Stop the run long enough to establish its real state, inspect the last tool result or error, and check whether the requested change already happened. Then classify the failure: transient service problems can justify a limited retry; invalid input, credentials, permissions, or tool configuration require a correction; repeated actions require a stop condition; and sensitive or blocked work may need a human decision.
The sequence below works for developers and for people using consumer-facing agents, although the labels and controls differ between products. Do not blindly resubmit a failed request until you have checked for partial completion and possible side effects.
As an Amazon Associate I earn from qualifying purchases.
What do I do when my AI agent gets stuck?
Use this order every time. It prevents the two most expensive recovery mistakes: repeating an action that already succeeded and retrying an error that cannot be fixed by waiting.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Pause and capture the current state. Save the visible task, run or session status, the latest turn, the last tool call, its arguments, the returned result, the exact error text, and any output already produced. In an API or SDK, record the structured error code and message rather than relying only on a shortened dashboard message.
- Check the target before replaying anything. Look at the file, record, message, deployment, ticket, database row, or other object the agent was supposed to change. A failed turn can still have created a file, sent a request, or changed a remote system. Compare the actual target with the intended end state and note which steps are complete.
- Determine whether the run is still active. A stream disconnect, browser refresh, or apparent stall does not necessarily mean the session failed. Retrieve the latest run or turn status if the product exposes it. If it is still running, wait for a terminal state or stop it safely; if it is complete, use the recorded result instead of starting a duplicate run.
- Classify the failure. Separate a temporary timeout, rate limit, or server failure from malformed input, expired credentials, missing permissions, an unavailable endpoint, a tool error, a failed session, or an approval pause. The remedy depends on this classification.
- Apply one bounded correction. Fix the invalid field, credential, permission, endpoint, tool argument, or instruction that caused the problem. For a suitable transient error, wait as directed by the service and retry with backoff. Set a maximum number of attempts or a deadline before you begin.
- Look for new evidence after each attempt. A retry is useful only when it produces a new result, a changed status, or information that narrows the cause. If the same call returns the same output, stop rather than escalating an accidental loop.
- Resume or escalate from the saved state. Continue unfinished work in the existing run when possible. Ask a person to decide when the next step is sensitive, ambiguous, irreversible, or blocked by missing information. Give support or an operator the IDs and evidence they need to investigate.
How to classify the failure before you retry
These categories are practical branches, not a universal error taxonomy. Products expose different status names, and many consumer interfaces do not show run, turn, or session identifiers.
#1 Best Overall
- [Ultra-Lightweight Ear Buds Designed for Small Ears] Each earbud weighs only 3.7g and features a compact, ergonomic in-ear design made especially for small ears. Secure, low-profile, and comfortable for workouts, all-day wear or overnight sleeping.The ultra-slim profile sits flush in your ear with no hard edges poking or rubbing, so you can even sleep on your side without discomfort.
- [Immersive Stereo Sound with TOZO OrigX Technology] TOZO OrigX tuning delivers clear vocals, balanced mids, and natural stereo sound for music, podcasts, and videos.
- [Long Battery Life for Daily Use] Get up to 7 hours of playtime on a single charge, with up to 32 hours total using the charging case—ideal for workdays, commuting, and extended listening sessions.
- [Bluetooth 5.3 & Stable Connection] Bluetooth 5.3 provides fast pairing, stable wireless performance, and reduced dropouts as you move around home or office.
- [Deep Bass with Clear Vocals] High-performance drivers produce punchy bass while keeping vocals clean and detailed for everyday listening.
| What you see | First response | Safe next step |
|---|---|---|
| Rate limit, timeout, temporary 5xx, or service unavailable | Check whether the operation completed and read any server retry timing. | Honor Retry-After when supplied; otherwise use exponential backoff with jitter and a maximum attempt count or deadline. |
| Invalid request or malformed parameters | Read the field-level error and compare it with the tool contract. | Correct the value or schema, then submit once. Replaying unchanged input will not repair it. |
| Credentials, permission, billing, or configuration failure | Verify the identity, scopes, project, account, quota, and endpoint configuration. | Fix access or configuration first; do not treat an authorization error as transient. |
| Tool error or unavailable endpoint | Inspect the tool name, arguments, integration health, and returned payload. | Repair or replace the tool, or return a clear failure to the agent so it can choose another valid path. |
| The agent repeats an action without progress | Stop the loop and compare successive tool calls and results. | Enforce an iteration limit, provide the missing environmental fact, change the approach, or ask a person. |
| Turn fails or a stream disconnects | Retrieve the latest session or turn status and inspect saved work. | Verify side effects and terminal status before resubmitting unfinished work. |
| Approval or safety-sensitive request | Treat the pause as deliberate, not as a crash. | Approve, reject, or clarify the action, then resume the saved run state. |
How to tell whether the agent already completed the action
A tool call can reach its destination even when the model receives a timeout, disconnect, or malformed response. Before a replay, inspect the destination using an independent read operation where possible.
- For files, check existence, modification time, size, and a content or checksum comparison.
- For database or SaaS records, query the record by a stable identifier and inspect its audit history.
- For messages or requests, check the provider’s sent, delivery, or job-status view.
- For deployments or infrastructure changes, read the current resource configuration and deployment event rather than trusting the agent’s last sentence.
- For payments, deletions, permission changes, or other irreversible effects, require an explicit human check before any retry.
Use an operation identifier, idempotency key, or destination-side deduplication when the service supports one. These mechanisms reduce duplicate effects, but they do not replace checking the actual target. The safe principle is to establish the present state first and ask the agent to perform only the missing transition.
Should you restart an AI agent after an error?
Not automatically. Restarting can discard useful context, duplicate side effects, or hide the original cause.
Continue the existing run when
- the session is healthy or merely waiting for a tool result;
- the run has a resumable state and the missing step is clear;
- an approval was requested and the decision can be supplied; or
- a transient failure occurred and the operation has been verified as incomplete.
Start a new session only when
- the session itself is reported failed or unrecoverable;
- the cause of failure has been corrected; and
- you have copied the relevant context, saved outputs, identifiers, and completed-step record into the new session.
A new session is not a fix for bad input, revoked credentials, an unavailable tool, or a missing approval. Correct the cause first, then resume the smallest unfinished unit of work.
Rank #2
- 【IP68 Waterproof Multifunctinal Smart Watch】Torjalph fitness activity tracker smart watch is IP68 waterproof technology ,you can wear it when swimming,handwashing,it's rain-proof which means that you perform multiple exercise modes outside without worrying about sweat or rain corroding your fitness smart watch,and this watch equipped with a 230 mAh large capacity battery magnetic charging ,and this sports activity tracker watch can be used 5-7 days If fully charged.
- 【Larger 1.83'' Full Screen Watch & Multi-dial Selection】1.83 inch full HD color touch screen smart watch for men / women provides excellent 240*284 resolution image quality ,This Bluetooth smart watch offers a better touch experience and remarkable immersion and ,more than 200 watch faces for you to choose from App"Da Fit ", you can change the dial style according to your mood to wear keeping the world on your wrist
- 【Accurate Activity Tracker Smart Watch 】sports watch activity tracker uses sweat and dust resistant breathable silicone strap, our Fitness Tracker Smart Watch for Android iOS Phones supports 12 sports modes,walking,running,bicycle,swimming,badminton,basketball,football,rope skipping. This bluetooth calls smart watch for men women is lightweight and stylish, bringing more convenient and relaxing sports experience while also providing more protection.
- 【Your 24 Hour Personal Daily Health Monitor】Fitness watch for women men full touch screen IP67 waterproof bluetooth smartwatch for Android iOS Phones with heart rate blood pressure monitor multifunction Andriod smart watch with steps counter sleep monitor sports fitness watch, monitoring blood oxygen saturation allows you to understand your physical condition in real time to track fitness progress ,to know your health status in time.
- 【More Functions Bluetooth Call Smart Watch】This fitness activity smart watch full screen for iPhone and Android phones includes a bunch of new features including:The bluetooth smart watch can hang up the phone, receive and display WhatsAPP /Text /Facebook /Twitter /Messenger reminder.Support sedentary reminder, alarm clock, stopwatch, timer,weather information,music control, remote camera control compatible with Android 5.0 and above or iOS 9.0 and above smartphones.
What to do when an agent keeps repeating the same action
Repetition usually means the agent is not receiving a reliable state change, cannot interpret the tool result, or has no stopping rule. Do not allow an unlimited loop.
- Stop at the configured limit. Set a maximum number of tool iterations, retries, or elapsed time. Anthropic’s engineering guidance describes maximum-iteration limits as a control for keeping agents bounded; Google’s troubleshooting guidance likewise recommends a maximum retry count.
- Compare calls, arguments, and results. Determine whether the arguments are identical, whether a supposedly successful result changed the environment, and whether the agent is seeing stale or ambiguous output.
- Check the tool contract. A tool should state what it does and return structured inputs and outputs. If success is represented only by vague text, the model may repeat the call because it cannot verify completion.
- Supply ground truth. Return the current state, error code, resource identifier, or fresh read result. Anthropic’s official engineering article puts the principle this way: “During execution, it’s crucial for the agents to gain “ground truth” from the environment at each step (such as tool call results or code execution) to assess its progress.”
- Change the plan or ask for help. If a valid tool cannot produce the needed state, switch to an available method or pause for a human. Repeating the same call is not a recovery strategy.
How to retry safely
Use backoff only for suitable transient failures
For a timeout, rate limit, or temporary server failure, wait before retrying and add random jitter so many workers do not retry simultaneously. Honor a server-provided Retry-After value. Define both an attempt limit and a time deadline; either one should end the retry loop.
Google’s current Gemini Python SDK documentation describes automatic retries of transient errors up to four times by default, with an initial delay of approximately one second and a maximum delay of 60 seconds. Those figures are implementation details for that SDK and version, not a cross-provider rule or a study statistic. Other SDKs and consumer products can behave differently, so inspect the relevant product’s current settings.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsDo not retry unchanged client errors
Malformed parameters, invalid model or tool names, expired credentials, missing permissions, quota or billing blocks, and policy rejections require a correction. Capture the error, fix the identified cause, and submit a new request only after the correction. If the message is unclear, reduce the request to a minimal reproducible call and inspect the tool’s schema and account configuration.
Rank #3
- [Low Noise High Fidelity] New bee guitar cord with 22AWG oxygen-free-copper center conductors provides enhanced signal transfer, the PVC insulation material and the high-density stranded copper double shield provide noise rejection for quiet operation
- [Ideal for Choice] New bee 1/4 straight and right angle metal connectors, ensures a secure connection in any instrument, pedal, or amp jack. L shape great fro on stage guitar stands
- [Wide Compatible] Suitable for mixer, power amplifier, microphone, effector, decoder, equalizer, electoronic organ, electric guitar, electric bass and other electronic equipments. Widely used in the installation of smart home, karaoke speaker, professional stage audio, etc
- [Easy Disassemble] Making incidental repairs quick and easy! Very nice, flexible cable
- [Dont Hesite to Order] 24 hours customer services and professional technology team are standing by
Make external operations safe to replay
Before retrying a write, use an idempotency key or a natural unique key when the endpoint supports it. If it does not, read the target and design the next request around the observed state. For a non-idempotent action such as sending, charging, deleting, or granting access, a human should confirm the result before replay.
What developers should build into an agent
A clear tool contract
Document each tool’s purpose, required fields, allowed values, authentication assumptions, side effects, and success and failure schemas. Anthropic’s tool-use documentation emphasizes that the application executes client-side tools and sends their results back in a tool_result; the model does not execute that application code by itself. Your application therefore needs to validate inputs, execute the operation, and return an unambiguous result.
Observable state and tracing
Persist request, run, session, and turn identifiers; tool calls and results; status transitions; errors; approvals; and saved outputs. Operators should be able to answer what was attempted, what the environment did, and what remains without asking the model to reconstruct history from memory.
Bounded execution
Wrap retries and agent loops with maximum attempts, deadlines, cancellation, and a clear terminal state. Emit a specific reason when the limit is reached, such as retry_limit_exceeded or awaiting_human_decision, rather than silently starting another turn.
Rank #4
- Digital Family Calendar: BSIMB digital calendar wall touch screen puts an end to your chaotic life. Designed for busy families, the digital calendar keeps you organised and in step with your family. Assign different colours and tasks to each family member to make family scheduling easier and more intuitive.
- Easy Setup and Auto-Sync: Plug in, connect to WLAN and auto-syn your electric calendar in minutes. Digital wall calendar compatible with Google calendar, iCloud calendar, Outlook calendar, Cozi calendar and Yahoo calendar. Add events directly on the device or through the ecalendar app. Electronic calendar stays current whether you're at work or traveling—no manual updates needed.
- All-In-One Home Assistant: Smart calendar builds habits and encourage independence with personalised chores. Display daily meal plan to reduce the daily ‘What's for dinner?’ stress. To-do lists keep track of everything you want. Turning chores into valuable achievements with stars and rewards, making life easier and more enjoyable.
- Manage Anytime Anywhere: BSIMB electronic calendar for family stay connected with the powerful ecalendar app. Access and manage your digital wall calendar anytime, anywhere with free features like stars & rewards, photo screensaver, chore charts, meals plans, to-do lists etc. Electronic planner paid features like Magic Import tailor advanced tools for users with special needs.
- Stay in Touch with Loved Ones: Share photos or videos from your phone to electronic wall calendar with app, instantly update your life moment, and keep in touch with your family, especially those who are far away; you can also surprise your smart calendar digital wall by sending him a wish card; it's also great for keeping tech-illiterate elders in touch with the youngsters.
Human checkpoints
Require approval for sensitive tool calls, irreversible changes, ambiguous instructions, and blockers that cannot be resolved from available data. Save the run state while waiting. When the person responds, resume that state instead of creating a fresh task that might repeat earlier work.
Safe replay and recovery tests
Test timeout-after-commit, duplicate delivery, stale reads, revoked credentials, unavailable tools, malformed tool output, stream disconnects, and approval pauses. Verify that the recovery path reads the target state and continues from the first incomplete step.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What if you are using a consumer-facing agent?
You may not see the underlying session or turn IDs. Translate the same controls into the interface you have:
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match- Stop the current generation or automation if it is actively looping.
- Copy the visible error and note the task, time, and last displayed action.
- Check the target app, file, inbox, repository, or account for a partial change.
- Correct the prompt, attachment, login, permission, or connected-tool setting identified by the error.
- Resume only the unfinished step, stating what has already happened.
- If the agent still repeats or the action is sensitive, stop and contact the product’s support team with the error, timestamp, account or run identifier, and evidence of any side effect.
Not every consumer product supports resumable runs, approval workflows, idempotency keys, or detailed tracing. Where those controls are absent, manual verification and a smaller, reversible task are the safest substitutes.
When should you stop retrying and ask a person?
Escalate when the maximum retry or iteration limit is reached, the same result repeats without new evidence, the target state is uncertain, an external side effect may have occurred, credentials or permissions cannot be changed by the agent, the tool or endpoint remains unavailable, or the next action is legally, financially, or operationally sensitive.
Include the request, session, and turn IDs when available. Also provide the tool name, sanitized input, returned error, timestamps, current target state, completed actions, number of retries, and whether an external side effect occurred. This evidence lets an operator resume or compensate for the failed action instead of starting from an unreliable assumption.
A compact recovery checklist
- Have I recorded the exact error and latest tool result?
- Is the run still active, complete, paused for approval, or failed?
- Did the target change despite the error?
- Is the cause transient, invalid input, access, tool, state, or approval?
- Will another attempt create a duplicate side effect?
- What new evidence will this retry produce?
- What is my retry count or deadline?
- Can I resume the existing state, or has the session itself failed?
- What information must a human receive if I stop?
The reliable habit is simple: observe first, classify second, correct or retry within limits, and resume only unfinished work. An agent that pauses with evidence is safer than one that appears busy while repeating an unverified action.
Recommended Free Tools
Quick Recap
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.
