plans. This phenomenon, albeit anecdotal, highlights an essential 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 in depth context historical past.

For smaller fashions, the distraction ceiling is way 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 reality retrieval. If you happen to’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 strong mannequin, related to all your providers 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 best way, because it’s principally software definitions or directions for utilizing instruments.

However even when consolidation and competitors don’t gradual 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 answer prompts. Now on its third model, the leaderboard exhibits that each mannequin performs worse when supplied with a couple of software.⁴ Additional, the Berkeley workforce, “designed situations the place not one of the supplied features are related…we anticipate the mannequin’s output to be no operate name.” But, all fashions will often name instruments that aren’t related.

Looking the function-calling leaderboard, you possibly can 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 totally different instruments. When the workforce gave a quantized (compressed) Llama 3.1 8b a question with all 46 instruments, it failed, despite the fact that the context was nicely throughout the 16k context window. However after they solely gave the mannequin 19 instruments, it succeeded.

The issue is, for those who put one thing within the context, the mannequin has to concentrate to it. It could be irrelevant data or useless software definitions, however the mannequin will take it under consideration. Massive fashions, particularly reasoning fashions, are getting higher at ignoring or discarding superfluous context, however we regularly see nugatory data journey up brokers. Longer contexts allow us to stuff in additional data, however this capability comes with downsides.

Context Conflict

Context conflict is if you accrue new data and instruments in your context that conflicts with different data within the context.

It is a extra problematic model of context confusion. The dangerous context right here isn’t irrelevant, it instantly conflicts with different data within the immediate.

A Microsoft and Salesforce workforce documented this brilliantly in a latest paper. The workforce took prompts from a number of benchmarks and “sharded” their data throughout a number of prompts. Consider it this manner: Generally, you would possibly sit down and kind paragraphs into ChatGPT or Claude earlier than you hit enter, contemplating each vital element. Different instances, you would possibly begin with a easy immediate, then add additional particulars when the chatbot’s reply isn’t passable. The Microsoft/Salesforce workforce modified benchmark prompts to appear to be these multistep exchanges:

All the data from the immediate on the left aspect is contained throughout the a number of messages on the suitable aspect, 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 workforce 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 data is gathered in levels reasonably than abruptly?

The reply is context confusion: The assembled context, containing the whole thing of the chat trade, comprises early makes an attempt by the mannequin to reply the problem earlier than it has all the data. These incorrect solutions stay current within the context and affect the mannequin when it generates its remaining reply. The workforce writes:

We discover that LLMs typically make assumptions in early turns and prematurely try and generate remaining options, on which they overly rely. In less complicated phrases, we uncover that when LLMs take a unsuitable flip in a dialog, they get misplaced and don’t get well.

This doesn’t bode nicely 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, if you hook up with MCP instruments you didn’t create there’s a better 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 all the things an agent would possibly want into the immediate impressed visions of superintelligent assistants that might entry any doc, join to each software, and keep excellent 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 reasonably than push ahead. Context confusion results in irrelevant software or doc utilization. Context conflict creates inside contradictions that derail reasoning.

These failures hit brokers hardest as a result of brokers function in precisely the situations the place contexts balloon: gathering data from a number of sources, making sequential software calls, partaking in multi-turn reasoning, and accumulating in depth histories.

Luckily, there are answers!

Mitigating and Avoiding Context Failures

Let’s run by means of the methods we will mitigate or keep away from context failures totally.

Every little thing is about data administration. Every little 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 data 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 useless” debate is born. The final vital occasion was when Llama 4 Scout landed with a 10 million token window. At that dimension, it’s actually tempting to suppose, “Screw it, throw all of it in,” and name it a day.

However, as we’ve already coated, for those who 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 stage, match, or spherical. Often, your loadout is tailor-made to the context—the character, the extent, the remainder of your workforce’s make-up, and your personal talent set. Right here, we’re borrowing the time period to explain choosing probably the most related instruments for a given job.

Maybe the best approach 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 workforce discovered that choosing the suitable instruments turns into essential when you could have greater than 30 instruments. Above 30, the descriptions of the instruments start to overlap, creating confusion. Past 100 instruments, the mannequin was nearly assured to fail their take a look at. Utilizing RAG methods 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 challenge, the workforce behind “Much less is Extra” developed a approach to dynamically choose instruments utilizing an LLM-powered software recommender. The LLM was prompted to motive about “quantity and kind of instruments it ‘believes’ it requires to reply the person’s question.” This output was then semantically searched (software RAG, once more) to find out the ultimate loadout. They examined this methodology 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—decreased energy consumption and velocity—essential metrics when working on the edge (that means, operating an LLM in your cellphone or PC, not on a specialised server). Even when their dynamic software choice methodology failed to enhance a mannequin’s outcome, the ability financial savings and velocity features have been definitely worth the effort, yielding financial savings of 18% and 77%, respectively.

Fortunately, most brokers have smaller floor areas that solely require just a few hand-curated instruments. But when the breadth of features or the quantity of integrations must develop, 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 approach 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 publish detailing its multi-agent analysis system. They write:

The essence of search is compression: distilling insights from an unlimited corpus. Subagents facilitate compression by working in parallel with their very own context home windows, exploring totally different points of the query concurrently earlier than condensing an important tokens for the lead analysis agent. Every subagent additionally offers separation of considerations—distinct instruments, prompts, and exploration trajectories—which reduces path dependency and allows thorough, unbiased 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 quickens the data gathering and distillation (if there’s compute accessible), but it surely retains every context from accruing an excessive amount of data or data not related to a given immediate, delivering greater high quality outcomes:

Our inside evaluations present that multi-agent analysis methods excel particularly for breadth-first queries that contain pursuing a number of unbiased 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 inside analysis eval. For instance, when requested to establish all of the board members of the businesses within the Data 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 gradual, 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 methods to 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 significantly 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 data from the context.

Brokers accrue context as they hearth off instruments and assemble paperwork. At instances, it’s value pausing to evaluate what’s been assembled and take away the cruft. This might be one thing you job your fundamental LLM with or you might design a separate LLM-powered software to evaluation and edit the context. Or you might select one thing extra tailor-made to the pruning job.

Context pruning has a (comparatively) lengthy historical past, as context lengths have been a extra problematic bottleneck within the pure language processing (NLP) discipline previous to ChatGPT. Constructing on this historical past, a present pruning methodology is Provence, “an environment friendly and sturdy context pruner for query answering.”

Provence is quick, correct, easy to make use of, and comparatively small—only one.75 GB. You possibly can name it in just a few strains, 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, reducing 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 the whole 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 turn out to be 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 can 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 may 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 amassed historical past reasonably than coaching. Summarization can assist it “begin over” and keep away from repeating context-based actions.

Summarizing your context is straightforward to do, however onerous to excellent for any given agent. Realizing what data ought to be preserved and detailing that to an LLM-powered compression step is essential for agent builders. It’s value breaking out this operate as its personal LLM-powered stage or app, which lets you gather analysis knowledge that may inform and optimize this job instantly.

Context Offloading

Context offloading is the act of storing data 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 would work.

Once more, Anthropic has a great write up of the approach, which particulars their “suppose” software, which is principally a scratchpad:

With the “suppose” software, we’re giving Claude the power to incorporate an extra pondering step—full with its personal designated area—as a part of attending to its remaining reply… That is significantly useful when performing lengthy chains of software calls or in lengthy multi-step conversations with the person.

I actually respect the analysis and different writing Anthropic publishes, however I’m not a fan of this software’s identify. If this software have been known as scratchpad, you’d know its operate instantly. It’s a spot for the mannequin to jot down down notes that don’t cloud its context and can be found for later reference. The identify “suppose” clashes with “prolonged pondering” and needlessly anthropomorphizes the mannequin… however I digress.

Having an area to log notes and progress works. Anthropic exhibits pairing the “suppose” software with a domain-specific immediate (which you’d do anyway in an agent) yields vital features: as much as a 54% enchancment in opposition to a benchmark for specialised brokers.

Anthropic recognized three situations the place the context offloading sample is beneficial:

1. Device output evaluation. When Claude must rigorously course of the output of earlier software calls earlier than appearing and would possibly must backtrack in its method;
2. Coverage-heavy environments. When Claude must comply with detailed tips and confirm compliance; and
3. Sequential resolution making. When every motion builds on earlier ones and errors are pricey (typically 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 excellent,” well deploying instruments, data, and common context upkeep, is the job of the agent designer.

The important thing perception throughout all of the above techniques is that context shouldn’t be free. Each token within the context influences the mannequin’s conduct, for higher or worse. The large context home windows of contemporary LLMs are a strong functionality, however they’re not an excuse to be sloppy with data administration.

As you construct your subsequent agent or optimize an current one, ask your self: Is all the things on this context incomes its preserve? If not, you now have six methods to repair it.

---

Footnotes

1. 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.
2. The “Lengthy type textual content” part within the Gemini docs sum up this optmism properly.
3. The truth is, within the Databricks examine cited above, a frequent manner 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.
4. If you happen to’re on the leaderboard, take note of the “Stay (AST)” columns. These metrics use real-world software definitions contributed to the product by enterprise, “avoiding the drawbacks of dataset contamination and biased benchmarks.”
5. Hell, this whole checklist of techniques is a robust argument for why you must program your contexts.