Actually Works · Failure modes

What breaks, written down.

Every episode ships with its own limits section — the part most tutorials leave out. This is all of them in one place: 243 documented so far, across 80 episodes.

Episode 82 — Some say Claude got dumb on purpose!

Not confirmed, so not said as fact: that Claude Opus 5.5 got worse, that anything was done on purpose, or that it saved money. Anthropic's report covers Sonnet 4.6, Opus 4.6 and Opus 4.7 in Claude Code, not Opus 5.5.
The report's three changes affected Claude Code, the Agent SDK and Cowork. The API was not affected.
The roughly 3% drop in the video is Anthropic's number for one of its own evals after the short-answer rule. It is not a general measure of quality.
Anthropic makes Claude, and this video was made with help from Claude. This channel did not test Claude's quality before or after the changes.
The screens in the video are drawn in our own design, not recordings, and the quality line in the opening is an illustration, not data. Menu looks may differ on your machine.

Episode 81 — ChatGPT is a yes-man. Here's the setup that fixes it.

One question, one try for each chat, and one push. Another day, another account or another model can answer differently. This is not a measurement.
In chat B I did not push a second or third time, so I do not know if it keeps holding. I do not claim the text always works.
The screens in the video are drawn in my own design, not recorded: the settings menu rows and the starting text are illustrations. I did not publish my real recordings because they show account details.
I also pasted a different, older Custom instructions text before test A; it was removed for A. I did not test other texts.

Episode 80 — Does your ChatGPT cave when you push? A 30-second test

I have not shown that this text stops ChatGPT from caving. It is the text I plan to test next, and your own result with your own account is the real answer.
The ChatGPT answers in this video are from my October 2026 recording of two chats about a phone I did not need (reused from episode 79); I did not record a new chat for this episode. The Settings and typing screens are drawn illustrations, not a recording. In that test ChatGPT gave in once and held once, two chats, one try each, so it is not proof of how ChatGPT always behaves.
Research on this topic (for example the SYCON-Bench paper, EMNLP 2025 Findings) suggests prompts help only partly; I read abstracts, not the full papers, so I cite no numbers.
One question, three pushes and one try is a quick check, not a measurement. Run it a few times on different questions before you trust the result.

Episode 79 — I asked ChatGPT for permission to do two dumb things (real test)

Two chats, one try each. Another day, another account or another model can answer differently.
This is a test of how one chatbot answers a push, not financial or career advice.
I did not test other plans, other phrasings or the paid version.

Episode 78 — I told ChatGPT I'm texting my ex at 2am. Smart plan? (real test)

One chat, one try, one day. Another day, another account or another model can answer differently, including agreeing with you.
This is a test of one tone of answer, not advice about relationships.
I did not test other plans, other phrasings or the paid version.

Episode 77 — Your phone hides a map pin inside your photos (iPhone privacy)

I did not run any ChatGPT test for this episode and did not record a screen. The ChatGPT claim is only what OpenAI's help page says about the model; it does not tell what an app or website does with a file before or after it reaches the model.
Apple says people you share with may be able to see the location. Whether they can depends on how you send the photo and which app receives it. I did not test each app, so check before you rely on it.
The menu names follow Apple's current guide and can differ on other iOS versions. Android phones have their own setting, which this page does not cover.
Removing the location does not hide what is in the picture. In my earlier tests, a photo with no location still let ChatGPT guess a city.
Next on the channel: what ChatGPT does when I hide even more of the photo. That test needs a new real recording that does not exist yet, so there is no date.

Episode 76 — I blurred even more of the photo. Could ChatGPT still place it? (privacy test)

One photo, one try, one account, one day. Another photo, another day or another model can give a different answer, a more exact one or a refusal.
In this test ChatGPT still got the city and the neighbourhood from the picture itself. Blurring signs did not hide the general area.
I do not know whether ChatGPT's web search used the photo or only text, and I did not check how its answer was produced.
I could not verify whether ChatGPT uses a photo's metadata. Do not assume it ignores or reads it; remove the location before sharing either way.
The photo is not a picture of anywhere I live, and its location was not published.

Episode 75 — I covered the sign. Could ChatGPT still find the street? (privacy test)

One photo, one try, one account, one day. Another photo, another day or another model can give a different answer, a more exact one or a refusal.
In this test ChatGPT still got the city and the neighbourhood right. Covering a sign did not hide the general area.
I do not know whether ChatGPT's web search looked up the photo itself or only searched text. The app does not say.
Not tested: covering more of the photo (the next test), other AI tools, other photos.

Episode 74 — ChatGPT named my street from one photo (no GPS)

One photo, one try, one account, one day. Another photo, another day or another model can give a different answer, including a wrong one or a refusal.
Turning Location off only removes the GPS location for that share. Other photo information may still be sent, and anything visible in the picture can still give the place away.
Not tested: cropping out the signs, other AI tools, other photos. That test comes next.

Episode 73 — 3 ways to make ChatGPT stop agreeing with you

Not measured by us: nobody here tested how much each try reduces agreement. The sources are guides and help pages, not studies.
The preset names and where they sit in Settings can change between app versions. If the menu looks different, search Personalization in the app.
A critical answer is not automatically a correct answer. Check any claim that matters.

Episode 72 — I told ChatGPT I'm skipping sleep to study all night. Then I told it to just agree with me.

One run, one account, one phone, one day. ChatGPT may answer differently for you, on another day, in another language or with another model.
The test does not show that ChatGPT never agrees with wrong plans. It only shows what it answered to these two messages.
Not checked: other chatbots, other kinds of plans, and whether the answer changes if you push a third or fourth time.

Episode 71 — One switch in ChatGPT decides if your chats get used. Mine was on.

OpenAI's own help page was not reachable (it returned an access error), so these steps are verified against the live app, not against OpenAI's documentation. Menu names and the exact path can differ by account, plan, country and app version.
Whether the switch is on by default for your plan or in your country was not checked. It was on for the recorded account. Check yours on the screen.
Not checked and not said in the video: whether switching it off affects chats that were already used, how long chats are kept, and how this differs on the website or on Android.

Episode 70 — Never tell ChatGPT anything personal until you check this.

OpenAI's own help page was not reachable (it returned an access error), so these steps are verified against the live app, not against OpenAI's documentation. Menu names and the exact path can differ by account, plan, country and app version.
Not checked and not said in the video: how long deleted or switched-off memories are kept, whether switching Enable memory off removes anything already saved, and how this differs on the website or on Android.
ChatGPT learns about you only to the extent Enable memory is on and you have shared things in chats; an empty or short summary is possible.

Episode 69 — Your iPhone rewrites your notifications, and it has gotten them wrong.

Not said in the video because the sources do not settle it: whether news summaries are back and under what warning, how often summaries are wrong today, and whether a given country or language has the feature.
On iOS 27 one press report says there is no single master switch for all of Apple Intelligence any more; the Summarize Notifications switch above is what this guide covers. Summaries inside Messages and Mail are separate settings that were not checked against an Apple page.
This channel did not test this on a live iPhone. Menu names are taken from Apple's help page; your screen may differ slightly depending on your iOS version.

Episode 67 — YouTube ignores the birthday you typed.

Not said in the video because the help page does not say it: which signals the model uses, how accurate it is, how many accounts it flags, and whether users get a notification. YouTube's July 2025 blog mentions signals such as the types of videos searched and watched; that is not repeated here.
Whether Belgium or any single country is covered was not checked beyond the list on the page.
This channel did not test this on a live account. Page names are taken from Google's help pages; your screen may differ slightly.

Episode 66 — Is that Amazon sale real? Check what it cost last month!

Not said in the video, because it was not confirmed on a first-party page: any Prime requirement, a cancellation window or duration for auto-buy, and spending limits. One search summary mentioned them; no Amazon page read here does.
Whether Amazon's price history is accurate is unknown, and this video does not claim it is.
This channel did not test the feature on a live account. Link names are taken from Amazon's pages; your screen may differ slightly.

Episode 65 — Gmail's AI writes your replies. Not for everyone.

Exact plan names and prices were not checked, and the video does not say them. Whether a free personal account has access outside the trusted tester program was not confirmed.
The ChatGPT Gmail connector, Smart Reply and Smart Compose are separate features not covered here.
This channel did not test the feature on a live account. Button names are taken from Google's page; your screen may differ slightly.

Episode 64 — Your Copilot chats can shape your ads.

Not said in the video because it could not be confirmed for the new Copilot app: whether chats are used for training by default, the Training on conversation activity toggle, the 18-month retention, and human review of some chats. Microsoft's pages that describe them say they apply only to the older Copilot app.
Work or school accounts and Microsoft 365 apps for home are treated differently by Microsoft (prompts are not used to train foundation models), so this guide is for a personal Microsoft account.
This channel did not test the switch on a live account. Menu names are taken from Microsoft's help pages; your screen may differ slightly.

Episode 63 — Is the AI chat you use one of the risky ones?

The ranking scores what each company says in its own policies, not what it does. Incogni sells personal-data removal, so it has a commercial interest in the topic. Exact per-platform scores were not read here; only who sits at the two ends is claimed.
Older Incogni studies (9 platforms) and one article placing DeepSeek at the bottom do not match this 2026 ranking and are not used.
This channel did not test the switch on a live account. Menu names are taken from OpenAI's help page; your screen may differ slightly, and the phone app's path is slightly different from the web.

Episode 62 — Delete a ChatGPT chat, and OpenAI can still keep it for up to 30 days.

This channel did not test this on a live account. OpenAI's own page about the lawsuit could not be opened (HTTP 403), so the court details come from press reports, not from OpenAI.
OpenAI does not say, in the pages read, whether Delete all chats also removes archived chats. Deleting a chat does not delete files saved to your Library.
Menu names are taken from OpenAI's help pages and public guides; your screen may differ slightly.

Episode 61 — Someone at Google can read your chats with Gemini, even after you delete them.

This channel did not test this on a live account. Google does not say how many chats are picked — only "a subset".
The steps above are for the Gemini website. Google's activity page links a separate article for Android and iPhone that this channel did not read, so no phone steps are claimed.
The Turn off button labels come from Google's help page; the interface may differ slightly by country or account type (for example a work or school account).

Episode 60 — Anything you type to Instagram's AI chat can come back as an ad.

This channel did not test this on a live account. Nothing newer than the October 2025 announcement was found, so it may have changed since.
Meta does not say whether deleting your chats changes the ads you see. The video says so and does not promise it.
If you are in the EU, the UK or South Korea, reports say this does not apply to you — check your own app's notices.

Episode 59 — Your LinkedIn profile is already teaching an AI — and no one asked you.

This channel did not test the toggle on a live LinkedIn account for this episode — the path above comes from LinkedIn's own help page and independent step-by-step guides, and menu labels move between app versions and languages.
Switching it off does not delete anything already used for training, and LinkedIn's help page does not say how much of any one person's data that is.
The setting controls generative-AI training only. LinkedIn also uses member data for other things (ads, recommendations, job matching) that this switch does not govern.

Episode 58 — That photo in your WhatsApp chat right now — type one sentence, and Meta's AI will redraw it for you, before you even hit send.

This feature's real-world accuracy (how well it follows a specific instruction, how it handles complex edits) was not independently tested by this channel — the claims here are sourced to WhatsApp's own rollout and independent reporting, not a first-hand benchmark.
It's a fast in-chat touch-up, not a replacement for a real photo-editing app — for anything beyond a quick fix before sending, use a dedicated editing tool.
Exact menu labels and button placement can shift between WhatsApp app versions — if you don't see the edit button described above, make sure your app is fully up to date first.

Episode 57 — Google quietly turned your phone's camera into a live AI you can talk to — point it at anything, and it answers you back, right away.

Search Live's real-world accuracy (handling of ambiguous objects, noisy environments, less common languages) was not independently tested by this channel — the claims here are sourced to Google's own announcement and independent reporting, not a first-hand demo.
It is not a substitute for expert judgment on anything high-stakes — a medical symptom, a legal document, a safety-critical repair. Treat it as a fast first look, not a final answer.
Some newer additions inside Google's AI Mode (described in press coverage as "information agents") are gated to Google AI Ultra subscribers — the base Search Live feature itself is free, but not every future AI Mode feature necessarily will be.

Episode 56 — Your headphones can already translate anyone talking to you — out loud, as they speak.

Available in 12 countries so far, per Google's own March 2026 announcement — not the whole world yet. Check the Google Translate Help page for the current list before assuming your country is covered.
This is a real-time conversational aid, not a substitute for a human interpreter in anything high-stakes — legal proceedings, medical consultations, or anywhere a mistranslation carries real consequences.
Live Translate performance in practice (speed, accuracy in noisy environments, handling of accents) was not independently tested by this channel — the claims above are sourced to Google's own announcement and independent reporting, not a first-hand demo.

Episode 55 — Every word you say to Alexa goes straight to Amazon now.

This does not stop your voice request from being sent to Amazon's servers — that capability was removed in March 2025 and nothing in the app restores it. It only lets you delete recordings after the fact.
The voice-delete commands only work after "Enable Deletion by Voice" has been turned on in the app. Said before that, they're just ignored.
The physical mute button is the only thing that blocks a recording from being made (and sent) in the first place — it is not a substitute for the deletion toggle, and vice versa.

Episode 54 — Your mom calling you on video, crying, begging you to send money right now — that crying face might not be real.

This is specifically about real-time face-swap filters used live on a video call — not photo deepfakes, not voice-only cloning (that's a different mechanism, covered in episode 38), and not pre-recorded deepfake video.
The three-finger test is a real, still-useful first check against cheaper scam tooling, not a guaranteed detector — treat a clean pass as inconclusive, not as proof of anything, per the verified July 2026 reporting above.
No visual test is currently reliable against the most advanced real-time deepfake tools, per the cited experts — the callback-on-a-known-number step is the part of this guide with no known bypass, and it's the one to actually rely on.

Episode 53 — That job you never heard back from? Software read your resume first — and it might not have actually understood a word of it.

This is about ATS resume-parsing specifically — not a claim that formatting is the only reason any single application doesn't get a response, or that every employer's ATS behaves identically.
The 78%/97.4%/71% adoption figures and the recruiter-review-time figures are averages from the cited surveys and studies, not a claim about any one specific employer's actual process.
Fixing formatting makes a resume parse correctly and reach a person intact — it does not guarantee a job offer, and does not fix an underlying mismatch between a candidate's real qualifications and a role's requirements.

Episode 52 — ChatGPT has been quietly building a profile on you — even when you're not talking to it.

This is about ChatGPT's Memory system specifically (Dreaming plus the visible summary) — not a claim about every AI product's memory behavior.
The exact wording, detail, and accuracy of what comes back varies by account, plan, region, and how much history exists — OpenAI's own rollout was staged by plan and country starting June 2026, so exactly what you see may differ from what's shown here.
Asking the prompt shows you a summary of what the model has inferred and retained — it is not guaranteed to be a complete, itemized list of every underlying data point OpenAI's systems hold about you.

Episode 51 — AI never tells you when it's just guessing — not about your health, not about your money.

This is about general-purpose AI chatbots answering health and financial questions specifically — not a claim about every AI product or every possible use case.
The confidence-check prompt does not make the underlying answer more accurate. It only makes the chatbot state a confidence level and name specific risks instead of answering with flat, undifferentiated certainty — verify anything health- or money-related that actually matters through a real professional or primary source regardless of what number it gives back.
The 49.6%/57%/88%/134-out-of-200 figures are study and test averages from the specific sources above, not a claim about what any one specific answer you personally receive will be.

Episode 49 — Google now puts an AI-written answer above your search results — the way most people search the internet just changed, and it's wrong more often than you'd think.

This is about Google Search's AI Overviews specifically — not a claim about every AI product, every search engine, or every other Google product.
The 'Web' filter removes the AI Overview for that one search only — it is not a permanent account setting, and it has to be tapped again on the next search.
The ~90%/1-in-10 figure is an average from one independent analysis, not a claim that any specific answer you personally see is wrong — there's no way, from inside the page, to know which answers fall in that 10%.
The Search Labs toggle (the optional step above) is real but incomplete by Google's own stated fine print — it does not fully disable AI Overviews outside of Labs, so don't rely on it alone.

Episode 48 — Your kid's phone can now blur nudity and violence for them. It will never tell you when it does.

Requires your child to be signed in with a Child Account inside your Family Sharing group — it does not work on an adult Apple ID, and a teenager signed in with their own adult-converted Apple ID can turn it off themselves.
It only inspects photos and videos sent or received in Messages and FaceTime video calls on Apple devices — it does not scan third-party apps (WhatsApp, Snapchat, Instagram DMs, etc.), text-only messages, or web content.
It does not notify a parent when it blurs something, by Apple's own design — the analysis and the block happen entirely on the child's device, and no signal leaves that device, even to Apple itself.
The Screen Time passcode gate described above is specifically what Apple's support page states for children under 13 — Apple's own documentation notes that age requirements and defaults vary by state, country, or region, so don't assume the exact same gate applies identically to an older teen without checking your own child's device.

Episode 47 — An AI agent already texts you before your forgotten trial turns into a real charge.

Rowan is a limited rollout with a waitlist as of this episode's verification — it is not available to every Rocket Money user today, and no public full-rollout date has been announced.
Rowan requires Rocket Money's Premium+ tier, $15/month — it is not part of the free tier or the older, standard Premium tier (which already offers non-AI subscription cancellation on a pay-what-you-think-is-fair basis, roughly $7-$14/month).
Every sourced action Rowan takes — cancelling, negotiating, chasing a refund — happens after the user replies to its text. It is not described anywhere in the sourcing as acting on its own without that reply.

Episode 46 — Your Copilot chats at work don't vanish when you close the tab.

Covers Microsoft 365 Copilot / Copilot Chat on a work or school account only — not the free consumer Copilot, and not every workplace AI tool in general.
Not every manager or IT admin can read an employee's actual chat content — Microsoft's own documentation states that reading prompt/response text needs specific Purview permissions (such as eDiscovery or Communication Compliance roles), not general admin or manager status.
This is not live monitoring. The documented capability is retention plus access under a formal compliance or legal process, not someone watching chats in real time.
Retention length is not one fixed number — it depends on the organization's own configuration and license; deleting your own activity history in My Account does not necessarily remove your organization's separately-retained copy, and this guide does not claim that it does.

Episode 45 — Your phone carrier is already screening calls before you answer.

US carriers only — Verizon, AT&T, and T-Mobile. Not a claim about carriers in other countries, or about carriers not named here.
None of these systems catch every unwanted or fraudulent call — a scammer calling from a real, verified number can still get through.
STIR/SHAKEN is a separate system from the scam-detection covered here. It authenticates that a caller ID hasn't been spoofed; it does not determine whether the caller's intent is legitimate, and this guide does not present caller-ID authentication as proof a call is safe.

Episode 44 — AI can already read your private messages.

Rollout is limited to select markets, starting with the US in English — Meta states it is expanding to more countries and languages, with no fixed date given.
Meta's own materials acknowledge a generated summary 'might be inappropriate or inaccurate' — treat it as a starting point, not a substitute for reading anything you need to act on.
This guide covers Message Summaries specifically, not any other Meta AI feature inside WhatsApp, Instagram, or Facebook.

Episode 43 — Windows can now take a screenshot of your entire screen every few seconds — and save it.

Recall requires a Copilot+ PC (specific NPU-equipped hardware) and a supported Windows 11 build — it is not available on every Windows PC, even if the toggle exists in Settings on some machines it will do nothing without the right hardware.
Turning the setting off does not undo whatever a bypass tool may have already copied before you turned it off — this setup only controls whether Recall keeps saving snapshots going forward, and lets you delete what it's already stored.
This guide is Windows/Recall only — it does not cover any other AI feature's own screen-recording or activity-history behavior on macOS, ChromeOS, or other platforms.

Episode 42 — Your phone might already have an AI that can catch a scammer mid-call — it's just turned off.

Pixel-only, and only specific models: in the US, Pixel 6 and later plus Pixel 9a/10a; internationally (Australia, Canada, France, Germany, India, Ireland, Italy, Japan, Mexico, Singapore, Spain, UK), Pixel 9 and later, with a SIM and device location in one of those countries.
This is a different feature from Google's separate spoofed-caller-ID detection (which works across Samsung/OnePlus but only flags number spoofing, not conversation content) — this guide covers Scam Detection specifically.
It's off by default — it does nothing until you turn it on in Settings.

Episode 41 — ChatGPT can now see every subscription that's draining your bank account behind your back.

As of this recording, Finances is live for ChatGPT Plus and Pro subscribers in the United States only — not the Free tier, and not yet available globally.
It launched as a Pro-only US preview on 15.5.2026 and was reported to expand to Plus users by around 25.6.2026 — availability by plan may keep changing as OpenAI continues the rollout.
It cannot move money, pay bills, or place trades — it only reads and displays data from accounts you've connected; it does not replace a budgeting app that can act on your behalf.

Episode 40 — Do you know who's already grading how you drive, every time you get in the car?

Exact app menu wording and location vary by car manufacturer and model year — this guide describes what to look for, not a guaranteed identical screen for every viewer.
The GM-to-LexisNexis/Verisk pipeline specifically was already shut down by GM as of March 2026 — this guide is about the ongoing, broader industry practice at other manufacturers, and about your standing right to check your own file regardless of which company held it.
Verisk's current, active involvement in collecting driving data is unclear in available sources — this guide treats LexisNexis as the primary, confirmed contact for a consumer disclosure request.

Episode 39 — Your smart TV isn't just playing what's on. It's watching what you watch — and sending that back.

Exact menu wording and location can vary by TV model year and firmware version — this guide describes what to look for, not a guaranteed identical screen for every viewer.
Turning ACR off does not delete data already collected and reported before you changed the setting.
This episode covers ACR specifically, not other data a smart TV or its apps may collect (account data, app usage inside a streaming app, voice assistant recordings), which have their own separate settings.

Episode 38 — Just ten seconds of your voice is all AI needs to clone it now — and scammers are already using it.

A safe word stops the specific 'fake emergency call' scam pattern described here — it doesn't protect against every kind of scam or fraud.
This only works if it's actually agreed upon in advance and everyone involved remembers it exists — it protects nothing if only one person set it up.
This episode covers the voice-cloning phone-scam pattern specifically, not broader AI-deepfake risks (video, images) which use different mechanisms and defenses.

Episode 37 — Every photo you've ever taken in Google Photos has already been looked at by AI — before you ever asked it to.

The exact toggle wording and menu path can vary by app version, region, and account type (personal vs. Workspace) — this guide describes what to look for, not a guaranteed identical screen for every viewer.
Turning off human review of your queries does not delete or undo the AI's existing analysis of your photo library — there is currently no publicly documented Google Photos setting that reverses that.
This episode covers Google Photos' Ask Photos specifically, not Google's broader AI/Gemini data policies across its other products.

Episode 36 — Every private thing you've ever typed into ChatGPT trains its next model — and that's the default.

Turning the toggle off is not retroactive — anything sent before you turned it off may already be part of a training run and cannot be pulled back out by this setting.
Temporary Chat still stores the conversation on OpenAI's servers for up to 30 days for safety monitoring before permanent deletion — it skips history and training, not all server-side storage.
This episode covers OpenAI's ChatGPT specifically; other AI tools (Claude, Gemini, etc.) have their own separate data-training settings, not covered here.

Episode 35 — An automation built in n8n can already say "done" — and still do the exact same thing again.

This stops duplicate runs caused by a slow reply; it does not protect against a sender that retries for a different reason (e.g. its own network error) — de-duplication is the real fix for that, not response timing alone.
Moving the reply earlier means the caller can no longer see the workflow's own output in that response — if the caller genuinely needs the result synchronously, this tradeoff needs a different design (e.g. polling a status endpoint).
This does not undo damage from duplicate runs that already happened — check for and clean up any real double-sends before assuming the fix alone is enough.

Episode 34 — Your n8n node runs once for every item you send it — not once, total.

This explains and manages per-item execution; it doesn't remove real per-call cost or rate limits — batching controls the pace, it doesn't make the calls free.
Batch size is a real tradeoff: too large risks hitting a rate limit anyway, too small barely helps — the right number depends on the specific API's own limits.
Test with a set close to your real production volume before trusting a workflow at scale — 3 test items behaving fine says nothing about 300.

Episode 33 — Nothing's wrong with your n8n AI agent. It's forgetting on purpose — once per 5 messages.

A larger context window means a longer, more expensive prompt on every message — there's a real cost/quality tradeoff, not just 'bigger is always better.'
This fixes forgetting within the configured window; it does not give the agent persistent memory across completely separate sessions unless the memory node itself is set up for that.
Postgres/Redis memory requires that datastore to already be reachable from your n8n instance — this is a bigger setup step than Simple Memory's zero-config default.

Episode 32 — By default, n8n never tells you when a workflow dies.

The Error Workflow has to be set per-workflow (or applied to all active workflows via n8n's own template) — it is not automatically on for every workflow in an account by default.
This tells you a workflow failed; it doesn't diagnose why on its own — the Error Trigger's output data (the error message, the failing node) is what you use to investigate.
Test on a duplicate or test workflow first — deliberately breaking a production workflow to test this is not worth the risk.

Episode 31 — ChatGPT needs the internet to respond. This AI doesn't — and it just proved it.

The one-time model download does need real internet — this is an offline-after-setup tool, not a zero-download one.
A modern laptop with 16GB RAM comfortably runs 7-8B parameter models; larger models (Llama 4, some Qwen 3 variants) need considerably more.
Model names move fast — this episode uses llama3.2, the current official quick-start model as of this episode; check ollama.com's model library for the current recommended default before assuming an older name (llama3, phi3) is still the best starting point.

Episode 30 — AI chats love padding the answer before they actually answer you.

It resets every new conversation — this is not a persistent account setting like ChatGPT's Custom Instructions, so it has to be pasted again in each new chat.
For questions that genuinely need nuance (medical, legal, anything where a caveat actually changes what you should do), the instruction already tells the AI to keep caveats that "change the answer" — but always read past the first sentence when the topic actually calls for it.
Not tested identically across every model version — behavior demonstrated on ChatGPT; the underlying mechanism (instruction-following) is standard across current chat models, but exact phrasing sensitivity can vary slightly between them.

Episode 29 — Right now, your n8n workflow can fail completely — and still say "Success."

This is about the "On Error: Continue" and "Continue (using error output)" settings specifically — a node left on the default "Stop Workflow" behaves as expected and does flag the workflow as failed.
The fix shown (an IF node checking for an error field) has to be added to every node you've set to Continue — it isn't a global setting that protects a whole workflow at once.
Verified against n8n's current official docs and community reports as of this episode, not personally re-tested against every n8n version — if n8n changes this behavior in a future release, re-check before relying on this exact setup.

Episode 28 — Four rows in my database started with "test_". Only three of them were real test data.

This one test showed correct judgment on one specific, well-commented trap — it isn't a guarantee that every ambiguous instruction gets caught correctly on every codebase, especially one with no explanatory comments at all.
Free-plan accounts can't run this at all without upgrading first — the $20/month claim is specifically about Pro-and-up, not every Claude account.
Working on a git branch only protects code already in a git repository — it does nothing for a request made directly against a live database or production system with no version control.

Episode 27 — I told Claude Code the wrong bug. On purpose.

This one test showed the tool looking past a wrong description on one small, clear-cut bug (an operator-precedence mistake) — it isn't a claim that every wrong description gets caught on every bug, especially subtler or more ambiguous ones.
Free-plan accounts can't run this at all without upgrading first — the $20/month claim is specifically about Pro-and-up, not every Claude account.
Needs a real project folder and an actual broken file to point it at — it doesn't diagnose a bug from a description alone with no code to read.

Episode 26 — This is a real AI model. My WiFi is off.

Ollama separately offers optional paid cloud tiers for cloud-hosted inference — this episode is about the free, local, offline path specifically, not Ollama's whole product line.
A small local model won't match the biggest paid models on hard reasoning tasks — the claim here is a real, private, free, offline assistant, not a like-for-like replacement.
The first ollama run of any model needs an internet connection to download it once — "offline" describes every run after that, not the very first one.

Episode 25 — Your n8n retry doesn't just try again. It can do the exact same thing twice.

This protects against an exact retry of the same input — it does not fix a different, genuine failure elsewhere in the workflow.
The lookup step needs a real, persistent store (a Sheet, Airtable, or a database) — n8n's own in-memory workflow data isn't guaranteed to survive a restart on every hosting setup.
Adds one Code node, one IF node, and one write-back node per external call that needs it — it isn't free, and skipping it on even one call leaves that one unprotected.

Episode 24 — Editors squeeze the picture to fit the audio. We do the opposite.

This does not fix a bad recording — if the narration itself is rushed or unclear, matching cuts to its pauses just preserves that pacing exactly as-is.
Doing this by hand, clip by clip, is slower than a fixed-slot template — the payoff is fewer overlaps and cut-off words, not less editing time.
This channel's own version (retime.py) automates the matching across an entire script at once; doing it manually in a general editor means finding each gap yourself, one at a time.

Episode 23 — 22 episodes in, one file keeps this from breaking.

The file only helps if it names real, specific mistakes — a vague wishlist ("write good code," "be careful") gives an agent nothing concrete to check itself against.
It's read once at session start, not enforced like a lint rule — nothing physically stops an agent from breaking a written rule anyway. What it removes is the excuse of not knowing the rule existed, not the possibility of a mistake.
The exact install command shown here can change as Claude Code updates — re-check code.claude.com/docs if this episode is more than a few months old.

Episode 22 — n8n's AI Agent can lie to you — and still show green

This is a design characteristic of n8n's current AI Agent node, confirmed as unresolved as of this episode's publishing (one fix proposal was closed "not planned" in March 2026) — re-check n8n's own GitHub if this episode is more than a few months old.
The fix adds real setup work per tool, per agent — it does not come free, and skipping it on even one tool leaves that one silent.
This episode covers one specific, documented failure mode (a tool call failing silently inside an Agent) — not a general audit of n8n's reliability.

Episode 21 — Can an AI browser actually run your errands? We tested the real number

It does not reliably chain many steps together — a wrong turn early in a long task compounds, and everything after it goes wrong too. Shorter, single-purpose tasks are where it's actually reliable today.
A paid "background assistant" tier exists for running tasks without watching them live — the free tier expects you to stay present and check in.
This episode names one specific browser (Comet) and one specific benchmark, both current as of publishing — re-check both if this episode is more than a few months old; this category is moving fast.
Not a security review: agentic browsers as a category have documented risks around a malicious page hijacking an agent's actions — this episode doesn't cover that side, only whether the everyday-task claim holds up.

Episode 20 — ChatGPT was going to buy things for you — then they quietly killed it

This isn't a criticism of ChatGPT generally — it's one specific, named feature (Instant Checkout) that launched and was retired within about six months.
Agentic checkout hasn't disappeared industry-wide — other players (Google, Perplexity) have their own versions; this episode covers what ChatGPT itself does today, not the whole category.
Check ChatGPT's own current shopping behavior if this episode is more than a few months old — this space is changing fast.

Episode 19 — Claude keeps your files now — not just this one chat

This fits recurring work with the same files and instructions — a one-off question doesn't need a project.
Files still count toward the model's context the same as anything else — a project doesn't make your files free to include.
Anthropic's own limits here can move — check Claude's current project settings if this episode is more than a few months old.

Episode 18 — ChatGPT can use a website now — not just talk about one

This is for tasks with a clear, checkable finish line (compare, look up, summarize across a few pages) — not open-ended research or anything where a wrong answer is costly and hard to catch.
It won't act inside an account you're already signed into — a task needing that isn't a fit for it.
Names and limits in this space change fast — OpenAI retired the previous "agent mode" without much notice days before this episode was recorded; check ChatGPT's own current mode picker if this episode is more than a few months old.

Episode 17 — One word can wreck a whole explanation.

This catches whether a sentence is understandable, not whether it's factually correct — a wrong but clearly-worded claim still needs separate fact-checking (see episode 15's own correction).
It works best with a second person; testing it on yourself only works if you can genuinely forget what you meant to say, which is harder than it sounds.
This rule caught real gaps in the back catalog too (episodes 3, 6, 7, 9, 10, 11 all had unexplained jargon or steps too vague to actually follow) — those have since been rewritten to the same standard, not left as an exception.

Episode 16 — Someone open-sourced our own video pipeline.

This is a tool for building videos with code/agent instructions — it doesn't replace a camera or footage of a real event; it's for the same kind of screen-recording-and-graphics video this channel already makes.
We tested a small, simple example, not a full multi-scene production — a longer, more complex video will take longer to render and may need more setup than shown here.
Open-source projects change; if a command in this episode no longer matches what you see, check the project's own current documentation rather than assuming this episode is still exact.

Episode 15 — Gemini fixes your broken formula, already built in.

This fixes mistakes in the instructions you write yourself — it doesn't know whether the numbers you typed in are correct, only whether the instruction is valid.
The exact wording and location of the Fix button can change as Google updates Sheets — if it isn't where this episode shows it, look for an error indicator on the box itself.
Not free for every account: Google's own rollout (June 2026) lists this for Business, Enterprise, Education, AI Pro and AI Ultra Workspace plans. On a personal Gmail account, the equivalent needs Google One AI Premium — check your own account's access before assuming it's there.

Episode 14 — Everyone's sharing this claim that AI always lies to please you.

Three tests on one model is evidence about that model in that situation, not a universal claim about all AI — a different framing or a different model could behave differently.
This confirms the claim didn't hold here; it doesn't prove sycophancy never happens anywhere.

Episode 13 — ChatGPT can recall things about you, even in a brand new chat.

It also applies to chats you'd rather it left alone — memory is per-account, not something you switch on per conversation.
This is memory of what you've said, not a guarantee of accuracy — it can carry forward something wrong just as easily as something true.

Episode 12 — Most people think Claude Code is only for programmers.

Starting with one big ask instead of one small change is where this goes wrong first — the discipline is the small-step part, not the tool.
It still needs you to check the result each time — describing a change and trusting it blindly is a different, riskier habit.

Episode 11 — AI narration has a flaw you can't consciously name — a word loses its ending, an S goes dull or too hot.

This measures against the narration's own median, not a fixed external standard — a whole file recorded badly could pass its own bad baseline.
It catches what it's built to measure — pacing, rate, sibilance, endings. It is not a general "does this sound good" check.

Episode 10 — Most leads go cold before anyone even replies.

This only helps if the split is actually correct — auto-sending something that needed a real decision is worse than a slow reply.
A draft nobody reviews is the same as no reply at all — the human step still has to happen.

Episode 9 — This agent can send emails by itself. It never does.

This slows down anything that genuinely needs to go out instantly — the trade is deliberate, not free.
It only protects the send step. A draft with a wrong fact still needs an actual human read, not just an approval click.

Episode 8 — A check said it passed. It lied.

Passing on one known-bad case doesn't prove the check catches every bad case — only that it isn't blind to that one.
This has to be repeated after any change to the check itself, not just after changes to what it checks.

Episode 7 — Your n8n agent doesn't know it's wrong

This only catches what the agent itself can flag as uncertain — it does not catch a confidently wrong answer.
"Route to a human" only helps if someone actually reviews that queue — an unread inbox is the same as no check.

Episode 6 — Three things your AI agent still breaks on.

These are three instructions, not three features — the agent still has no real memory, no login handling, and no undo. The lines only change what it's told to do about each.
None of this needs a rebuild, but it does need you to actually add the lines — an agent left on its defaults still has all three problems.

Episode 5 — Your captions are hiding behind Instagram

This is Instagram and TikTok's current UI. Platforms change their layouts, and the exact numbers can shift.
Decorative background elements can still use the full frame — only text and anything meant to be read needs to respect the safe box.

Episode 4 — This video almost shipped broken

Testing a check against a known-bad case does not guarantee it catches every bad case — only the ones like the one you tested.
This is a discipline, not a one-time fix. The three defects here happened after checks already existed; the checks just weren't checked.

Episode 3 — Your AI agent is already lying to you

One prompt line does not make an agent reliable. It changes what it reports, not what it can actually do.
Nothing here is unique to one platform — this is what the failure modes look like across every agent we've tested.

Episode 2 — What an AI agent actually is (not the marketing definition)

Agents do not fail like chatbots. A chatbot gives you a bad answer; an agent fails confidently, halfway through, having already done part of the work.
That is why the first one you build should touch something reversible — a draft, not a send.
Nothing here makes an agent reliable. It makes the word mean something, which is what the rest depends on.
The loop is the definition, not a product. Two tools can both have it and one can still be useless for your job.

Episode 1 — Your ChatGPT keeps giving you the obvious

It cannot browse the web and it cannot read your files — it changes how the model reasons, not what it can reach.
On a brand-new chat the first answer is sometimes still generic. Ask once more and it settles.
It is not a jailbreak and it does not raise any usage limit.
Custom instructions are per-account, not per-device: it follows you, and it also applies to chats you would rather it left alone.