L
Week one · the habits that hold

Your first week with an AI tool

Eighty-eight habits for the week the novelty wears off and the work gets real.

The rules, section by section

Before the rules

The first day is about not losing anything. The first week is different, because by now the tool is in the work rather than beside it.

What breaks in week one is quieter than what breaks on day one. Nothing is destroyed. Instead a summary quietly omits the section that mattered, a job runs somewhere nobody is looking, or a conversation drifts far enough that the answer is about a task you abandoned twenty messages ago.

Eighty-eight habits, in the order they tend to arrive. Every one carries the reason behind it and the same request phrased two ways.

A. Proving it read what it says it read

A claim to have read something costs nothing to produce and arrives in the same confident tone either way.

You attach a 60-page contract and ask what the termination clause says. Back comes an answer about termination, fluent and specific, drawn from every contract ever written rather than from yours.

Twelve habits here, and they share one move: ask for something only a reader could produce. A file list, a size, a sentence from page forty.

A46 · What to do

Ask the tool to display a list of the files it says it read, along with a rough size for each file.

Why

When you see a list with file names and sizes, you can check quickly if the right files were actually read. A plain claim is easy to miss or fake, but a list is proof you can verify.

What not to do

Don't accept a vague reply like 'I read your files' without any details. Without a file list and sizes, you cannot check if the tool skipped something or made a mistake.

Say it like this“Show me the files you read, with sizes: Rahul_Report.docx, Budget2024.xlsx.”
Not like this“Did you read my files?”
A47 · What to do

Ask the tool to confirm it has read each of the N files separately during your current session, not in a batch or silently in the background.

Why

When the tool says it read all files but does not show each one, you might wait for answers that never come. Separate reads make sure nothing is missed or delayed without you knowing.

What not to do

Don't trust a single 'all done' message if you gave it several files. When files are skipped without warning, you might waste days waiting for results that never appear.

Say it like this“Did you read all five files in this chat—ClientList.csv, Notes.txt, PlanA.docx, PlanB.docx, and Tasks.xlsx?”
Not like this“Did you read everything I uploaded?”
A48 · What to do

Check if the tool actually read your file itself, not repeated what another assistant or tool reported about it.

Why

A summary from another system can hide mistakes or missing parts. If the tool does not read the file itself, wrong details might get copied over and block you from seeing the real picture.

What not to do

Relying on a summary from another helper is risky, because if the tool skips reading the file itself, errors from the first summary stay hidden and keep causing trouble.

Say it like this“Did you open Q1_Presentation.pptx yourself, or did you use Maya's summary?”
Not like this“What did the other tool say about my file?”
A49 · What to do

Split large files into smaller parts and make the tool read every part, so nothing is skipped or missed.

Why

When a tool guesses the pattern from only part of a big file, it can miss bugs or important details that appear elsewhere. Reading each part closely avoids hidden mistakes.

What not to do

Don't let the tool read only the start or a few sections of a big file. Skipping parts leads to missed errors that could have been found if all sections were checked.

Say it like this“Read every section of Operations_Annual_Report_2023.pdf, not only the first 10 pages.”
Not like this“Scan the first part of my big report.”
A50 · What to do

Ask the tool to read the whole reference file, not the beginning, so you get a complete understanding.

Why

When only the start of a file is read, you might see small details but miss the overall structure. Reading the entire file helps avoid repeated mistakes and shows how everything fits together.

What not to do

Don't stop after reading the first part of a file. Missing the rest means you only see a small piece, which can lead to repeating errors or missing important sections.

Say it like this“Read the full TeamGuide2024.pdf, not the introduction.”
Not like this“Check the start of the TeamGuide.”
A51 · What to do

Pick a few important files to read with care, instead of trying to read every single file right away.

Why

Trying to read too many files at once means you pay less attention to each one. Focusing on the key files helps you get the details you need, without getting lost in less important ones.

What not to do

Don't ask the tool to read all files in a folder if you only need a few. Reading too many files at once wastes time and makes it harder to focus on what matters.

Say it like this“Read Budget2024.xlsx and MeetingNotes_March.txt first, before the rest.”
Not like this“Read every file in my Documents folder.”
A52 · What to do

Tell the tool to read the entire file or message, not only a section, when you need full understanding before work starts.

Why

Missing sections at the start means the tool relies on guesses or gaps, which leads to confusion, missed context, and errors. Reading everything once saves time compared to fixing misunderstandings later.

What not to do

Don't ask for a summary or answer before the tool has read the complete file. Skipping parts now means you will spend more time fixing mistakes that come from missing key information.

Say it like this“Read all of Q3_Report_2023.pdf before you answer.”
Not like this“Summarise Q3_Report_2023.pdf quickly.”
A53 · What to do

Tell the tool exactly which folder or file to search, not your whole computer or company drive, when you want it to find something.

Why

Large storage spaces slow the tool down or cause it to freeze. Searching a specific folder or file saves time and keeps your computer responsive, while searching everything risks long delays or crashes.

What not to do

Don't let the tool search your entire drive or cloud. A broad search can take an hour, stall your work, and might not even find the right file among thousands.

Say it like this“Search only the 2023_Invoices folder for 'Rohit Sharma'.”
Not like this“Find 'Rohit Sharma' in my files.”
A54 · What to do

Ask the tool what it already knows from a file before telling it to read that file again, unless you have updated the file since.

Why

Reading the same file twice without changes wastes time and does not add new understanding. Checking first avoids unnecessary rereading and keeps things moving efficiently.

What not to do

Don't tell the tool to reread a file unless you have changed something. Rereading without updates slows you down and does not improve results.

Say it like this“Tell me what you know from ProjectPlan2024.docx before reading it again.”
Not like this“Read ProjectPlan2024.docx again.”
A55 · What to do

Open and read the files or emails that will use your output before you start creating your work, so you know what is needed.

Why

Not checking what will use your work leads to missed requirements, wasted drafts, and extra calls to explain or fix things. Five minutes of reading avoids hours of rework.

What not to do

Don't start writing or creating before checking the email or slide deck where your work will go. Guessing what is needed usually means doing it twice.

Say it like this“Read the briefing email from Anjali before drafting the summary.”
Not like this“Write the summary now, then check the briefing.”
A56 · What to do

Look through all files or creators in the folder, not only the first few, when searching for the right version or source for your work.

Why

Some folders have many versions from different people. The correct file might be hidden further down. Skipping files risks using the wrong version and causes failed attempts.

What not to do

Don't pick the first or most recent file without checking others. The best version might be from a different creator, and missing it means wasted effort.

Say it like this“Check all project proposals in the ClientX_2024 folder before choosing one.”
Not like this“Use the first proposal you find in ClientX_2024.”
A57 · What to do

Ask the tool to show you the source file or section before it copies from it, so you can check it is using the right material.

Why

If the tool cannot display the start of the source, it might be guessing or paraphrasing instead of copying. Seeing the source lets you confirm accuracy before using the content.

What not to do

Don't let the tool copy from a source unless it shows you the actual section first. Otherwise, you risk errors or made-up content.

Say it like this“Show me the first paragraph of Budget2023.docx before copying.”
Not like this“Copy from Budget2023.docx.”

Try these — proving it read what it says it read

No score, and nothing to finish. Every wrong answer opens with why the wrong idea was tempting.

Question 1 of 12
 

B. Doing the same job many times over

The second time you run a job is where the money and the mistakes both live.

Three hundred rows worked yesterday. Today the same script runs against nine thousand, and the part that was cheap at three hundred is the part that empties the account.

Seventeen habits for repeated work: what changes when the count goes up, what has to be checked once rather than every time, and where a small run stops predicting a large one.

B58 · What to do

Send each call, or request, at the same time instead of waiting for one to finish before starting the next.

Why

Waiting for each call to finish before starting the next means every call takes as long as the slowest one. Sending them together means you only wait for the slowest, not the total of all.

What not to do

Don't write a loop where each call waits for the last to finish, since making every job wait can slow things down, especially when some calls are much slower than others.

Say it like this“Start all 12 calls to report.xlsx together, then wait for them all.”
Not like this“Send each call to report.xlsx one after another.”
B59 · What to do

Arrange your script so that more than five calls, or requests, run at the same time instead of one after another.

Why

Running calls one after another means the total time is the sum of every slow call. Letting them run at the same time means only the slowest matters, which can save hours for large jobs.

What not to do

Don't let your script process each of 20 files one by one, because waiting for one slow or stuck file could mean hours of delay.

Say it like this“Run all 15 calls for the March invoices at the same time.”
Not like this“Send 15 invoice calls in a row, one after the other.”
B60 · What to do

Send tasks, or requests, spaced out over a few seconds instead of all at once, to avoid overloading the system.

Why

Sending everything in one point creates a traffic jam. The system gets too many requests, slows down, and may even block itself, so everything takes longer than if spaced out.

What not to do

Don't hit 'send' on 40 updates to team_roster.csv in one second, since doing so can overwhelm the system and cause all of them to slow down or fail.

Say it like this“Send 10 updates to team_roster.csv every 10 seconds.”
Not like this“Send all 40 updates to team_roster.csv at once.”
B61 · What to do

Test how many workers, or parallel calls, your job can run before failures go above one percent, instead of guessing the number.

Why

Guessing the number of workers can cause too many failures. Testing lets you find the point where adding more workers makes things break, so you can pick a safe number.

What not to do

Don't set eight workers because it sounds right, as picking a number without checking can cause failures and wasted effort.

Say it like this“Test with four workers on sales_data.csv until failures stay below one percent.”
Not like this“Set workers to eight for sales_data.csv without testing.”
B62 · What to do

Reduce the number of workers, or parallel calls, when you see failures, instead of adding time gaps between requests.

Why

Adding pauses only spreads out the problem, but too many workers still overload the system. Cutting workers means fewer requests at once, which prevents overload and reduces failures.

What not to do

Don't add a five-second pause between 12 calls to payroll.pdf hoping it will fix errors, since this only delays things and doesn't solve the real problem.

Say it like this“Cut workers from 12 to six when payroll.pdf calls start failing.”
Not like this“Add a pause between each payroll.pdf call when failures happen.”
B63 · What to do

Let each planned call, or request, try only once in the main job. Handle failures separately, not by repeating inside the loop.

Why

Retrying failed calls inside the main block creates more requests during a problem, making the block worse and last longer. Handling failures in a separate step keeps things clear and manageable.

What not to do

Don't retry failed calls to summary.docx inside the main loop, because doing so can overwhelm the system and make everything much slower.

Say it like this“Log failed summary.docx calls and retry them after the main job finishes.”
Not like this“Retry summary.docx calls inside the main script loop.”
B64 · What to do

Organise your task so you can pick up from where it stopped if something fails, instead of starting over from the beginning.

Why

Network connections can drop for a few seconds, or the tool might time out. Not every failure means you hit a limit. A careful resume lets you continue without missing anything, instead of repeating everything.

What not to do

Don't restart the whole process every time something fails. Doing so wastes time and can double the workload, especially when only a few items were missed.

Say it like this“Resume processing invoices_2023.csv from line 1201 if the tool stops.”
Not like this“Start over with invoices_2023.csv if anything fails.”
B65 · What to do

Break your job into batches, test one batch first, and set a limit on each batch every time you run the task.

Why

A crash in the middle of a long run can erase hours of work. Batching and limiting each run means you only lose a small, recent chunk, not the whole job.

What not to do

Don't run a single massive job without limits or testing a batch. If it crashes, you will lose everything and have to repeat the whole task.

Say it like this“Process 100 rows from sales_q1.xlsx, test, then do the next 100 rows.”
Not like this“Process all 5,000 rows from sales_q1.xlsx at once.”
B66 · What to do

Save your progress after each batch finishes, instead of waiting until the whole job is done.

Why

Saving after each batch means you only lose the last batch if something fails. Waiting until the end risks losing all your work if the tool crashes before the final save.

What not to do

Don't wait to save until the entire job is finished. If something goes wrong at the last step, you will lose everything you have done.

Say it like this“Save processed data from clients_mumbai.csv after every 50 rows.”
Not like this“Save clients_mumbai.csv only after all rows are processed.”
B67 · What to do

Estimate how big the job is before you start, especially for structured tasks like files or lists.

Why

A long file might stop halfway if the tool fails, but still say it finished. You could end up with a broken file without any warning. Splitting into parts helps you spot and avoid silent failures.

What not to do

Don't start processing a huge file without checking its size or breaking it up. You might miss incomplete results if the tool fails silently.

Say it like this“Check report_2022.pdf is 4,000 pages. Split into four 1,000-page parts.”
Not like this“Process report_2022.pdf in one go without checking.”
B68 · What to do

Set the tool's output format using its settings, not by describing the format in your message.

Why

Choosing a structured output in the tool’s settings works reliably every time. Asking with words alone can lead to mistakes, as the tool may not follow your instructions exactly.

What not to do

Don't rely on telling the tool to format data in your message. The tool might ignore your instructions, leading to inconsistent or messy results.

Say it like this“Use the tool’s settings to export answers from tasks_list.csv as a table.”
Not like this“Say 'Please give me a table' in your message.”
B69 · What to do

Send big lists or many items directly to the tool, instead of passing them through your own system or software.

Why

Letting your own system handle each item first can slow everything down. Every item waits in line, making the whole process take much longer than if you went straight to the tool.

What not to do

Don't route every item through your own software before sending it to the tool, as this creates a bottleneck and wastes time.

Say it like this“Upload 300 images directly to the tool for processing, not through Outlook.”
Not like this“Send 300 images one by one through Outlook to the tool.”
B70 · What to do

Check the number the tool gives you for things like word count, tasks, or items, and use that number in your report or summary.

Why

People often ignore the count the tool already shows and guess a number instead, usually because they are in a hurry or think their guess is close enough. The tool's number is available without extra work and is almost always more accurate.

What not to do

Don't estimate the number of lines in 'budget_2023.csv' by eye, because your guess will probably be wrong and the tool already shows the exact count.

Say it like this“Show me the number of entries in 'budget_2023.csv' from the tool.”
Not like this“I think there are about 200 lines in the file.”
B71 · What to do

Update your cost estimate every time a price changes or a step takes longer or shorter than planned. Use the latest information, not the original plan.

Why

Costs can shift when suppliers change prices or tasks take more or less time. If you keep using the first estimate, your numbers will quietly become wrong, and nobody will notice until it is too late.

What not to do

Don't keep using the original cost for 'May marketing plan' if the ad rate changes, because your total will be out of date and decisions based on it could be wrong.

Say it like this“Update the cost for 'May marketing plan' after the new ad rate.”
Not like this“Stick with the old cost even after the ad rate changed.”
B72 · What to do

Spend time to ask your question clearly and completely, so you get a good answer the first time instead of fixing mistakes in several drafts.

Why

Fixing and redoing three times takes twice as much work and delays your result. A clear, well-prepared question gets you the right answer in one go, saving effort and time.

What not to do

Don't send three quick drafts of the 'Q2 sales summary' task, hoping to fix errors after each one, because you will waste time and energy on corrections.

Say it like this“Write a clear prompt for 'Q2 sales summary' before sending it.”
Not like this“Send the first thing that comes to mind, then fix it later.”
B73 · What to do

When you need to check if results are right, ask for one item at a time instead of a list, so the tool can check each answer properly.

Why

When asked for many answers at once, the tool skips some of its own checks to finish quickly. Asking for one at a time lets it check each answer, making mistakes much less likely.

What not to do

Don't ask the tool to summarise 10 reports in one go for 'weekly review', because it will likely miss errors it would catch if you asked for each report separately.

Say it like this“Summarise 'sales_report_07.pdf' for me, then stop.”
Not like this“Summarise all 10 reports at once.”
B74 · What to do

Store your secret key, like an API password, in only one place. If the tool cannot find the key, stop and fix the problem before going on.

Why

Two copies of a key can mean one gets out of date or is wrong. The tool might keep running with an old or unknown key, causing silent failures that are hard to find later.

What not to do

Don't keep a copy of the 'api_secret.txt' key on your laptop and another in email, because you might use the wrong one and not notice until something fails.

Say it like this“Check for 'api_secret.txt' in the main folder and stop if missing.”
Not like this“Try using any key you find, even if you are not sure it's current.”

Try these — doing the same job many times over

No score, and nothing to finish. Every wrong answer opens with why the wrong idea was tempting.

Question 1 of 12
 

C. Checking that actually checks

Checking is the step that gets dropped first, because the work already looks finished.

The summary reads well. Nobody notices the source document had a section the summary never mentions, and the section was the one that mattered.

Twelve habits that put a number where a feeling was. Count what came out. Open the source. Ask what was left out rather than whether it is right.

C75 · What to do

Count the number of items in your source file, then check that the output shows the same number of items.

Why

A missing or extra item in the output often means the process skipped something or duplicated a section. Empty spots or missing details can be marked as a success unless you check the count matches.

What not to do

Don't assume the output is correct because the process finished without errors. Skipped items or empty sections can pass through unnoticed if you do not check the count.

Say it like this“The products.csv file has 12 rows; the web page should show 12 products.”
Not like this“Check if the page loads without errors.”
C76 · What to do

Check the size of your output file and compare it to the input before trusting the result.

Why

A much smaller file size in the output can mean missing content, broken images, or incomplete data. File size changes often happen without warnings, so you need to check before you accept the result.

What not to do

Don't skip checking file sizes because the file opens or looks fine. Files can be incomplete or missing large sections even if they load.

Say it like this“Report-final.pdf should be at least as large as report-source.docx.”
Not like this“Download the PDF and send it to Ravi.”
C77 · What to do

Check the final output file itself, not the script or tool that created it.

Why

Scripts or tools can change, but the output file might not update. Sometimes, people review the script and assume the output is fresh, but the file could be old or still generating.

What not to do

Don't review the script or process and assume the output is up to date. Output files can be leftover from an earlier run or incomplete.

Say it like this“Open summary-2024.xlsx and check the figures before sending.”
Not like this“Review the script and mark the task as done.”
C78 · What to do

Compare the date and time on the source files and the output to see if the build actually ran.

Why

If the source file is newer than the output, the process did not run or failed. Checking timestamps prevents sending out old work and saves time spent on debugging missing updates.

What not to do

Don't skip date checks because the output looks right. Files can look fine but still be based on old data if the build did not run.

Say it like this“Check that updated-logo.png is newer than assets.zip after the build.”
Not like this“Assume the build ran because the output file is there.”
C79 · What to do

Check the actual display page for errors, even if data and API checks pass.

Why

Data can be perfect in the backend, but the display layer might show old names, empty sections, or missing images. No warning appears if you only check the data, so you must look at the page.

What not to do

Don't trust API or data checks alone. Front-end issues can hide behind perfect data if you do not view the display.

Say it like this“Open the staff directory page and look for missing names after updating staff.json.”
Not like this“Check the API response and call it done.”
C80 · What to do

Open a live web page to check when features load data while running, not from local files.

Why

Some features fetch data while running. If you only check files in the folder, the feature might use default values and pass static checks, hiding problems that only appear on the live page.

What not to do

Don't test data loading by opening the file directly from the folder. Live features might not fetch real data and you will miss errors.

Say it like this“Visit http://intranet/leave-calendar and check if new requests appear live.”
Not like this“Open leave-calendar.html from the downloads folder.”
C81 · What to do

Add the needed data or files when you set up the tool, instead of waiting until you are using it live.

Why

When the tool has all the information it needs from the start, it does not need to fetch anything over the internet while you work. Missing network calls during use can hide errors, making problems hard to spot.

What not to do

Don't wait to upload 'report-template.csv' until you are already filling it in. Doing so can cause the tool to fail quietly if the network is slow or down.

Say it like this“Upload 'team_list.xlsx' during setup, not when editing records.”
Not like this“I'll add the data later when I need it.”
C82 · What to do

Run your test using a real file from your team, not data you made up yourself.

Why

Self-made examples usually fit what you expect, hiding mistakes. Real team files often have missing fields, extra columns, or different formats, and differences like these break the tool in ways made-up data never does.

What not to do

Don't check the upload tool with a fake 'sales_data.csv' you typed yourself. Real files from Priya or Ben usually have quirks that your test will miss.

Say it like this“Test with last quarter's 'client_feedback.xlsx' from the shared folder.”
Not like this“I made a sample file to see if it works.”
C83 · What to do

Check the results with your own eyes, not only with an automatic tool or tick-box.

Why

Automatic checks can pass files or images that look wrong to a person. A file might meet technical rules but still be confusing, messy, or unreadable in real work.

What not to do

Don't rely on the tool saying 'all clear' for the new icon set. Real people like Meera or James might still find them ugly or unclear.

Say it like this“Open 'logo_set_v3.png' and check how it looks in the header.”
Not like this“The test says all icons passed.”
C84 · What to do

Look at your content in the exact place your colleagues will first see it, not only in the review or edit screen.

Why

Hints or messages can feel helpful in one view but harsh or confusing in another. Where and when people see them changes how they react, especially after a mistake.

What not to do

Don't only check the hint in the admin panel. When someone like Sandeep gets it after a wrong guess on the main page, it might feel rude.

Say it like this“Preview the error message on the login page after a failed attempt.”
Not like this“I checked the hint next to the answer in review.”
C85 · What to do

Ask for a browser check only when you want to be sure the website and its server match up, and expect the check to be real.

Why

A real browser check loads the page as a person would. Only this shows if the website and its server are sharing the same data. Skipping this leaves hidden mistakes that only appear later.

What not to do

Don't say you checked in the browser if you only refreshed the dashboard. A mismatch in 'inventory.html' might hide until a real browser check is done.

Say it like this“Open 'order_summary.html' in Chrome and check the totals match the server.”
Not like this“I refreshed the page and it looked fine.”
C86 · What to do

Write your report with five parts: what you saw, what caused it, how you fixed it, what you checked, and what you did not check.

Why

A clear report with all five points stops guessing or fake fixes. Each part shows what happened, why, how you fixed it, and what remains untested, so others can trust or check your work.

What not to do

Don't write a vague note like 'fixed the upload bug'. Without symptoms, cause, fix, checks, and gaps, nobody knows if the problem is really gone.

Say it like this“Symptom: upload failed for 'staff_list.csv'. Cause: wrong column name. Fix: updated header. Checked: upload works now. Not checked: export function.”
Not like this“Bug fixed, upload works now.”

Try these — checking that actually checks

No score, and nothing to finish. Every wrong answer opens with why the wrong idea was tempting.

Question 1 of 12
 

D. Traps that come back in every project

Some failures are not events at all. Certain patterns arrive in every project, wearing a different costume each time.

A conversation drifts. A reviewer's note gets relayed without the context that made it make sense. An instruction set in message two stops being honoured by message forty.

Eighteen habits, and they are the ones that repay attention most, because each one is a thing you will meet again next week and the week after.

D87 · What to do

Ask the tool to show the exact error message and the original input, so you can see what went wrong.

Why

A spelling mistake in something like a file name or field often leads to a vague error message that does not help you find the real problem. The actual mistake hides behind a general error, wasting time.

What not to do

Don't rely on the tool's default error message alone, because it might not tell you where the mistake started.

Say it like this“Show the full error and the input for report_23may.csv.”
Not like this“Why did it fail?”
D88 · What to do

Check if the helper tool can see the information from the main process before you trust any error it gives.

Why

Helper tools often run in a separate place and do not have access to the details from the main job. Errors that show up may look real, but they come from missing information, not a real problem.

What not to do

Don't assume the helper tool can see everything from the main process, or you might chase errors that do not exist.

Say it like this“Confirm if helper_script.py can access invoice_2024.json from the main task.”
Not like this“Why can't the tool read the file?”
D89 · What to do

Check the exact value the client sent before you make rules or filters about it.

Why

Clients sometimes add extra spaces, change case, or include extra characters in the value. If you only test with your own sample, you miss what actually arrives and your rule fails in practice.

What not to do

Don't make a rule based only on your own test value, or you might miss changes the client makes.

Say it like this“Show me the actual value sent by client for order_id in May_Orders.xlsx.”
Not like this“Assume the value is always 'Order123'.”
D90 · What to do

Open the file and look at how it starts before you write new data to it.

Why

Some files begin with a bracket or a specific marker, while others start with a record. Writing without checking can break the format, making the file unreadable or corrupt.

What not to do

Don't guess the file's layout or write new data without checking, or you might damage the file.

Say it like this“Check the start of summary_2024.json before adding new entries.”
Not like this“Add records to the file without opening it.”
D91 · What to do

Check the key that skips duplicates before you change it, using a copy and counting how many records remain.

Why

Changing the duplicate-skipping key without testing can hide too many entries or miss duplicates. Counting the results on a copy shows if your change works as expected or causes data loss.

What not to do

Don't change the duplicate key in the real data first, or you might accidentally remove important records.

Say it like this“Try the new duplicate key on a copy of sales_Q2.csv and count the results.”
Not like this“Change the key and expect it works.”
D92 · What to do

Stop after two failed fixes and review the problem from the beginning before trying again.

Why

If two different attempts do not solve the problem, the original guess about the cause is likely wrong. Pushing ahead with more fixes adds confusion and makes the issue harder to solve.

What not to do

Don't keep trying new fixes after two failures, or you'll make the situation more confusing and harder to untangle.

Say it like this“Pause after two failed attempts to fix export_31May.csv and review the steps.”
Not like this“Try a third quick fix without checking.”
D93 · What to do

When you have to fix the same problem again, go back to the start of the file or process and read through to find what you missed before making another change.

Why

Repeating a fix means something deeper was not spotted the first time. Going back to the start lets you see the whole picture, so you do not waste time patching what was never fully fixed.

What not to do

Don't keep applying the same quick solution to the same error. Repeating patches hides the real problem and means you spend more time fixing the same thing.

Say it like this“I’ve fixed the date format in invoices.csv twice; I’ll reread the import steps from the top.”
Not like this“I’ll patch the date field again and expect it sticks.”
D94 · What to do

After you fix one example of a bug, make a list of every other place where the same bug could show up, so you don’t miss any hidden problems.

Why

Fixing only the obvious problem leaves others hidden, which can cause errors later. Listing every spot where the bug might appear helps you track and remember what still needs checking.

What not to do

Don't fix only the bug you see and move on. Ignoring other spots lets more problems slip through, which can break things later when nobody remembers where to look.

Say it like this“I fixed the formula in Q2_budget.xlsx—now I’ll note every other sheet using that formula.”
Not like this“I fixed it in one file, so that should be enough.”
D95 · What to do

When you leave a bug unfixed, write a note above the problem in the code itself explaining why, so the reason is clear for anyone who reads it later.

Why

A note in the code stays visible through changes and helps everyone remember why something was left as it is. Conversations or chats about it often disappear or get forgotten.

What not to do

Don't rely on memory or old emails to explain why you left a bug. Future readers will not know the reason and may waste time or repeat mistakes.

Say it like this“# Leaving this error in report.py because Rajiv’s script depends on it for now.”
Not like this“I’ll tell Rina about the bug in a call and leave it.”
D96 · What to do

When you rename something, search for any names that are put together from smaller parts while the program is running, not only the exact old name.

Why

Searching only for the exact name misses names that are built from pieces, so those parts can break quietly and cause errors that are hard to spot later.

What not to do

Don't search only for the old name and change it everywhere you find it. Names made from parts will not match, and can leave broken links or errors.

Say it like this“I’m renaming ‘project_alpha’ in code and in any place it’s built from ‘project’ + ‘_alpha’.”
Not like this“I changed ‘project_alpha’ everywhere, so I’m done.”
D97 · What to do

Before removing a safety note or warning in the code or process, read it carefully to understand what problem it was put in to prevent.

Why

Every safety step or lock was added after a real mistake or loss. Removing one without understanding can quietly bring back old problems that cost time or money.

What not to do

Don't take out safety notes or warnings because they seem unnecessary. Skipping the reason can let old errors come back without anyone noticing until it’s too late.

Say it like this“I’ll read the warning in payroll.py before deleting it, in case it’s tied to an old issue.”
Not like this“I’ll remove this old warning; nothing bad has happened recently.”
D98 · What to do

When something breaks, fix only the part that is actually broken instead of changing the whole system, so you do not hide the real cause or create new problems.

Why

One mistake might be behind several safety steps. Changing too much hides which part mattered, and can remove protections that were stopping old errors from coming back.

What not to do

Don't remove or rewrite everything when one thing breaks. Over-fixing covers up what really went wrong and risks undoing useful protections.

Say it like this“I fixed line 45 in summary.js and left the rest of the checks in place.”
Not like this“I deleted all three checks since one was broken.”
D99 · What to do

Write down why you made a change in the project file, not only what you did or when.

Why

Future readers need to know the reason behind a change, not the date or label. Without an explanation, nobody can tell if the change was a quick fix, a workaround, or a permanent solution. Mistakes repeat.

What not to do

Don't add a note like 'Fixed on 5 March' or 'Updated by Priya' with no reason. People will not know what problem was solved or why the change matters.

Say it like this“Added check for missing ID in orders.csv to stop duplicate rows.”
Not like this“Updated 2024-03-05 by Priya.”
D100 · What to do

Record every design or approach you tried and abandoned, including what did not work and why.

Why

Colleagues will try the same failed ideas again if nobody records what did not work. Teams waste hours repeating the same steps and hitting the same problems in the same order.

What not to do

Don't skip writing down failed attempts on the design for the report generator. Others will waste time trying the same layout and getting stuck on the same formatting bug.

Say it like this“Tried table layout for summary.docx, but headings overlapped with page numbers.”
Not like this“Used new layout for summary.docx.”
D101 · What to do

Split large files before opening them in your editor, especially files over 100 MB.

Why

Editors like Notepad or VS Code freeze without warning when opening big files. The program looks like it is waiting for your input, but it is actually stuck and will not recover until you force it to close.

What not to do

Don't open 2023_salesdata.csv directly in your editor when the file is huge. You will lose time waiting for it to load, and sometimes the editor crashes.

Say it like this“Split 2023_salesdata.csv into 10 parts using splitfile.exe first.”
Not like this“Open 2023_salesdata.csv in Notepad.”
D102 · What to do

Use a step-by-step program to break, tag, match, or reformat files, instead of asking the AI tool directly.

Why

A program follows clear steps and gives the same result every time. The AI tool might guess or miss details, leading to mistakes or inconsistent formatting, especially with large or complex files.

What not to do

Don't ask the AI to 'clean up' log_17.txt in one go. The AI may skip lines or change the format in ways you do not expect.

Say it like this“Run clean_logs.py on log_17.txt before uploading.”
Not like this“Ask AI to fix log_17.txt formatting.”
D103 · What to do

Remove or replace any non-ASCII characters, like arrows or curly quotes, from log files on Windows systems.

Why

Windows programs often fail when they see non-standard characters in logs. One strange symbol, like an arrow or smart quote, can cause every later step to break, and the error is hard to spot until everything fails.

What not to do

Don't leave arrows or curly quotes in build_log.txt. The next process may crash without a clear error, and you will spend hours searching for the cause.

Say it like this“Replace all curly quotes in build_log.txt with straight quotes before saving.”
Not like this“Leave special characters in build_log.txt.”
D104 · What to do

Treat a hash (#) inside quotation marks as part of the text, not as a comment.

Why

A hash inside quotes is sent through as regular text. If you treat it as a comment, you risk leaking private notes or internal messages to customers or partners, because the text is not removed as expected.

What not to do

Don't assume that 'Error: "#urgent"' in messages.log is hidden. The phrase will go out to everyone who reads the log.

Say it like this“Keep '#urgent' inside quotes in messages.log; do not remove or comment it.”
Not like this“Strip all text after # in messages.log.”

Try these — traps that come back in every project

No score, and nothing to finish. Every wrong answer opens with why the wrong idea was tempting.

Question 1 of 12
 

E. Jobs running where you cannot see them

Work you cannot see is work you cannot stop, and it is where the largest bills come from.

A job was started in a window that has since closed. Nothing reports on it. The first sign is a figure at the end of the month, or a file that changed while you were somewhere else.

Eight habits that keep long-running work visible, interruptible, and writing down what it has already achieved.

E105 · What to do

Check the running process list, not only the open ports, before deciding a server has stopped.

Why

A process can keep running even after its network port is closed. Memory and old connections may still be active, which can interfere with later work if not properly cleared.

What not to do

Don't rely on seeing a free port to assume everything stopped. Old processes might still be using memory and holding connections, causing confusion when you restart.

Say it like this“Ps aux | grep my_server.py before starting my_server.py again.”
Not like this“I don’t see port 8000, so the server is gone.”
E106 · What to do

Stop a process by finding and killing the one using the port number, not by searching for the program name.

Why

Killing by program name can hit unrelated processes, disrupting other work. Some old processes might survive if their names differ slightly, leading to outdated code running unnoticed.

What not to do

Don't kill every process with the same name, since doing so can break other people’s work and leave some old copies running, which confuses everyone.

Say it like this“Fuser -k 3000/tcp to stop the server on port 3000.”
Not like this“Killall node for every Node.js process.”
E107 · What to do

After restarting a server, confirm that exactly one instance is running before you continue with your work.

Why

Zero instances means nothing is working; more than one means requests go to the wrong place or get lost. A quick check prevents hours of confusion and hidden bugs.

What not to do

Don't skip checking for duplicate or missing processes, because missing this step leads to silent failures where requests disappear or the wrong code answers.

Say it like this“Ps aux | grep server.js and see only one line for server.js.”
Not like this“Restarted, so it must be fine now.”
E108 · What to do

When your server is slow but shows no errors, restart it to clear any old, stuck processes.

Why

A leftover or zombie process can keep old connections waiting behind the live server. Requests pile up but logs show nothing wrong. Restarting clears the queue and restores speed.

What not to do

Don't assume no errors mean no problem. Slowness with no errors often hides a zombie process clogging the system.

Say it like this“Sudo systemctl restart flask-app when flask-app is slow but logs are clean.”
Not like this“No errors in logs, so the server is healthy.”
E109 · What to do

Start a server using a separate launcher, not from inside the model’s shell, if you want it to stay up after closing the shell.

Why

A server started from the model’s shell will shut down when the shell closes. The port disappears, so it looks like a crash, but the process was tied to the shell.

What not to do

Closing the window after launching the server in the same shell as your model shuts down the server without warning. It's easy to forget the server is tied to that shell, but keeping them separate avoids surprise shutdowns.

Say it like this“Nohup python3 serve.py & to keep serve.py running after closing the shell.”
Not like this“Python3 serve.py in the same shell, then close it.”
E110 · What to do

Use a single, reliable command or script to start your server, not a mix of different start commands.

Why

Multiple start commands can compete for the same port. One team had four outages in one day because a test script replaced the real server, confusing everyone.

What not to do

Using different commands or scripts to start the same project can cause port clashes, leaving a fake server running instead of the real one. Each command may launch its own process, so stick to one method.

Say it like this“Always use ./start_project.sh to launch dashboard.py for the team.”
Not like this“Sometimes run python dashboard.py, sometimes use npm start.”
E111 · What to do

Set up your dev server so one slow call to /api/report.pdf does not block other requests. Run heavy tasks in the background, or use a separate worker process.

Why

A single-threaded dev server handles only one thing at a time. When Priya downloads a large report, every other request waits. Her browser tab spins, and the page looks frozen. Nothing else loads until that slow job finishes.

What not to do

Don't run every task on the main thread, or let big downloads like /api/report.pdf freeze the whole page. The server cannot respond to anything else while stuck on one job.

Say it like this“Move /api/report.pdf generation to a worker so my dashboard loads fast.”
Not like this“Why is my dashboard frozen when I download a report?”
E112 · What to do

Count updates on your side instead of letting an open page keep polling the server after every change. Stop the page from asking again if nobody is watching.

Why

A page that keeps polling the server after each update uses more power and network. When Alex leaves the tab open overnight, the server keeps sending updates nobody reads. The bill grows, and the laptop fan never stops.

What not to do

Don't let your page keep polling for updates when nobody is looking. Ignoring this wastes electricity and money, and fills logs for no reason.

Say it like this“Pause update polling on sales-stats.html when the tab is hidden.”
Not like this“Keep polling updates even if nobody is watching.”

Try these — jobs running where you cannot see them

No score, and nothing to finish. Every wrong answer opens with why the wrong idea was tempting.

Question 1 of 12
 

F. Finding a working rhythm

Rhythm is not a productivity idea here. Rhythm is what stops a day of AI work from becoming a day of waiting and re-reading.

You ask, you wait, you read the whole answer, you find one thing wrong, you ask again. Four rounds later the afternoon is gone and the draft is worse than the second version.

Thirteen habits about when to look, when to leave it alone, and when a thing is finished enough.

F113 · What to do

Approve the full plan for the project at the start, then allow the tool to complete the task without asking for more permissions until the end.

Why

Pausing at every step to get approval makes the work drag out, and colleagues may think you are avoiding responsibility. A single approval at the start keeps work moving and shows confidence in the plan.

What not to do

Stopping every hour to ask if you should continue turns a quick 20-minute task into a half-day wait. You might want reassurance, but Priya expects the report by lunch, so keep going unless told otherwise.

Say it like this“Approve the July_data_cleanup_plan.xlsx once, then let it run to finish.”
Not like this“Should I continue after step 2, or wait for your OK?”
F114 · What to do

Check with your manager or colleague which option to use before starting the task, not after you have begun or partway through.

Why

Deciding between choices like format or template upfront helps avoid confusion. Asking for a decision during the process interrupts the flow and can force a redo of work already done.

What not to do

Starting to clean the spreadsheet and then asking Ravi about the date format halfway breaks up your work and can mean redoing the first half. It's tempting to begin quickly, but check the format first to save time.

Say it like this“Before I start, should I use DD/MM/YYYY or MM/DD/YYYY in Q2_sales.csv?”
Not like this“I’ve finished half—should I use DD/MM/YYYY or MM/DD/YYYY?”
F115 · What to do

Interrupt the workflow only if you find a real cost issue, risk of losing information, or a genuine change in the task, not for routine checks.

Why

Stopping work for every small question or check slows progress and makes it hard to see what really matters. Only big issues like lost data or new instructions should break the flow.

What not to do

Pausing every 10 minutes to ask if you should keep sorting the contacts list keeps the job half-done and hides if something really needs attention. You may want feedback, but steady progress shows real issues sooner.

Say it like this“Stop processing contacts_2024.xlsx only if data is missing or instructions change.”
Not like this“Should I keep going, or do you want to check now?”
F116 · What to do

When you answer a question, give a clear response and then stop. Do not keep replying or using the tool further until new information is needed.

Why

Extra replies after a question can mean the answer was not understood or accepted. If you keep working after asking, you might guess wrong and create mistakes.

What not to do

Replying with three follow-up messages after getting the answer about the file name confuses the team and can start a chain of unnecessary edits. You might want to be thorough, but one clear reply is enough.

Say it like this“Rename the file to final_minutes_12June.docx and wait for further instructions.”
Not like this“Should I rename it, or do you want another name, or maybe something else?”
F117 · What to do

When you know the right way to fix a mistake, give the correction directly to the tool, without listing options or making a plan.

Why

Explaining a long fix or giving choices hides the fact that you already know the answer. Direct fixes save time and prevent confusion about what should happen next.

What not to do

Sending three ways to fix the typo in the invoice when you know which is right wastes time and keeps the error longer than needed. It's tempting to show options, but picking the best fix speeds things up.

Say it like this“Correct 'Surbhi' to 'Surabhi' in invoice_1042.pdf.”
Not like this“Should I use Surbhi, Surabhi, or another spelling?”
F118 · What to do

Give only the correct choice if the answer is clear, instead of presenting two options when one is obviously right.

Why

Offering a false alternative when the answer is clear wastes time and can bring back a problem you already solved, confusing both you and your colleague.

What not to do

Asking if you should keep the mistake in the time sheet or remove it, when you know it needs correcting, could make the error return. You may want confirmation, but fixing it straight away prevents confusion.

Say it like this“Remove the extra row in timesheet_May23.xlsx.”
Not like this“Should I remove the extra row, or leave it?”
F119 · What to do

Ask the tool to put items in order of importance, not as a list where all items are equal.

Why

When all points are marked as equally important, you end up deciding what matters most. Sorting should be done by the tool, so you only need to review or act, not decide the order.

What not to do

Don't ask for a list without telling the tool to rank the items. Leaving everything as equal means you must figure out what comes first, which takes more time.

Say it like this“Rank the tasks for the Q3_Report.docx review from most urgent to least.”
Not like this“List all tasks for Q3_Report.docx review.”
F120 · What to do

Show the quickest way to reach your goal first, then give other options clearly after, not mixed together.

Why

When multiple plans or steps are mixed together, it gets confusing to tell which one to follow. Starting with the shortest clear solution helps you get started, and alternatives are easier to compare.

What not to do

Don't ask for all possible ways at once or let the tool mix steps from different plans. You will waste time sorting out which steps go together.

Say it like this“Show the fastest way to export Budget2024.xlsx to PDF, then list other methods.”
Not like this“How do I export Budget2024.xlsx to PDF?”
F121 · What to do

Use everyday words for instructions and ask the tool to define any work-specific terms clearly the first time.

Why

Reports and tools often use workplace terms that sound familiar but hide what you need to do. Clear, simple language means you know exactly what action is needed, without guessing.

What not to do

Don't let the tool fill answers with jargon or skip explaining terms. You could miss what to do, or waste time looking up meanings.

Say it like this“Explain each step to update the 'Sales_Leads.csv' file using plain words.”
Not like this“Guide me through the workflow for Sales_Leads.csv.”
F122 · What to do

Ask the tool to answer only the question you asked, and stop there. Do not give extra steps or checks.

Why

Extra advice or steps that you did not ask for make your task bigger and less clear. Unwanted details can add confusion and extra work you did not plan for.

What not to do

Don't let the tool add tips or related tasks you did not request. You might end up with more to do, or miss your real goal.

Say it like this“Show how to change the font in ClientLetter.docx, only that.”
Not like this“How do I change the font in ClientLetter.docx? Any other tips?”
F123 · What to do

When fixing feedback with more than one part, answer every part fully, not the first one.

Why

If you only answer the first part of feedback, it looks like you did not read the whole message. Missing points means extra back-and-forth, and more time spent.

What not to do

Don't respond to only the first question or request in feedback. You will leave the rest undone and create more follow-up work.

Say it like this“Update the table in Q2_Summary.docx and fix the date in the heading.”
Not like this“Update the table in Q2_Summary.docx.”
F124 · What to do

Always get both the file's saved location and its live web address when a file is finished.

Why

A file's saved folder tells you where to find it later, but the web address is what you share or use to show the file working. Missing either one means you will have to ask again.

What not to do

Accepting only a file name, folder path, or web link might seem enough, but you risk losing track or having to chase up missing details if you do not get both.

Say it like this“CompletedBudget2024.xlsx: folder path and web link.”
Not like this“Send me the link for CompletedBudget2024.xlsx.”
F125 · What to do

Ask the AI tool to show the full folder path for every file it changes, right after finishing the task.

Why

When you work across many folders, the AI might only mention file names or partial locations. Without the full folder path, you waste time looking for the changed files, especially if there are several with similar names.

What not to do

Accepting a summary that lists only file names or leaves out folder locations leads to confusion and extra searching. When your project has many folders with files like 'draft.docx', location details save time.

Say it like this“List the full path for every file you changed, like /marketing/2024/June/summary.docx.”
Not like this“Show me what files you changed.”

Try these — finding a working rhythm

No score, and nothing to finish. Every wrong answer opens with why the wrong idea was tempting.

Question 1 of 12
 

G. When a session goes wrong

Every session goes wrong eventually. The difference is whether you can say what went wrong.

The answer stopped being useful somewhere around the twentieth message and you cannot point at where. So you start again from nothing, and lose the four decisions that were right.

Eight habits for the recovery: what to carry across, what to leave behind, and how to tell a bad session from a bad task.

G126 · What to do

Compare your current version of the file to the last one that worked before starting to fix a bug.

Why

Guessing or searching for the problem wastes time and often misses the real issue. A side-by-side comparison shows exactly what changed, so you can spot the cause of the break quickly.

What not to do

Don't start changing things without checking your last working copy. Random fixes often make things worse and hide the main problem.

Say it like this“Show me what changed in InvoiceReport_v3.xlsx since last Friday.”
Not like this“Why is it broken now?”
G127 · What to do

Check what has changed in your file before restoring an old backup, so you do not lose your latest work.

Why

A backup might be out of date, missing recent edits or additions. Restoring it without checking risks losing hours of work, or overwriting fixes you made since the backup.

What not to do

Don't restore a backup right away when something goes wrong. You might erase new work or miss the actual problem.

Say it like this“Show changes in Q2Targets.docx since yesterday before restoring backup.”
Not like this“Restore the backup now.”
G128 · What to do

Pause and review what happened after a mistake, before you try to fix it.

Why

Rushing after an error often leads to more mistakes, or missing the real cause. Taking a point helps you understand what really went wrong and decide how to fix it properly.

What not to do

Don't hurry to fix mistakes as soon as you see them. Acting too fast usually hides the root cause and wastes time.

Say it like this“Stop and show me what changed in Budget2024.xlsx before I fix it.”
Not like this“Fix it now, it's urgent.”
G129 · What to do

Set up your task so that if it fails, it stops and tells you why, instead of trying again without warning.

Why

When a task keeps repeating itself without an error message, you waste time waiting and guessing. A clear stop and message lets you talk to a colleague or fix the problem right away.

What not to do

Don't let a failed task keep retrying quietly. You lose time and may not notice the error until much later.

Say it like this“Stop processing Orders_June.csv and show the reason if it fails.”
Not like this“Try again if it doesn't work.”
G130 · What to do

Fix a mistake and make sure you do not repeat it in the same way again.

Why

People expect you to learn from errors. If you repeat a mistake after fixing it once, colleagues lose trust and may think you do not care or pay attention.

What not to do

Don't repeat the same mistake after fixing it once. Others notice and start to doubt your reliability.

Say it like this“Check I do not send the wrong attachment again for HR_Policy2024.pdf.”
Not like this“Oops, I sent the wrong file again.”
G131 · What to do

Adjust how you run the tool or process if the results are off, instead of changing the whole method.

Why

Bad results often come from using the tool incorrectly or with the wrong settings. Changing everything wastes effort when a small tweak in how you start or use the process would fix it.

What not to do

Don't throw out the whole method when results are wrong. The problem is often in how you run it, not the method itself.

Say it like this“Try running the sales report on Q2_Summary.xlsx with different filters first.”
Not like this“Rewrite the whole report process.”
G132 · What to do

Show the results of any change to agreed limits or settings—like filters—before saving them, so everyone can see what will happen.

Why

Changing a setting, such as a filter for an email folder or a project deadline, can have unexpected effects. If nobody sees the result first, important messages or tasks might get lost, delayed, or wrongly sorted.

What not to do

Don't update the spam filter rules for the shared mailbox without showing Priya and Mark what emails will be caught, or you might hide their client messages by accident.

Say it like this“Show me what happens if I change the filter in 'Q2-invoices.xlsx' first.”
Not like this“Update the filter settings for the invoices file now.”
G133 · What to do

Write down what is happening, who is involved, and what has been tried before the meeting gets busy or the conversation moves on.

Why

When the team starts talking, details can be skipped or forgotten. If you do not record the situation early, the reasons behind decisions and the steps taken can disappear from memory and the notes.

What not to do

Don't wait until after the meeting to write up what happened with the 'April-leave-requests.docx' file, because you will lose why certain requests were approved or denied.

Say it like this“Before we talk, let me write down what happened with 'April-leave-requests.docx'.”
Not like this“Let's discuss first, I’ll write it up later.”

Try these — when a session goes wrong

No score, and nothing to finish. Every wrong answer opens with why the wrong idea was tempting.

Question 1 of 12
 

The questions people still ask

Why does my AI tool give different answers to the same question two minutes apart?
Repeat answers

AI tools often use random sampling when generating answers. Asking the same question twice, even in the same conversation, can produce different responses. Settings like temperature or randomness affect this. If you need the same answer every time, look for a way to set the tool to a more predictable mode.

How do I know if my AI tool is actually using the latest version of my spreadsheet?
File confusion
  1. Check the tool’s file list to see which version it loaded.
  2. Look at the timestamp or file size for a match with your latest copy.
  3. If the version is old, upload or sync the new file before running your task.
  4. Ask the tool to confirm the file name and date before you start work.
Why does my AI tool keep losing track of the conversation when I ask about something from earlier?
Memory limits

AI tools have a limited memory for each session. After a certain number of messages, the tool forgets earlier parts of the conversation. If you need it to remember something, repeat key details or attach the relevant file again when you ask your next question.

What should I do if my AI tool says it cannot see a file that I know is there?
Missing files
  1. Check the folder or path you told the tool to use.
  2. Make sure the file name matches exactly, including capitals and spaces.
  3. If the file is open in another program, close it and try again.
  4. Refresh or resync the tool’s file list, then check again.
How can I stop my AI tool from using the wrong version of a report when I have more than one file with a similar name?
Version clash
  • Rename older files with the date or version in the name.
  • Store finished reports in a separate folder from drafts.
  • Delete or archive outdated versions after sending the final copy to your manager.
  • Double-check the file path before giving it to the tool.
Why does my AI tool sometimes miss emails from Priya when searching my inbox?
Search misses

Email search depends on how the tool matches names and addresses. If Priya uses more than one email address, or if her name is spelled differently in some messages, the tool might skip those. Check the search settings and try using her full email address instead of her name.

How do I know if my AI tool is running tasks in the background while I work on something else?
Background jobs

Most AI tools show a progress bar or status message for tasks running in the background. Some send a notification when the job finishes. If you do not see any sign, check the task manager or activity log for ongoing work.

What is the difference between giving my AI tool a folder to search and giving it a whole drive?
Search scope
ScopeFolderWhole Drive
SpeedFasterSlower
Risk of wrong filesLowHigh
PrivacyMore controlLess control
Setup neededPick one folderPick the whole drive
Why does my AI tool keep asking for my login details every time I open it?
Login trouble

Some tools do not store your login details for security reasons. If you clear your browser cookies or use private mode, you will need to enter them again. Check if the tool offers a secure way to save your credentials, or use a password manager.

How can I tell if my AI tool really read the document I uploaded, or if it only scanned the first page?
Partial reading
  • Ask the tool to summarise a section from the end of the document.
  • Request a list of all headings or key points.
  • Check the document size the tool reports after uploading.
  • If answers seem incomplete, try splitting the file and uploading each part separately.
Why does my AI tool keep giving me answers about last week’s schedule when I ask about next week’s plan?
Outdated info

AI tools often use the most recent data they have seen. If the calendar file or planner is not updated, the tool may give answers based on old information. Check that the AI has access to the latest version of your schedule before asking about future plans.

How can I check if my AI tool is using the right spreadsheet when I have ‘Sales_Q1.xlsx’ and ‘Sales_Q1_Final.xlsx’ in the same folder?
File confusion
  1. Open the folder and note the exact file names and sizes.
  2. Ask the AI tool to list the files it sees in that folder.
  3. Check the file path the tool is using in its answer.
  4. Confirm the file name matches the one you want to use.
Why does my AI tool keep missing calendar invites from my Outlook when I ask it to find meetings with Ravi?
Missing data

AI tools sometimes cannot see every calendar or mailbox by default. Check if the tool is connected to the right Outlook account, and if it has permission to read shared calendars. Some tools only scan the main calendar, not all folders.

How do I stop my AI tool from running the same task twice when I click the button again by mistake?
Duplicate task
  • Wait for the task to finish before clicking again.
  • Check if the tool shows a progress bar or status update.
  • Look for a ‘cancel’ or ‘stop’ button.
  • Refresh the page to clear stuck requests.
Why does my AI tool say ‘done’ but I cannot find the file it was supposed to create in my ‘Reports’ folder?
Missing file

AI tools sometimes save files in a default folder, not the one you expect. Check the tool’s settings for the save location. Search your computer for the file name, or ask the tool to show the full file path where it saved the report.

How can I tell if my AI tool is still working on a task, or if it has frozen?
Stuck process
  • Look for a loading symbol or spinning circle.
  • Check if the tool’s status bar is moving.
  • See if the tool’s log or activity panel is updating.
  • Try opening another part of the tool to see if it responds.
  • Wait a few minutes before restarting the program.
Why does my AI tool keep using the wrong template when I ask it to create a new slide deck for the Monday meeting?
Template mix-up

AI tools often pick the first or default template they find. Check if you have named your preferred template clearly, like ‘Monday_Meeting_Template.pptx’. Tell the tool the exact file name or path to use for new slide decks.

How do I check if my AI tool has finished updating all the files in a folder before I start working on them?
Update check
  1. Ask the AI tool to list all files it has updated, with timestamps.
  2. Compare the list to the files in your folder.
  3. Check if the timestamps match today’s date.
  4. Open one or two files to confirm the changes are there.
What is the difference between ‘reading’ and ‘indexing’ when my AI tool says it has processed my project folder?
Process type
ActionWhat it doesWhat you get
ReadingOpens and checks the content of each fileDetailed answers about the file’s content
IndexingScans file names and basic info onlyFaster search, but less detail from inside files
Why does my AI tool keep mixing up my June and July expense reports when I ask it to summarise my spending?
File naming

AI tools often match file names by keywords, not by month. If both files are named ‘Expenses_Report.xlsx’, the tool may pick the wrong one. Rename files to include the month, like ‘Expenses_June2024.xlsx’ and ‘Expenses_July2024.xlsx’, to avoid confusion.

Why does my AI tool miss attachments when searching emails from my manager in my Outlook inbox?
Missing data

Some AI tools only scan the main text of emails, not the attachments. When searching, mention the file type or attachment name. If the tool cannot access attachments, download them to a folder and point the tool there instead.

How can I check if my AI tool is using the latest version of ‘MarketingPlan.docx’ when I update it after uploading?
Version trouble

Upload the updated ‘MarketingPlan.docx’ again, then ask the tool to list the version or timestamp it sees. If the date matches your last save, the tool has the latest version. If not, refresh or re-upload before working further.

Why does my AI tool keep timing out when I ask it to summarise all the PDFs in my ‘ClientDocs’ folder?
Timeout

Large folders with many PDFs can overload an AI tool, causing timeouts. Split the task into smaller groups, such as five files at a time. Summarise each batch separately, then combine the results in a final summary.

How do I check if my AI tool finished processing all the slides in ‘Q2Review.pptx’ before I send the summary to my team?
Incomplete work

Ask the tool for a list of slide titles or numbers it has processed from ‘Q2Review.pptx’. Compare this list with the slide count in PowerPoint. If any slide is missing, prompt the tool to read the file again before sharing.

What steps should I take if my AI tool keeps skipping over blank rows in my ‘LeadsList.csv’ when cleaning up data?
Data cleaning
  1. Open ‘LeadsList.csv’ and check how blank rows are formatted.
  2. Ask the tool to display its reading of the first 20 rows.
  3. If blank rows are missing, reformat them to contain a placeholder, such as ‘N/A’.
  4. Run the clean-up task again and check if all rows appear.
How can I make sure my AI tool does not overwrite ‘Budget2024.xlsx’ when updating ‘Budget2023.xlsx’ in the same folder?
File safety
  • Rename files clearly before uploading.
  • Use unique names, such as ‘Budget2023_Final.xlsx’.
  • Tell the tool the full file name each time.
  • Check the folder after each update to confirm the right file changed.
Why does my AI tool answer questions about ‘HR_Policies.pdf’ even after I delete the file from my drive?
Cache trouble

Some AI tools keep a memory or cache of files they have read. Deleting a file from your drive does not always remove it from the tool’s working memory. Clear the tool’s cache or restart the session to remove old content.

What should I do if my AI tool keeps confusing emails from ‘Priya Sharma’ and ‘Priya S.’ when searching my inbox?
Contact mix-up
  • Search using full names or email addresses.
  • Filter by sender address, not only display name.
  • Check if both contacts appear in your address book.
  • Ask the tool to show which emails it found for each name.
How do I check if my AI tool finished processing all the rows in ‘LeadsList.csv’ before I start sending follow-up emails?
Processing check
  1. Ask the tool to count the number of rows it processed.
  2. Compare this with the total rows in ‘LeadsList.csv’.
  3. If the numbers do not match, ask the tool to process the missing rows.
  4. Confirm by sampling a few entries from the start, middle, and end.
What is the difference between asking my AI tool to ‘scan’ a file and to ‘read’ a file when working with ‘ClientNotes.txt’?
Feature difference
ActionWhat happensWhen to use
ScanTool checks for keywords or structure, not detailQuick search or file list
ReadTool loads and processes the full contentDetailed analysis or summary

What the rules give back

None of these eighty-eight are about slowing down. Every one of them is about being able to go faster without checking twice, because the check is built into how the work is asked for.

Copyright © Pawan Nayar · LLOS.ai · 2026 — Original pedagogy, voice, and design — all rights reserved.