Hardware FixRecommendedDevice not working? Your driver may be the problemCheck updates for common hardware issues.Fix DriversThanksgiving DealsAmazon USThanksgiving week: Black Friday picks start earlyAmazon US: current deals, useful picks and tech finds.Check DealsWindows FixRecommendedWindows errors stealing your time? Find the fix fastScan stability, cleanup and performance issues.Fix Now×
FREEDOM251

Freedom251 › Blog

Why Coding Agents Consume So Many Tokens—and How to Stop Runaway Loops

By Freedom251 Team · · 12 min read

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.

AI coding agents consume tokens across an entire workflow, not just in the prompt you type. An agent may plan, inspect a repository, call tools, edit files, run checks, interpret failures, and retry; every model turn can resend instructions, conversation history, and tool output. The practical answer is to treat each run like a bounded workflow: define observable completion criteria, cap steps and retries, limit context and tools, track progress, and enforce budgets outside the model.

A high token total is not automatically a malfunction. A difficult migration or debugging task can require legitimate exploration. The warning sign is rising usage without corresponding state changes, such as the same tool call repeated, the same error returned, or no additional files and checks completed.

As an Amazon Associate I earn from qualifying purchases.

Free tools Windows power users keep installed

One-click scans. No signup required.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Why do AI coding agents cost more than simple LLM prompts?

A simple LLM prompt is often one request and one response. An agent is a control loop. It decides what to do, invokes a tool, reads the result, updates its working context, and decides again. Planning, repository search, file reads, edits, test runs, error analysis, and retries each add work and usually another model call. JetBrains’ cost-management guidance describes this difference as the reason agentic work can cost more than an isolated prompt.

Suppose an agent receives a task, lists the repository, opens several files, edits one module, runs a checker, sees an error, searches for a related configuration, changes the module again, and reruns the checker. The visible request may be short, but the run contains many decisions and tool results. If each later turn includes the original instructions plus the accumulated transcript, the agent repeatedly processes material it has already seen.

Popular on Amazon
Sale
Silkland 40Gbps USB C 90 Degree Right Angle Adapter 2 Pack, 240W USB4 | Male to Female Extender, 8K@60Hz, Thunderbolt 4/3, Steam Deck, ROG Ally, MacBook, Switch 2, iPhone 18/17/16/15, Quest 3 VR
  • 【90 Degree Full Function USB C Adapter】Supports the latest USB4 standard for up to 40Gbps of bandwidth, 8K video support, and up to 240W Fast charging; Support USB PD 3.1 Extended Power Range (EPR) including 140W, 180W and 240W. Your productivity is easily doubled. *Warm Tip: For the best fit, please make sure the USB-C port is fully exposed; thick or bulky cases may affect compatibility.
  • 【Reliable Quality】Made from high-quality, scratch-resistant aluminum alloy, delicate metal frame. get excellent wear resistance and heat dissipation. Non-Slip Design for easier insertion and removal.
  • 【Small and portable】Ultra thin size, ultra compact design, a petite physique that matches the minimalist design of your modern laptops. Coming with a slim figure that won't block the port next to the one it's plugged into.
  • 【Compatibility】Widely work with most usb c, thunderbolt 4/3 devices. Compatibilit for ROG Ally, Steam Deck, Switch, MacBook Pro, Quest 3 VR headset, Surface Book 2, iPhone 18/17/16/15, Samsung Galaxy S23/S22/S21, Note 20/10, Google Pixel, ChromeBook, etc.
  • 【Excellence Customer Support】If you experience any issues or cable failure, simply reach out to our dedicated customer service team. We provide a prompt response within 18 hours, providing you with the necessary technical support to resolve any concerns.

Costs therefore depend on the full harness: model pricing, input and output token rates, whether prompts are cached, context-compaction rules, tool schemas, parallel workers, retry policy, and the task itself. Token count is a useful operational signal, but it is not a universal price or a proof that a run is broken.

How does the agent loop turn a small request into a large bill?

The common sequence is:

  1. Plan: translate the request into subtasks and choose an initial action.
  2. Inspect: search directories, read files, inspect configuration, or query version-control state.
  3. Act: call a shell, editor, build tool, test runner, or another service.
  4. Observe: parse the tool’s output and compare it with the goal.
  5. Revise: edit files, change the command, or investigate a new dependency.
  6. Validate: run tests, linters, type checks, or a deployment preview.
  7. Retry or escalate: repeat a failed action, ask for missing information, or stop.

Each transition can produce input and output tokens. Tool output is especially easy to underestimate: a recursive search, compiler trace, generated diff, or large log can become part of the next prompt. A broad repository scan repeated after every failure multiplies that overhead.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

There is also hidden decision overhead. The model must select among available tools and interpret their schemas. Exposing ten tools when only two are relevant increases the instructions and choices carried through the loop. Parallel agents add their own prompts, results, coordination messages, and often a final synthesis turn.

Does accumulated context really make token use grow faster?

It can. In a full-context replay pattern, turn 2 includes turn 1’s material, turn 3 includes turns 1 and 2, and so on. The total amount of input processed over many turns can then grow roughly quadratically with the number of steps rather than linearly. Vercel’s LLM cost-management guide uses this model to explain why long loops can become disproportionately expensive.

That is a model of one context strategy, not a billing law. Providers may cache stable prefixes, frameworks may summarize or compact history, and an agent may select only relevant files for each call. Input and output prices also differ by model. Actual spend can therefore be lower or higher than a simple quadratic estimate.

Context reduction must be done carefully. Removing an irrelevant log or duplicate file listing usually helps. Removing the acceptance criteria, security constraints, migration assumptions, or the failure state can force the agent to rediscover information or make an unsafe change, creating more work than the trimming saved.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

What makes a loop runaway instead of merely thorough?

A runaway loop lacks a reliable exit or cannot recognize progress. JetBrains identifies repeated actions, missing stop conditions, and vague goals as common loop failure modes. Typical causes include:

Also popular on Amazon
Sale
Anker USB C Hub, 5-in-1 USBC to HDMI Splitter with 4K Display | 85W Max PD, 1 Powered USB-C 5Gbps & 2 Powered USB-A 3.0 5Gbps Data Ports, for MacBook Pro, MacBook Air, Dell and More (Black)
  • 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
  • Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
  • Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
  • HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
  • What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
  • Underspecified success: “Fix the API” gives no observable condition that ends the run.
  • Unbounded retries: the same command is attempted after an identical error without changing inputs or strategy.
  • Unstable validation: a checker emits noisy or nondeterministic failures, so the agent never receives a stable success signal.
  • Lost state: each turn forgets what was already searched or changed and starts over.
  • Tool failures: a timeout, permission error, or malformed response triggers repeated calls.
  • Over-broad scope: the agent keeps expanding from one bug into unrelated cleanup.
  • Uncontrolled delegation: agents create more agents, communicate without limits, or run many branches in parallel.

A hard task can legitimately have many calls. The difference is whether calls change the state toward completion. Activity without progress is the operational definition worth monitoring.

How can I tell that a coding-agent run is wasting tokens?

Look at a trace, not just the final invoice. Useful warning signs include repeated tool names with identical inputs, recurring error signatures, increasing retry counts, no new file diff, an unchanged task-state record, and a flat count of resolved subtasks while token usage rises. A single large response is not enough evidence; correlate usage with artifacts and outcomes.

Capture at least these fields for every iteration:

Field What it tells you
Iteration and timestamp How many turns occurred and whether time is accumulating.
Selected action and tool input Whether the agent is repeating the same operation.
Tool output summary and error signature Whether failures are identical, new, or resolved.
Input, output, and cached-token counts Which part of the call is driving usage.
Files, checks, or subtasks changed Concrete evidence of progress.
Retry count and branch/agent count How repetition or fan-out is multiplying work.
Stop reason Whether the run completed, failed, timed out, or was manually halted.

When repeated actions coincide with no measurable state change, pause the run. Inspect the latest tool output and the last successful state transition. Then correct the prompt or tool configuration, resume from the preserved state if the next action is clear, or stop and escalate with the failure evidence attached.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Do not ask the model to predict its own final spend and treat that prediction as a guarantee. The 2026 preprint “How Do AI Agents Spend Your Money?” reports substantial run-to-run variation and weak prediction of total token use in its evaluated coding tasks. An alert based on observed usage and progress is safer than an optimistic forecast.

What limits should every coding-agent workflow have?

Use several independent limits. A step cap catches logical loops; a deadline catches hung tools; retry and tool budgets prevent one failure from consuming the run; and a spend quota protects the wider account. Any one limit can miss a different failure mode.

Control Recommended definition When it fires What to preserve
Maximum steps A fixed number of model/tool iterations per task The next iteration would exceed the cap Transcript summary, files changed, and last tool result
Wall-clock deadline An elapsed-time limit independent of model requests Tools hang or the run runs unusually slowly Active command, logs, and checkpoint
Retry budget A small bound per command or error class The same failure repeats without a changed input Error signature and attempted remedies
Tool-call budget A total and, where useful, per-tool maximum Searches, builds, or external calls dominate Tool outputs needed for a human or later run
Concurrency and recursion caps Maximum parallel workers, child agents, and delegation depth Fan-out or agent creation starts multiplying Parent/child task map and partial results
Token or spend quota A provider or organization-level hard ceiling Usage crosses the permitted budget Usage records and a resumable checkpoint

These controls should be enforced outside the model where possible. Microsoft’s security guidance for unbounded AI consumption says: “For Agent DoS, rate-limit agent creation, cap the total number of concurrent agents and subagents, bound recursion depth, throttle and cap inter-agent communication, and place strict limits on fan-out, queue depth, tool calls, and workflow invocations.” The same principle applies to accidental wallet-draining loops.

How do I define a stopping rule that the agent can actually use?

Write completion as observable evidence, not an intention. For example: “The parser accepts the three supplied fixtures, the type checker passes, and the diff is limited to the parser and its tests.” Also define failure and escalation: “After two attempts with the same error signature, stop, summarize the error, and ask for a decision.”

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Also popular on Amazon
Sale
Anker USB C Hub, 7in1 Multi-Port USB Adapter, 4K@60Hz USBC to HDMI Splitter | 85W Max PD, 2 USB 3.0 & 1 USBC Data Ports, SD/TF Card Reader, for Laptops/Mac and Other Type C Devices (Charger Not Included)
  • Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
  • Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
  • Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
  • Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
  • What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.

A useful task contract contains:

  • Scope: repositories, directories, files, and dependencies the agent may touch.
  • Acceptance checks: exact commands or artifacts that demonstrate success.
  • Forbidden changes: generated files, public interfaces, production data, or unrelated formatting.
  • Retry policy: what input must change before another attempt is allowed.
  • Escalation rule: when to request credentials, clarification, a design choice, or human review.
  • Stop output: a concise summary of changes, checks, remaining failures, and the next safe action.

State should be durable rather than hidden in a long transcript. A small task manifest can record completed subtasks, files inspected, commands run, known failures, and the current hypothesis. Before selecting another action, the agent reads that manifest and updates it after meaningful work. This prevents a context reset or retry from redoing the same exploration.

How should I control context without making the agent less reliable?

Start by measuring which context is actually used. Keep the task contract, relevant code, dependency versions, security requirements, recent diffs, and the latest failure evidence. Remove duplicate directory listings, stale logs, unrelated files, and verbose tool output after extracting the facts needed for the next decision.

Prefer structured summaries to raw transcripts. A test runner can return failing test names, the first useful stack-trace section, and a count of passing checks instead of thousands of repeated lines. A repository search can return file paths and matching snippets with a limit. Store the full output for audit, but send only the relevant slice to the next model call.

Use checkpoints or compaction at natural boundaries: after discovery, after an edit, and after validation. Each checkpoint should state what was learned and what remains, not merely truncate the oldest messages. If a safety-critical constraint might be lost, repeat it explicitly in the compacted context.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Context selection is usually a better first optimization than blindly shortening the user prompt. A shorter prompt that causes an incorrect edit, a failed build, and three retries is not cheaper overall. JetBrains recommends measuring per-run usage and removing unnecessary context and tools while retaining information needed for reliable work.

Which tools and delegation patterns create avoidable spend?

Expose only tools that can contribute to the stated task. Every additional tool adds schema and decision overhead, and a tool that can mutate production systems raises the cost of mistakes. Separate read-only discovery from write or deployment actions, and require an explicit approval or policy check before high-impact operations.

Bound fan-out when using multiple agents. Set a maximum number of children, a recursion depth, a concurrency limit, and a message or token budget for coordination. Give each child a narrow deliverable and a deadline; otherwise several workers may inspect the same files and return overlapping reports that a parent must synthesize.

Also popular on Amazon
Sale
BENFEI USB C Hub 5-in-1 with 4K HDMI(Certified), 100W Power Delivery, 3 USB-A, Silicone Cable, Aluminum Case Compatible with MacBook Pro/Air, iPad Pro, iMac, iPhone 15 Pro/Pro Max, XPS, Thinkpad
  • Portable and powerful USB-C HUB: BENFEI USB Type-C HUB, with super-soft and knot-free silicone woven design cable, meets most mobile office needs. Compact, lightweight, stylish, and powerful portable USB C Hub equipped with 1 x HDMI port, 1 x 100W charging, and 3 x USB ports. 18-month warranty, 24-hour response, to ensure you feel at ease when using our product.
  • Design centered on comfort and reliability: Thanks to BENFEI's end-to-end in-house cable production capability, in-house PCBA and assembly capability, using the industry's most advanced silicone woven design and process, 20cm cable in length, no knots, super-soft, the HUB is easy to use in all scenarios: laptop, tablet, stand etc. Super-soft, 25000+ life cycles, to meet your daily carrying and office needs.
  • 100W Charging: Support up to 90W USB C pass-through charging via Type-C port to keep your laptop powered. 10W is reserved for other interface operations. No data and video function on the Type-C port.
  • 4K HDMI Display: The HDMI port supports media display at resolutions up to 4K 30Hz, keeping every incredible moment detailed and ultra vivid. Please note that the C port of the Host device needs to support video output.
  • Transfer Files in Seconds: Transfer files and from your laptop at speeds up to 10 Gbps with USB A 3.2 port. Extra 2 USB A 2.0 ports are perfectly for your keyboards and mouse.

Parallelism can reduce elapsed time for independent checks, but it can also multiply token use and produce conflicting edits. Use it for genuinely independent, read-heavy work, then designate one process to integrate results. Do not allow a child agent to create further agents unless that recursion is explicitly required and bounded.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Where should budget enforcement live?

Use layered enforcement because no single layer sees every risk.

  1. Inside the framework: implement step, retry, tool, state, and recursion limits close to the loop. This layer understands task progress and can return a useful stop reason.
  2. At a gateway or provider: apply token, rate, concurrency, and model-access quotas. This catches loops that bypass an application’s local counters.
  3. At the organization level: set per-team or per-workload spend ceilings, alerts, and circuit breakers. This prevents one project from consuming a shared budget.
  4. In observability: retain traces, usage, error classes, and outcomes so limits can be tuned from evidence.

Alert thresholds should be based on representative historical runs, with separate ranges for small fixes, migrations, and large refactors. A soft alert can request human review; a hard quota should stop requests regardless of what the model says. Preserve a resumable checkpoint so a budget stop is a controlled pause rather than lost work.

Should I switch to a smaller model or reduce reasoning effort?

Possibly, but model changes are a quality decision as well as a cost decision. A cheaper model or lower reasoning setting may reduce tokens per call while increasing retries, incorrect edits, or failed validation. Compare total cost to successful completion on representative tasks, not token count in isolation.

Use the least expensive configuration that reliably meets the acceptance checks, and reserve a stronger model for ambiguous design work, difficult debugging, or a recovery attempt. Keep the same external limits for every model. A more capable model can still loop if the goal is vague or the tools are unbounded.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

The preprint “How Do AI Agents Spend Your Money?” reports up to 1,000 times as many tokens for evaluated agentic tasks as for its code-reasoning and code-chat baselines, and up to a 30-fold difference between runs of the same task. Those are results from the authors’ evaluated settings, not a multiplier to apply to every coding agent. They support measuring real workloads and installing safeguards rather than relying on a single estimate.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

What savings are realistic from workflow optimization?

Do not promise a fixed percentage. Savings depend on the harness, task mix, provider, caching, and whether quality remains comparable. In the evaluated configurations of the 2026 TokenEconomizer preprint, the authors report average token-cost reductions of 18.4–30.9% for coding-agent settings, with 6.4–21.2% reductions in their evaluated web-development and security-auditing settings. Those figures describe the paper’s interventions and experiments, not an individual user’s guaranteed result.

Also popular on Amazon
𝟒𝐏𝐚𝐜𝐤 𝐔𝐩𝐠𝐫𝐚𝐝𝐞𝐝 USB to USB C Adapter,Type-C Female to A Male Charger Converter for iPhone 18 Charger,iWatch Ultra SE 9 10,iPhone 17 Pro Max 16 15 14,Samsung S25,AirPods,iPad,Car,Xbox
  • 【Read before You Buy】 Note: This adapter is USB 2.0 version and cannot transmit HDM, VGA, or other video signals. Nevertheless, they are fully capable of charging devices,duplicating files, and can be seamlessly integrated with USB-C earphones, wireless network adapters, and SD card adapters.
  • 【Easy to Use and Safe】Thanks to modern manufacturing techniques, we've made these adapters sufficiently compact so that they can remain inserted in your device without noticeable protrusion or inconvenience. This means you don't have to constantly plug and unplug them, reducing the likelihood of wear and misalignment, and making them more convenient to carry around.
  • 【Reliable Quality】Engineered for enhanced durability, durable aluminum alloy shell material,get excellent wear resistance and heat dissipation,the male USB C and female USB connectors & the male USB and female USB C connectors are designed to be plugged and unplugged up to 20000 times. Provides data transfer, device charging, memory space expansion, and additional computer peripherals.
  • 【Think more for you】Do you need to connect a USB-C-enabled device to an odd device that only has USB-A ports? Then these USB-A male to USB-C female adapters are made for you.Simply insert the adapter into the USB-A port on the device, and you can instantly connect your USB-C device to the device that originally didn't have the correct connection port.
  • 【Trusted xiwxi Quality and Service】As a responsible brand seller, you will receive xiwxi's 12-month Quality assurance and lifetime friendly customer service, you're not just buying a product,you're investing in peace of mind with all-round quality and exceptional customer service.

In practice, the highest-leverage sequence is usually:

  1. Instrument input, output, cached tokens, tool calls, retries, elapsed time, and outcome.
  2. Find runs where usage rises while files, checks, or subtasks remain unchanged.
  3. Add or tighten an external step, deadline, retry, and spend limit.
  4. Fix the task contract and validator so completion and failure are unambiguous.
  5. Reduce redundant context and cap oversized tool output.
  6. Remove irrelevant tools and bound parallelism or delegation.
  7. Only then evaluate a different model or reasoning setting against the same quality criteria.

How do I configure a safe run before starting?

Use this preflight checklist:

  • Write one-sentence scope and measurable acceptance checks.
  • List allowed repositories, directories, tools, and mutation privileges.
  • Set maximum steps, wall-clock time, retries, tool calls, concurrency, and spend.
  • Define the repeated-error rule and the human-escalation path.
  • Enable per-run usage and trace logging, including a stop reason.
  • Choose an initial context set and a maximum output size for searches and logs.
  • Create a checkpoint location for task state, diffs, and validation results.
  • Decide which operations require approval before execution.

During the run, review progress at checkpoints rather than waiting for the final bill. If two successive iterations produce the same error and no state change, pause. If the agent reaches the step or spend limit, stop it, preserve the checkpoint, and resume only after changing the task, tool, or strategy that caused the stall.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

What is the practical rule for stopping runaway token use?

Stop treating token consumption as a prompt-length problem. It is a workflow-control problem: repeated model calls, accumulated context, retries, tools, and delegation determine the total. The safest design combines explicit success and failure criteria, durable progress state, bounded execution, selective context, limited tools and fan-out, and independent budgets with monitoring.

A run that spends more because it completes a difficult task can be valuable. A run that spends more while repeating the same action is a control failure. Measure both usage and progress, stop the latter early, and preserve enough state to recover without starting from zero.

Frequently Asked Questions

Can prompt caching eliminate the cost of a long agent loop?

Caching can reduce charges or processing for repeated prompt prefixes when a provider supports it, but it does not remove output tokens, tool calls, retries, or coordination work. Cache behavior and pricing differ by provider, so inspect per-call usage rather than assuming every repeated token is free.

What should an agent return when a budget limit stops it?

Require a machine-readable stop record containing the reason, completed subtasks, files changed, commands and results, unresolved errors, and a recommended next action. That record lets a person or a later run continue safely without replaying the entire transcript.

What’s actually slowing this PC down?

Pick the symptom - the matching free tool is one click away.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

How can I distinguish a slow legitimate build from a loop?

Compare state transitions and tool inputs over time. A long build that produces a new artifact or a different diagnostic is progressing; repeated commands with identical inputs, identical errors, and no diff indicate repetition even if each command eventually exits.

Are token limits enough to protect a shared development account?

No. A token quota may not cover concurrency spikes, child-agent creation, provider rate charges, or expensive external tools. Combine per-run limits with organization-level concurrency, model-access, spend, and workflow-invocation controls.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply. Prices as of November 20, 2026 12:44 pm GMT+0530.

Leave a Reply

Your email address will not be published. Required fields are marked *

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Recommended PC Tool
Recommended PC Tool
Crashes, No Sound, or Screen Glitches?Free driver scan
PC Slower Than It Used to Be?Free scan - under a minute

Two free Windows tools

One Free Minute Could Fix That PC

Before you go - each of these free tools takes about a minute and tackles what quietly slows a Windows PC down.

Special offer. View Outbyte info, uninstall instructions, EULA, and Privacy Policy.